Ageing in Epoxy Resin As a Precursor to Electrical Treeing

Ageing in Epoxy Resin as a Precursor to Electrical Treeing A thesis submitted to the University of Manchester for the degree of Doctor of Philosophy in the Faculty of Science and Engineering

2020

Harry McDonald

School of Engineering Department of Electrical and Electronic Engineering

Table of Contents

Abstract ...... 11 Acknowledgements ...... 13 1. Introduction ...... 14 1.1. Cables ...... 14 1.1.1. Cable Background ...... 14 1.1.2. XLPE Cables ...... 16 1.1.3. Electrical Tree Formation ...... 17 1.2. Electrical Treeing Background ...... 18 1.2.1. Stages of Electrical Treeing ...... 18 1.2.2. Standard Experimental Design ...... 19 1.2.3. Concerns regarding the needle-plane methodology ...... 20 1.2.4. Electrical Treeing In Different Materials ...... 21 1.3. Tree Initiation ...... 22 1.3.1. What is Tree Initiation? ...... 22 1.3.2. Incubation Period ...... 22 1.3.3. Tree Initiation Mechanism ...... 29 1.3.4. Impulse Treeing ...... 30 1.4. Tree Growth ...... 31 1.4.1. Tree Shape ...... 32 1.4.2. Bifurcation (Branching) ...... 32 1.4.3. Stages of Tree Growth ...... 33 1.4.4. Growth Mechanisms ...... 34 1.4.5. Electromechanical Fracturing ...... 35 1.5. Aims and Objectives ...... 37 1.6. Thesis Summary ...... 38 2. Plane-Plane Sample Testing ...... 39 2.1. Fine Tree Initiation Testing ...... 39 2.1.1. Introduction ...... 39 2.1.2. Background ...... 40 2.1.3. Methodology ...... 41 2.1.4. Fine Tree Testing Results ...... 57 2.2. Interfacial Tracking Tests ...... 61

2.2.1. Introduction ...... 61 2.2.2. Background ...... 63 2.2.3. Methodology ...... 64 2.2.4. Interfacial Tracking Test Results ...... 65 2.3. Water Absorption Testing ...... 68 2.3.1. Introduction ...... 68 2.3.2. Background ...... 68 2.3.3. Methodology ...... 70 2.3.4. Water Absorption Result...... 76 2.4. Conclusion ...... 78 3. Atomic Force Microscopy with Infrared Spectroscopy (AFM-IR) - Technique ...... 80 3.1. Introduction ...... 80 3.1.1. Background information on chemistry in electrical treeing ...... 80 3.1.2. Atomic Force Microscopy with Infrared Spectroscopy – How it works ...... 88 3.2. AFM-IR Chemical Analysis Testing Methodology ...... 91 3.2.1. Samples ...... 92 3.2.2. Preparation for AFM-IR ...... 96 3.2.3. AFM-IR Operation and Data Analysis ...... 100 3.2.4. Conclusion ...... 104 4. Atomic Force Microscopy with Infrared Spectroscopy (AFM-IR) - Testing and Results ...... 105 4.1. AFM-IR Tree Initiation Testing Results ...... 105 4.1.1. Pre-initiation Tests ...... 105 4.1.2. Post-initiation Tests ...... 118 4.2. AFM-IR Tree Growth Testing Results ...... 132 4.2.1. AFM(1)-A1(2) – Earliest Stages of Tree Growth ...... 132 4.2.2. AFM(1)-G1 – 100 µm from the needle tip ...... 135 4.2.3. AFM(1)-G2 (Channel 1 – 100 µm from the needle tip) ...... 138 4.2.4. AFM(1)-G2 (Channel 2 – 500 µm from the needle tip) ...... 142 4.3. Conclusion ...... 148 5. Discussion ...... 150 5.1. Plane-Plane Sample Configurations ...... 150 5.1.1. Sample Design Issues ...... 150 5.1.2. Water Saturated Samples ...... 150 5.1.3. Tracking Samples...... 151

5.1.4. Reverse Tree Samples ...... 151 5.1.5. Improved Sample Analysis Needed ...... 152 5.1.6. Application of AFM-IR to plane-plane electrode configurations ...... 153 5.2. Discussion on Tree Initiation Findings ...... 154 5.2.1. Degradation processes occurring prior to initiation ...... 155 5.2.2. Distribution of degradation around the needle tip ...... 157 5.2.3. Degradation necessary for tree initiation to occur ...... 158 5.2.4. Mechanisms occurring during tree initiation ...... 160 5.3. Discussion on Tree Growth Findings ...... 162 5.3.1. Raised Channels ...... 162 5.3.2. Exposure of channels during polishing ...... 164 5.3.3. Branching points ...... 165 5.3.4. Degradation products beyond the channel walls ...... 167 5.4. Proposed Model of Tree Growth ...... 168 6. Conclusion ...... 173 7. Future Work ...... 175 7.1. Planar Sample Experimental Configurations ...... 175 7.1.1. Reverse Tree Testing ...... 175 7.1.2. Water Absorbent Samples ...... 176 7.1.3. Alternative Sample Designs ...... 177 7.2. AFM-IR ...... 178 7.2.1. Sample Preparation...... 178 7.2.2. Alternative Materials ...... 179 7.2.3. Initiation testing performed with electroluminescence analysis ...... 179 8. References ...... 181 9. Appendix A- Energy conversion calculation from partial discharges for sample AFM(1)-A1(2) 188 Appendix B– List of Publications ...... 190 Chemical Analysis of Tree Growth in Epoxy Resin using AFM-IR Spectroscopy ...... 191 Chemical Analysis of Solid Insulation Degradation using the AFM-IR Technique ...... 199 High Resolution Chemical Analysis of Electrical Trees through AFM-IR Spectroscopy ...... 203 A High-Resolution Study of Chemical Aging Prior to Electrical Tree Growth ...... 207

Word Count: 56061

List of Figures:

Figure 1-1: Total installation cost of a 132 kV XLPE Cable for different insulation sizes, from [1] 15 Figure 1-2: Branch-type Tree (Left) and Bush-type Tree (Right). Such differences in structure can be caused by material differences or by differences in the voltage application. [4] .. 16 Figure 1-3: Electrical trees growing from the outer semicon layer of an XLPE cable following water tree growth (stained in methylene blue) [8] ...... 17 Figure 1-4: Simplified schematic of typical electrical treeing experimental setup ...... 20 Figure 1-5: Electrical Tree Initiation Times with logarithmic axis [25], using data from [26] showing initiation probabilites over ageing times in excess of 10000 minutes...... 23 Figure 1-6: Degraded region in LDPE – Imaged using Methylene Blue Dye. [33] ...... 25 Figure 1-7: Electroluminescence at Needle Tip [34] ...... 25 Figure 1-8: Intensity changes of electroluminescence (showing intensity in different phase quadrants) [31] ...... 26 Figure 1-9: Charge Absorbed by system (Experimental vs Theoretical) [40] ...... 28 Figure 1-10: Tensile stress orthogonal to the electric field. Taken from [70] ...... 37 Figure 2-1: Grown in epoxy, fine tree growth leads to reverse trees forming (circled). Growing from the planar electrode towards the needle tip. Breakdown typically occurs when the reverse tree bridges the insulation...... 40 Figure 2-2: The chemical structures of the molecules which make up LY 5052 epoxy resin and HY 5052 Hardener [72] ...... 43 Figure 2-3: Cured epoxy sample in the needle-plane configuration (2 mm gap from needle tip to base)...... 45 Figure 2-4: Circuit diagram for high voltage electrical ageing of epoxy samples...... 46 Figure 2-5: Picture of experimental setup (corresponding to circuit diagram), including labels. . 46 Figure 2-6: a) Schematic of test sample with sharp tip (3 µm radius), b) Schematic of dummy sample with rounded head (0.5 mm radius). Epoxy is not to scale, see Figure 2-3. ... 47 Figure 2-7: The UI (User Interface) for the OMICRON partial discharge detection system. Discharges are shown as dots on the graph on the left. Their magnitude is recorded and shown on the vertical axis. Their phase is shown on the horizontal axis in comparison to the green, voltage line...... 48 Figure 2-8: Projection of electrical trees in epoxy, captured on CCD camera with a backlight..... 49 Figure 2-9: Schematic of samples grown in needle-plane configuration. Dotted lines show different example areas for sectioning...... 51 Figure 2-10: OmegaPol grinding and polishing machine ...... 52 Figure 2-11: Schematic of experiment design for planar samples...... 53 Figure 2-12: Photo of experimental design showing planar samples within Perspex box and under silicone oil...... 53 Figure 2-13: Three samples containing fine trees of different optical clarity. a) is taken using a lower magnification than b) and c). This provides a lower resolution but wider field of view. In b) fine channels can be seen. In c) due to the quality of the sample very little can be observed. The dashed lines indicate the epoxy-electrode interfaces...... 54 Figure 2-14: Aluminium sputtering machine, works under vacuum with thermal source...... 55

Figure 2-15: a) Photo of a planar sample between a (steel) top electrode and a (brass) bottom electrode, b) Photo of the top electrode having been rounded at the edges to prevent discharges ...... 56 Figure 2-16: Typical partial discharge signal from samples without electrical trees ...... 56 Figure 2-17: Optical image showing electrical trees within a planar sample. Taken at a lower resolution the fine channels are not clear in this image. The dashed lines indicate the epoxy-electrode interfaces...... 57 Figure 2-18: In a) Optical image in which the fine channels close to the surface can just be seen, one is pointed out by an arrow. In b) the image is altered to make the channels more visible, however they remain difficult to see clearly. The dashed lines indicate the epoxy-electrode interfaces...... 58 Figure 2-19: Schematic of different testing orientations. a) The darker channels are exposed to the high voltage electrode, with the fine trees closest to the ground. b) The fine trees are approaching the high voltage electrode...... 59 Figure 2-20: Model used to calculate the effect of fine trees on permittivity of dielectric ...... 60 Figure 2-21: Illustration of trees initiating from interfacial tracks...... 62 Figure 2-22: Eletrical tree growth perpendicular from tracks in epoxy at different pressures. a) 0 kPa, b) 20 kPa, c) 40 kPa, d) 60 kPa. Taken from [9]...... 62 Figure 2-23: Experimental design for electrical tracking tests showing the preference for trees to grow into higher permittivity materials. Adapted from [10]...... 63 Figure 2-24: Schematic of interfacial tracking experiment. Adapted from [9]...... 64 Figure 2-25: Schematic of Planar Electrode Tracking testing configuration ...... 65 Figure 2-26: Optical image with 30 second exposure of discharges showing as a purplish light emission during tracking tests...... 66 Figure 2-27: Breakdown through the interface causes surface level carbonisation...... 67 Figure 2-28: Electrical trees forming in response to water trees forming within the insulation (stained with methylene blue). Taken from a) [8] and b) [85]...... 69 Figure 2-29: Model for calculating electric fields within insulations in which water trees have formed. Taken from [86] ...... 70 Figure 2-30: Graph of water absorption for epoxy samples against Hardener - Epoxy Resin ratio following 2 days of submersion in heated water ...... 71 Figure 2-31: Schematic of water absorbency test. A represents the base layer with the lowest water absorption and so the highest fields. B represents the layer with the highest water absorption, experiencing the lowest fields. A breakdown channel and tree growth is illustrated at the point at which layer A is thinnest, this is the intended growth mechanism...... 73 Figure 2-32: Three trialled moulds to create the depression within layer “A”...... 74 Figure 2-33: A finished sample. In a) Just visible (pointed at by the arrow) is the interface between the layers of epoxy. In b) this interface is highlighted by a black line...... 75 Figure 2-34: Optical imaging shows the difficulty in imaging samples. Within a) a shadow covers the areas of interest where the base layer is smallest. In b) the interface is visible however due to its small size difficult to see within. The dashed lines show the interface between the epoxy and the electrodes...... 75 Figure 2-35: Optical Images of the water saturated samples, planar electrodes above and below (seen on left). Potential breakdown channel (magnified in right hand image) ...... 76

Figure 2-36: Optical microscopy of the suspected breakdown channel. In a) the channel is viewed from a similar (but reversed) angle as the previous images, b) shows the channel from a different angle and at higher resolution...... 77 Figure 3-1: Autoxidation and chain scission reaction, taken from [34] ...... 82 Figure 3-2: Damage to polymer bonds due to hot electron collisions in oxygen rich environment leading to tree initiation [11] ...... 83 Figure 3-3: Conducting Polyethylene tree in a) The light emission from PD is visible at tree tips. In b) the same tree is visible. Circled in each image is the location of the discharges. (image courtesy of Dr. Hualong Zheng)...... 84 Figure 3-4: Non-conducting polyethylene tree, in a) light emission occurs along the length of the channels. In b) the tree is visible (Image courtesy of Dr. Hualong Zheng)...... 85 Figure 3-5: Norrish Type 2 reaction ...... 85 Figure 3-6: Polyethylene molecular chain ...... 86 Figure 3-7: Epoxide group ...... 86 Figure 3-8: Example spectra (from Sample AFM(1)-A2(1)), annotated with key spectral regions 87 Figure 3-9: Schematic illustrating the functioning of the AFM part of the AFM-IR (Image courtesy of Dr. Suzanne Morsch)...... 89 Figure 3-10: Schematic illustrating the operation of the AFM-IR ...... 89 Figure 3-11: The deflection reading taking from monitoring the position of the AFM probe can be Fourier transformed to give an amplitude of deflection...... 90 Figure 3-12: In a) an example spectra is shown above a surface profile map showing the locations the spectra were taken from, b) shows an absorbance map of the same region at 1726 cm-1 where red areas show higher absorbance. The map is 30 x 30 µm...... 91 Figure 3-13: Optical images of an electrical tree grown in epoxy in Sample AFM(1)-G1: a) lit from above, b) lit from below. Circled are two points at which the channels are exposed.98 Figure 3-14: Top: Example spectra taken by the AFM-IR from different locations on Sample AFM(1)-G1. Bottom: The locations from which these spectra are taken are shown on a surface profile image generated by the AFM-IR...... 101 Figure 3-15: a) Unnormalised spectra and the locations they are taken from, b) The same spectra having been normalised against the 1500 cm-1 peak. Taken from Sample AFM(1)-I1.102 Figure 3-16: The surface roughness of two lines across Sample AFM(2)-10 are shown in the graph in a) (the y-axis of which goes from +40 nm to – 40 nm centred around the average height. The lines tracked across the samples are shown in b)...... 104 Figure 4-1: A Schematic of the void created within the sample following the removal of the needle tip. The dashed line represents the cross-section created through polishing...... 106 Figure 4-2: Optical images from Sample AFM(1)-A2(1) taken a) during polishing, before exposing the void from the needle tip and using back-lighting, b) after polishing the void is exposed and the sample is front-lit emphasising the surface condition...... 106 Figure 4-3: Top: Spectra for Sample AFM(1)-A2(1). Bottom: AFM-IR height profile showing the location of the spectra ...... 107 Figure 4-4: AFM-IR absorption maps for Sample AFM(1)-A2(1); a) Surface Profile, b) 1056 cm-1, c) 1132 cm-1, d) 1248 cm-1, e) 1288 cm-1, f) 1448 cm-1, g) 1604 cm-1, h) 1656 cm-1, i) 1708 cm-1, j) 1726 cm-1, k) 1742 cm-1 ...... 108 Figure 4-5: Top: Spectra for Sample AFM(1)-A2(1) from within the needle tip void. Bottom: AFM- IR height profile showing the location of the spectra ...... 109

Figure 4-6: Top: Spectra for Sample AFM(2)-5. Bottom: AFM-IR height profile showing the location of the spectra ...... 110 Figure 4-7: AFM-IR Absorption maps for Sample AFM(2)-5 a) Surface Profile, b) 1056 cm-1,c) 1248 cm-1, d) 1360 cm-1, e) 1448 cm-1, f) 1604 cm-1, g) 1656 cm-1, h) 1726 cm-1, i) 1742 cm-1 ...... 111 Figure 4-8: Top: Spectra for Sample AFM(2)-10. Bottom: AFM-IR height profile showing the location of the spectra ...... 113 Figure 4-9: AFM-IR Absorption Maps for sample AFM(2)-10 a) Surface Profile, b) 1056 cm-1, c) 1132 cm-1, d) 1248 cm-1, e) 1288 cm-1, f) 1360 cm-1, g) 1448 cm-1, h) 1604 cm-1, i) 1656 cm-1, j) 1702 cm-1, k) 1726 cm-1, l) 1742 cm-1 ...... 114 Figure 4-10: Schematic illustrating the location of three distinct groups of degradation identified in Sample AFM(2)-10 ...... 115 Figure 4-11: AFM-IR absorption map for sample AFM(2)-10 showing absorption at a) 1248 cm-1 and b) 1726 cm-1 ...... 116 Figure 4-12: Optical Image of a tree formed in Sample AFM(1)-A1(2), approximately 15 µm in length...... 118 Figure 4-13: The partial discharges measured associated with AFM(1)-A1(2) ...... 119 Figure 4-14: Figure: Schematic illustrating polishing cross-section in relation to needle tip void and tree channel...... 120 Figure 4-15: Top: Spectra for Sample AFM(1)-A1(2). Bottom: AFM-IR height profile showing the location of the spectra ...... 120 Figure 4-16: AFM-IR Absorption Maps for Sample AFM(1)-A1(2) a) Surface Profile, b) 1604 cm-1 ratio against 1500 cm-1, c) 1604 cm-1 flattened, d) 1604 cm-1 original image, e) 1500 cm- 1 original image (increased contrast settings)...... 122 Figure 4-17: Sample AFM(1)-A1(2) AFM-IR Absorption Maps: a) Surface Profile, b) 1132 cm-1, c) 1248 cm-1, d) 1360 cm-1, e) 1448 cm-1, f) 1604 cm-1, g) 1656 cm-1, h) 1702 cm-1, i) 1726 cm-1, j) 1742 cm-1 ...... 123 Figure 4-18: Schematic illustrating polishing cross-section in relation to needle tip void and tree channel in Sample AFM(1)-A1(2)...... 124 Figure 4-19: Sample AFM(1)-A1(2), cross-section as in Figure 4-18. a) Top and Bottom Lit and b) AFM-IR Surface Profile. Circled is the point at which the channel reaches the surface in each image. Also visible in the optical image is the rest of the channel beneath the surface...... 124 Figure 4-20: Spectra for Sample AFM(1)-A1(2). Bottom: AFM-IR height profile showing the location of the spectra...... 125 Figure 4-21: Sample AFM(1)-A1(2). a) Surface Profile. Absorption amplitude at wavelengths b) 1650 cm-1, c) 1726 cm-1, d) 1752 cm-1. Circled in each image is the location at which the channel has been exposed at the surface...... 126 Figure 4-22: AFM-IR flattened absorption maps for sample AFM(1)-A1(2). Absorption wavenumber used for each map are a) 1504 cm-1, b) 1726 cm-1, c) 1752 cm-1, d) 1248 cm-1...... 127 Figure 4-23: Optical image of Sample AFM(1)-I1. Dashed line represents approximate cross- section...... 129 Figure 4-24: AFM-IR spectra for Sample AFM(1)-I1 corresponding to tip locations indicated by markers on the contact mode height image (right) of epoxy resin following the

initiation of an electrical tree from a needle tip. Spectra are normalized to 1600 cm-1 peak. Approximate needle position shown by white dashed line...... 129 Figure 4-25: AFM-IR images of the surface around the needle tip in AFM(1)-I1: a) surface profile. IR Absorption amplitude at wavenumbers: b) 1056 cm-1, c) 1248 cm-1, d) 1360 cm-1, e) 1448 cm-1, f) 1640 cm-1, g) 1708 cm-1, h) 1726 cm-1. Each map is ratioed against 1500 cm-1 (corresponding to an aromatic peak). The images represent a 30 µm x 30 µm surface...... 130 Figure 4-26: Sample AFM(1)-A1(2). A tree grown in epoxy in which the voltage was removed following initiation. Optically imaged (top and bottom-lit) ...... 132 Figure 4-27: Surface profiles taken on the AFM-IR, the size of the channels is measured in two locations at which the channel is exposed on the surface...... 133 Figure 4-28: a) The height profile from two lines across sample AFM(1)-A1(2). These lines are shown on the AFM-IR surface profile in b). Circled in both are regions above buried channels in which an increase in height is observed...... 134 Figure 4-29: Surface profiles for the channels from Figure 4-28 in Sample AFM(1)-A1(2) given to scale...... 135 Figure 4-30: An optical image (front-lit) of Sample AFM1-G1...... 136 Figure 4-31: A polished surface with an exposed tree channel: a) Optical Image (bottom-lit); sub- surface channel visible. b) Optical Image (top-lit); reflective material visible within channels below the surface. Circled are the locations at which the channel is exposed at the surface. [106] ...... 136 Figure 4-32: Spectra for Sample AFM(1)-G1 from within the needle tip void. Inset: AFM-IR height profile showing the location of the spectra [106] ...... 137 Figure 4-33: AFM-IR images of the surface around the channel in AFM(1)-G1: a) surface profile. IR Absorption amplitude at wavenumbers: b) 1664 cm-1, c) 1702 cm-1, d) 1726 cm-1, e) 1248 cm-1, f) 1604 cm-1, g) 3300 cm-1. Each map is ratioed against 1504 cm-1 (corresponding to an aromatic peak). The images represent a 30 µm x 15 µm surface. [106] ...... 138 Figure 4-34: AFM(1)-G1, optically imaged in the background with the 1702 cm-1 chemical map overlaid. Red areas indicate greater absorption at this wavelength. [106] ...... 138 Figure 4-35: Sample AFM(1)-G2 with an exposed channel. a) Optical Image (bottom-lit); channel structure visible below surface with exposed section of channel circled. b) Optical Image (top-lit); channel exposed at the surface. [106] ...... 139 Figure 4-36: Selected AFM-IR spectra of a channel and surrounding areas from Sample AFM(1)-G2 (normalized to 1504 cm-1 peak). Inset: AFM-IR height profile displaying location of spectra. [106] ...... 140 Figure 4-37: Sample AFM(1)-G2 a) Surface Profile. Absorption amplitude at wavelengths: b) 1604 cm-1 c) 1656 cm-1 d) 1702 cm-1 e) 1726 cm-1 f) 1742 cm-1. [106] ...... 141 Figure 4-38: Background: Optical Image of a tree channel with an overlaid AFM-IR absorption map from Sample AFM(1)-G2 (1742 cm-1) [106]...... 141 Figure 4-39: Composite optical images from sample AFM(1)-G2. Circled are the channels studied in this project...... 143 Figure 4-40: A sub-surface channel from Sample AFM(1)-G2. a) Optical Image (top-lit); channel exposed at the surface. b) Optical Image (bottom-lit); channel structure visible below

surface. Circled are points at the surface at which the channels have become exposed. [106] ...... 143 Figure 4-41: Sample AFM(1)-G2 with a sub-surface channel. Selected AFM-IR spectra of channel and surrounding areas (normalized to 1504 cm-1 peak). Inset: 50 x 50 µm AFM-IR height profile displaying location of spectra. [106] ...... 144 Figure 4-42: AFM-IR images of the surface of AFM(1) – G2 (channel 2). a) Surface profile. IR absorption amplitude at wavenumbers: b) 1248 cm-1, c) 1604 cm-1, d) 1640 cm-1, e) 1656 cm-1, f) 1708 cm-1, g) 1724 cm-1 h) 1742 cm-1. Each is ratioed against 1504 cm-1 (aromatic peak). The images represent a 30 µm x 30 µm surface [106] ...... 145 Figure 4-43: AFM(1) – G2 (channel 2) background images show the channel structure below the epoxy surface. Overlaid images use the AFM-IR chemical mapping at different wavelengths and contrast settings. a) Optical Image b) 1742 cm-1 and set to show the extent of the carbonyls c) 1742 cm-1 and set to show the highest concentrations of carbonyls d) 1656 cm-1 and set to show the highest concentrations of alkene bonds. [106] ...... 145 Figure 4-44: Sample AFM1-G2 surface profile captured using AFM-IR. Circled is a small tree channel raised above the surface...... 146 Figure 4-45: The height profile of two sections of a small channel from Sample AFM(1)-G2. The red line shows the profile across an unexposed section of channel, the blue line captures a partially exposed section of the channel. a) captures the location the profiles are taken from, b) the height profiles...... 147 Figure 4-46: To scale height profiles from Sample AFM(1)-G2. Considers the channels examined in Figure 4-45...... 148 Figure 5-1: a) 1742 cm-1 first testing of Sample AFM(1)-A1(2) and b) Spectra isolating the highest carbonyls ...... 159 Figure 5-2: 1752 cm-1 second testing of Sample AFM(1)-A1(2) ...... 159 Figure 5-3: Sample AFM(1)-A1(2) IR absorbance at a) 1132 cm-1 and b) 1650 cm-1...... 160 Figure 5-4: The channel from sample AFM(1)-G2. On the left a) shows the IR absorbance at 1742 cm-1, b) shows the AFM surface profile. Circled in each image is a small channel which rises above the rest of the surface ...... 163 Figure 5-5: Schematic illustrating the compression of channels at the surface during polishing163 Figure 5-6: Raman Scattering Illustration...... 168 Figure 5-7: Raman spectra taken from channel cores of a non-conducting tree at different distance from the needle electrode. Taken from [53] ...... 169 Figure 5-8: Illustrated model showing different stages of proposed electromechanical fracturing driven tree growth along with partial discharge and chemical degradation...... 172 Figure 7-1: Plane-needle-plane electrode configuration. Adapted from [126] ...... 177

List of Tables:

Table 3-1: Relevant peak assignments [96] ...... 88 Table 3-2: First Testing Set – Sample ageing voltages and comments ...... 94 Table 3-3: Second AFM-IR Testing Set – Sample ageing voltages and comments...... 96

Abstract

A common picture of electrical treeing is of a tree forming; generating channels which then steadily grow until it reaches a runaway stage of rapid growth as the tree approaches the ground conductor. In numerous cases however stages of treeing have been identified which do not conform to this picture. Properties such as tree chemistry, partial discharge magnitudes and phase dependency, channel widths, tree structure and growth speeds are regularly found to be dependent upon the radial extent of the trees. Also identified are changes of channels with time; such as when radial extension of channels is halted, giving way to channel widening, darkening or even the formation of new micro-channels. Each time the properties of the trees change, we are observing differences in the underlying process or processes which produce the electrical trees. This may be a temporary change, such as when channel growth halts only to restart after a waiting period. Or it can indicate a more fundamental shift in the nature of tree growth has occurred.

The objective of this project was to better understand electrical treeing at a more fundamental level. Electrical treeing tests are typically performed using needle-plane electrodes however this configuration will influence and change the properties of growth. To recreate the conditions of electrical treeing in cables, a plane-plane electrode configuration is tested; utilising a number of sample ageing techniques in attempt to accelerate tree initiation and growth. A localised breakdown was formed using these methods; however it was not repeatable nor was tree growth found to occur following it. These tests identified the need to determine ageing prior to tree initiation in long term tests in which channels do not quickly form.

A new chemical analysis technique is applied to electrical treeing for the first time, developing new insight into the nature of this ageing. Atomic force microscopy with infrared spectroscopy (AFM-IR) allows chemical characterisation with spatial resolutions of 50 nm. Using this an early stage channel is found to have formed without producing chemical degradation; based on this it is proposed tree initiation in this case was most likely the result of electromechanical fracturing. Meanwhile the ageing which precedes initiation is found to be more heterogeneous than previously believed. In the study of mature channels it was identified that channels at different distances from the needle tip have distinct chemical signatures. Using these results, along with results and discussions from the literature, electromechanical fracturing is proposed to be active in electrical tree growth in epoxy.

The power of the AFM-IR, demonstrated by these results, along with a large range of potential applications means its use is highly recommended in the future study of electrical treeing. The results obtained here should also be considered in the understanding of tree formation and growth.

Declaration:

No portion of the work referred to in the thesis has been submitted in support of an application for another degree or qualification of this or any other university or other institute of learning.

i. The author of this thesis (including any appendices and/or schedules to this thesis) owns certain copyright or related rights in it (the “Copyright”) and s/he has given The University of Manchester certain rights to use such Copyright, including for administrative purposes. ii. Copies of this thesis, either in full or in extracts and whether in hard or electronic copy, may be made only in accordance with the Copyright, Designs and Patents Act 1988 (as amended) and regulations issued under it or, where appropriate, in accordance with licensing agreements which the University has from time to time. This page must form part of any such copies made. iii. The ownership of certain Copyright, patents, designs, trademarks and other intellectual property (the “Intellectual Property”) and any reproductions of copyright works in the thesis, for example graphs and tables (“Reproductions”), which may be described in this thesis, may not be owned by the author and may be owned by third parties. Such Intellectual Property and Reproductions cannot and must not be made available for use without the prior written permission of the owner(s) of the relevant Intellectual Property and/or Reproductions. iv. Further information on the conditions under which disclosure, publication and commercialisation of this thesis, the Copyright and any Intellectual Property and/or Reproductions described in it may take place is available in the University IP Policy (see http://documents.manchester.ac.uk/DocuInfo.aspx?DocID=2442 0), in any relevant Thesis restriction declarations deposited in the University Library, The University Library’s regulations (see http://www.library.manchester.ac.uk/about/regulations/) and in The University’s policy on Presentation of Theses

Acknowledgements

I’d like to thank the many people that have helped me during my time on this PhD, who’ve made it possible and kept me sane throughout. Thank you to everyone at the Ferranti Building for all your help and support; in particular to my supervisor Prof Simon Rowland for all the help and guidance of the last few years. To Hualong, Ibrahim, Zepeng, Siyuan, Viddy, Rich and Robert who have always been generous with their time and help. Thank you to Dr Suzanne Morsch whose help and chemical expertise was invaluable.

Thank you to Hasti, Matt, James, Zong and Fang for making the office a bearable and at times even fun place to be with the jokes, support and shared complaining sessions which have been necessary to keep going through the project. Thank you to my friends outside the office Joanna, Mark, Katie, Ross, Kathryn, Sean and James.

I’d like to thank everyone at Run Wild, it has been a constant source of energy and enjoyment for me during the happy and not so happy days and that is due to you that have been there. There are too many to mention but I’d particularly like to thank Danny, Tom, Molly, Naomi, Jason and Shona for everything.

Thank you to my family who’ve always been there for me. To my partner Chrissy who has always been so supportive, motivating and encouraging; sharing the happiness and picking me up when I’m down.

Thank you to everyone, it has been a long journey but I made it in the end.

1. Introduction

Polymeric insulation is overtaking oil and paper as insulation within cables due to their low cost and strong dielectric performance with low dielectric losses and high breakdown strengths. With time polymeric insulation will age, and over decades electrical ageing becomes a serious issue. For polymeric insulation a common failure mechanism is assumed to be through electrical treeing. Steps taken to reduce tree formation have been successful. However as we look to increase lifetimes and use ever higher voltages more development is required. This has driven research in this area, as people endeavour to understand why the materials age and eventually breakdown, and look for ways to improve the properties of the materials being used.

Laboratory based studies typically use a needle-plane configuration for accelerated ageing, chosen as it produces electrical trees relatively quickly and consistently. However this is not representative of the more uniform fields expected to be found in a cable. As the conditions the polymer experiences are changed so too may the physical processes which are occurring during tree growth. Any such changes reduce the effectiveness and applicability of the tests as well the level of understanding gained from them.

This study looks to develop tools which will allow better understanding and contextualisation of results from lab-based testing. The approaches taken for this are to develop a more representative testing configuration, and to apply a novel chemical analysis technique to better understand the process involved in ageing.

1.1. Cables

1.1.1. Cable Background Cabling is widely used in the UK power system for the transmission and distribution of power as an alternative to overhead lines. Whilst being more expensive it offers a number of significant advantages that have resulted in use when overhead lines are not felt to be suitable. Examples of this are; a reduced visual impact, necessary in areas of natural beauty, being more resistant to environmental impacts allows easier and more reliable operation in rough terrain, the reduced difficulties in establishing right-of-way which is particularly relevant in urbanised areas and the ability to operate underwater allowing for interconnections to mainland Europe and for offshore generation.

The UK’s power networks are anticipated to expand in the coming years and already contain many assets which are soon reaching or are already beyond their expected lifetimes. There is need then to

14 both develop new cable technology to be cheaper, smaller or longer lasting, and to develop asset management techniques such that the usage of current assets can be optimised.

This project will look at understanding the processes behind electrical treeing, one of the foremost breakdown mechanisms afflicting power cable insulation. Research in this area offers the potential to improve the design of cabling both in terms of materials used and the manufacturing process. Improved understanding of treeing can also allow improved asset management of the cables once installed, with more accurate testing and analysis of the obtained results. Such improvements could increase reliability and so increase cable lifetimes. Alternatively they could be utilised to allow smaller cable designs which would save money in manufacturing and installation which are not inconsiderate. Bartlett et al [1] calculated that a reduction of insulation thickness from 18 mm to 14 mm would lead to a total saving of 29% of manufacturing and installation costs, as shown in Figure 1-1.

Similar savings stand to be made in terms of improved asset management techniques. These include cable testing (which is difficult due to cable lengths and lifetimes), interpreting the insulation condition from the test results (type and scale of the damage) and then taking action on this (can be concluding it is fine, it needs further monitoring, needs repairing or needs replacing). Each step requires a level of understanding and certainty in order for the overall decisions to be made with confidence. Given the consequences of an unexpected breakdown can include blackouts and large fines for a responsible company, the drive for improved methods by which to monitor and manage cable assets is strong.

Figure 1-1: Total installation cost of a 132 kV XLPE Cable for different insulation sizes, from [1]

1.1.2. XLPE Cables Until the 1960s oil and paper insulated cables were most common among high voltage power cables [2]. At this point polymeric, particularly XLPE, cables came into fashion due to their lower weights and reduced costs combined with strong dielectric performance. XLPE cables have continued to become more popular at the highest voltages and now represent the vast majority of cable insulation in the world.

While XLPE cables have strong insulating properties they have been shown to be vulnerable to a particular degradation mechanism known as electrical treeing. These are gaseous channels which can form within the insulation when subject to large electric fields. The exact conditions necessary for tree formation remain unclear; however it is known large, divergent fields will lead to tree formation [3]. These channels propagate through the dielectric, with the channel growth apparently driven by partial discharges, where localised areas of the insulation may breakdown due to exposure to high fields. These do not result in an immediate breakdown across the insulator but in material damage and erosion in that area. The gaseous voids create an electrical weak point, more susceptible to breakdown than the surrounding medium due to their free space encouraging electron avalanches. The avalanches may occur through the length of the channels in non- conducting trees or only at the channel tips in conducting trees where carbonisation has occurred.

Figure 1-2: Branch-type Tree (Left) and Bush-type Tree (Right). Such differences in structure can be caused by material differences or by differences in the voltage application. [4]

As the channels grow it is found they do not grow directly in the direction of the electric field as might be expected. Instead the channels will split and branch off stochastically as shown in Figure 1-2. The reasons for this branching are not fully understood although there has been significant work done in this area (see Section 1.4.2). The manner of this growth, often containing larger channels with smaller channels branching off from them leads to the name ‘electrical treeing’.

1.1.3. Electrical Tree Formation Early XLPE cables were found to be vulnerable to electrical trees. These often developed as the result of voids formed inside the cable during the manufacturing process. The introduction of a semi- conducting layer between the conductor and the insulation, performing extrusion in ultra-clean conditions and crosslinking at high pressure has largely eliminated these issues. A second cause of electrical treeing is water treeing which is a similar type of degradation to electrical treeing. In this case, water trees occur due to water ingress through the outer layers of the cable. Here the trees are formed of pockets of water around a micron in diameter, which appear to be interconnected by sub-micron tracks [5]. The presence of this water changes the dielectric’s insulating properties and alters the electric field distribution by changes in permittivity and conductivity. This encourages breakdown within the insulation ultimately resulting in electrical tree formation [6, 7]. A water tree leading to electrical tree growth is shown in Figure 1-3.

Figure 1-3: Electrical trees growing from the outer semicon layer of an XLPE cable following water tree growth (stained in methylene blue) [8]

The occurrence of water trees has been reduced in cables through adapting the outer layers of the cable to become waterproof by introducing metal sheaths. This has reduced the regularity with which electrical trees form within the bulk of cabling, though has not eliminated it. The most vulnerable area within a cable is now at its joints. While cables can be manufactured in controlled condition factories, they must be jointed in the field and this can introduce the risk of imperfect interfaces as well as that of contaminants entering the cable. Electrical tracking, or surface tracking, is when the degradation occurs along the surface of a material triggering discharges and forming a carbonised, conductive path. This can occur between the layers of two materials and either cause breakdown itself or lead to tree growth into the bulk of the material [9, 10]. Electrical treeing remains one of the most prominent issues afflicting cables and a limitation on their design and operation.

1.2. Electrical Treeing Background

As described previously; electrical trees are gaseous channels formed under high electrical stress in polymeric solid dielectrics. They do not cause immediate breakdown but act as a precursor to a total failure which occurs when the trees bridge the insulation between two conductors and cause it to short circuit. Electrical treeing is typically a long-term ageing mechanism, occurring after decades of operation. They are very hard to predict even when utilising asset monitoring techniques and remain one of the primary limitations on cable design.

1.2.1. Stages of Electrical Treeing There are a number of different stages of electrical treeing which precede total breakdown and asset failure. These can be largely divided into two groups; stages of initiation and stages of channel propagation.

Tree initiation itself seems to be an almost instantaneous process in which a channel of 5 - 15 µm will form within a previously solid region of insulation. Prior to this is an incubation period, a time in which the material is subject to the high electrical field but no visible tree exists. During this time the region surrounding the needle tip becomes chemically degraded [11-13]. It is widely thought that this acts as a precursor to initiation, weakening it until initiation is able to occur and a channel forms.

Once a channel has formed then it will begin to propagate across the insulation. As shown in Figure 1-2 this growth will not be directly in the direction of the field but will be tortuous and branching as it grows, creating the eponymous tree shape. The growth of tree channels is strongly associated with partial discharges in most cases and these are commonly believed to be main mechanism behind tree growth [14]. The properties of the trees being formed are voltage and material dependent,

18 meaning every possible different stage of tree growth cannot be fully described here. However more depth is provided in Section 1.4. What is typical is that for breakdown to occur trees must bridge the insulation between conductors and create a sufficiently conductive path for them to become short- circuited. When these conditions are met a breakdown channel will form virtually instantaneously.

1.2.2. Standard Experimental Design That channels may only form after years of service makes electrical treeing a difficult phenomenon to reliably recreate and test under laboratory conditions (which require testing to be predictable and favour shorter testing scenarios). Many different sample configurations have been tested [15] however the most consistent and effective in tree formation has been the needle-plane configuration which is used as standard. Work discussed in this literature review and thesis should be considered to use this configuration unless otherwise stated.

In needle-plane configuration a sharp point, typically from a precisely manufactured needle is utilised to cause field enhancement and create a divergent field, this is the high voltage electrode. It is embedded within the insulation; unfilled epoxy is a common choice for this as are variations of polyethylene due to their dielectric strengths and transparency, providing easy imaging. The needle is held at a set distance above a planar electrode which is grounded. This design and others of a similar nature are used due to their ability to more readily produce electrical trees in the short time frames suitable for research. Hours are required instead of the decades which are often required to produce trees in practice. This configuration is ideal for aiding the monitoring of tree initiation and growth. A sharp needle tip provides a high level of confidence on where the tree will first form, aiding optical imaging. The configuration also produces consistent samples for partial discharge monitoring, a technique which is typically used to study tree growth and identify that tree initiation has occurred. It is often believed that the needle tip electrode acts similarly to an inclusion from the inner semicon layer into the insulation in a cable. The use of this technique as a standard in the study of treeing does however make the implicit assumption that this is how trees are formed in practice.

Figure 1-4: Simplified schematic of typical electrical treeing experimental setup

1.2.3. Concerns regarding the needle-plane methodology Other methods to accelerate and make tree formation more consistent have been considered and attempted including void insertion [16] and higher frequencies [16, 17]. Such techniques are not always directly analogous; this is seen in the use of higher frequencies which have been shown to alter growth patterns. Because of this they are no longer used to accelerate power frequency treeing tests [17]. Similar concerns can be raised regarding the typical point-plane sample configuration.

The study of electrical treeing reported in the literature is now largely limited to tests performed using well-defined, sharp metal inclusions. The electrode arrangement dictates the electric fields the insulation is exposed to, both in magnitude and in terms of the field lines produced. With point electrodes yielding strongly divergent fields, a point-plane configuration is not analogous to the quasi-parallel electrode used in cabling, so any application of findings made using one configuration to another should be questioned.

Previous tests have demonstrated the importance of the size and shape of the electrodes used, even within the point-plane configuration. Alterations to the radius of the point have been shown to have significant effects upon the initiation and growth times as well as the structure of the tree produced. Auckland et al [17] investigated this looking at the time taken for tree initiation and growth of different tip radii (1, 3, 5, 10, 20 µm). It was found initiation time significantly increased with tip radius, such that no trees could be made to initiate using the 20 µm needle tips. To increase the treeing rate, standard practice has been to utilise needles with small radii manufactured to high tolerances for consistency. The authors also investigated the effect of needle radii upon the

20 structure of the trees produced with stark differences found. At the smallest radius (1 µm) a very direct channel which quickly bridged the conductors formed. At 5 µm a much more heavily branched tree was produced and with the larger radius (10 µm) a bush tree was produced instead. It should be noted that increases in electric field magnitude, through larger voltages, would be expected to correlate to an increase in branching [18, 19], whilst here a reduced needle tip radius would decrease the branching level. The indication from this is that there are parameters which impact tree generation beyond simply the magnitude of the field produced. This could be related to the divergence of the electric field or to other properties of the needles such as their ability to inject charges.

Mason [15] looked at a wider range of electrode designs, still simulating the effect of inclusions through the use of point electrodes but utilising a number of different designs and manufacturing methods. He also considered aspects such as the length of the needle embedded within the polymer and the gap between the electrodes. From this it was noted the geometry of the electrodes do affect channel growth, with the use of long needles potentially causing distinctly different behaviour from that which would be seen in cables.

With such wide variance observed when using relatively similar electrodes, it is clear that reliance upon the point-plane configuration may limit our understanding of the conditions under which electrical trees form and grow. In using such results to understand electrical tree growth in cables, an assumption is made of direct transferability to cables in practice which is not yet supported by evidence.

1.2.4. Electrical Treeing In Different Materials Treeing occurs in a number of different materials including polyethylene (LDPE, HDPE, XLPE), epoxy, EPR and silicone rubber. These are materials which span a variety of dielectric and mechanical properties. That the same degradation mechanism occurs in each case would suggest there is something quite fundamental which underlies this behaviour.

We should not however assume each material behaves identically. Indeed there are many differences which can be observed in the trees grown in these materials. This can be seen in the fine and reverse trees which have been found in epoxy resin samples [20] but not in any other material. Even within the same material changes such as ambient temperature can cause treeing to present significantly different properties. An example of this is epoxy which can be above or below the glass transition temperature which impacts the growth properties of the electrical trees [21].

Understanding both the similarities and the differences between how these materials experience the phenomenon of electrical treeing can give us an insight into the mechanisms which underlie it.

1.3. Tree Initiation

1.3.1. What is Tree Initiation? Tree initiation is here assumed to span the time periods from pre-treeing stages, in which some form of chemical degradation is generally understood to occur, until visible channel formation and the beginning of ‘tree-like’ behaviour. This will cover the earliest stages of the degradation of the insulation, in which a whole and ‘healthy’ region of insulation is subject to high electrical fields which gradually induce chemical or physical damage during the incubation period [11]. As the degradation builds there reaches a point at which a large visible channel (5-15 µm) is able to form [22], this appears to occur very quickly [23] and represents the end of the initiation stage and the beginning of tree growth.

The conditions necessary for tree formation, how these conditions are reached and by what mechanism a gaseous channel is able to form within a solid dielectric are all necessary aspects of tree initiation to be considered.

1.3.2. Incubation Period As discussed previously, electrical tree initiation is a process which generally does not occur immediately. Instead it requires a voltage application for a length of time before any optically visible damage can be shown to have occurred. This length of time is known as the incubation period, and during this time chemical degradation has been observed to occur within a region around the needle tip, known as the degraded or deteriorated region [11, 24]. There is strong evidence of a relationship between this chemical damage and eventual tree initiation however the actual mechanisms behind this remain unclear. This degradation, and the link between it and tree initiation are examined in this work.

1.3.2.1. Stochastic Nature of Initiation Due to the stochastic nature of electrical treeing it is at present impossible to know when initiation will occur. Similarly using a single test sample is not sufficient to describe the behaviour of a dielectric under certain conditions. Instead it is typically necessary to test a number of samples in order to build up an accurate picture of their behaviour and the probabilities of breakdown occurring. By understanding the probabilistic nature of the initiation much can be revealed of the underlying physical causes.

Figure 1-5: Electrical Tree Initiation Times with logarithmic axis [25], using data from [26] showing initiation probabilites over ageing times in excess of 10000 minutes

A typical plot of the cumulative probability of tree initiation occurring over time is given in Figure 1-5. It can be seen that seemingly identical samples will provide a range of times at which initiation will be achieved. A common way of describing such a sample distribution is to give the 50% value (median), the time at which half of the samples will have initiated. There can be issues with this however in that there can be large variations in the time taken for initiation even at the same voltage, as well as significant changes in the 50% value for small changes in voltage.

Such an analysis is made by Dissado and Hill [25], in which they suggest an increased focus upon the minimum time taken for initiation, though they note potential difficulties in obtaining such a value with small sample sizes and sample-to-sample variations. As well as providing an analysis of how such statistical distributions can be discussed, they also suggest a physical basis for this distribution. It is posited the cause of this distribution is the varying electric field strength across the needle electrode dependent on shape and the requirement of ‘characteristic sites’ which are sites suited for charge injection and therefore likely to trigger tree initiation. These sites will not necessarily exist at the point of highest field but may slightly off from this, the reduced field experienced here results in a larger initiation time. There is however little physical evidence directly supporting this theory. Work performed in Manchester does however indicate the potential for significant impact from slight variations within samples. These include variations between samples as well as within samples, in the areas surrounding the needle tip. One key aspect is the air gap which often forms at the interface between the needle tip and the polymer, shown to be impactful in [27]. Indications from tests at Manchester suggest smaller voids, on the scale of micrometres and imperceptible through optical imaging, are common and can also significantly alter tree initiation properties. This may have introduced a level of apparent randomness in tests in which such voids existed but were not detected.

Others have suggested alternative theories. Noto and Yoshimura [16] in consideration of such a distribution suggested a theory involving multiple breakdown mechanisms which each became more or less relevant at different voltage levels, including mechanical strain and joule heating as earlier discussed. Another such theory to explain the stochastic nature of tree initiation is, in cases without sufficient field enhancement for charge injection, the first free electrons required to trigger an avalanche may be produced by collisions with cosmic radiation. This theory however is not yet evidenced. Any complete theory, discussing electrical tree initiation, must account for this statistical distribution; whether it is introduced in the preceding chemical degradation or in the mechanism by which a channel forms.

1.3.2.2. Electroluminescence Observing what is occurring within a dielectric during the incubation period is not a simple task as there is no visible degradation occurring. Chemical analysis can be used to detect degradation (see Section 1.3.2.4) but not while a voltage is applied, and standard electrical analysis techniques such as partial discharge monitoring are ineffective given that the degradation is a PD-free process [23]. One phenomenon which can be used to observe that energy is being transferred to molecules is electroluminescence, a low level light emission which is detected surrounding the needle tip [28-30]. Not to be confused with light emissions related to partial discharge, the electroluminescence has a different origin and is closely correlated with eventual tree initiation [30-32]. The correlation between electroluminescence and tree initiation is observed in a number of factors; localisation, threshold voltage, electrode geometry and magnitude changes with time.

Treeing experiments performed in oxygen-deprived conditions found, using chemical testing (Figure 1-6) and SEM/TEM imaging, that prior to initiation occurring damage was caused to the molecular bonds around the needle tip [11, 24]. This creates a ‘degenerate’, ‘degraded’ or ‘deteriorated’ region around the needle tip which is further discussed later in this report (Chapter 3). Tests have also been performed to identify the extent of the electroluminescence around the needle tip; this can be seen in Figure 1-7.

10 µm

Figure 1-6: Degraded region in LDPE – Imaged using Methylene Blue Dye. [33]

Figure 1-7: Electroluminescence at Needle Tip [34]

That the degenerate region and the electroluminescence are found to occur over similar volumes, as can be observed in Figure 1-6 and Figure 1-7, suggests a link, direct or indirect, between electroluminescence and the material damage which may also be responsible for tree initiation [35].

It has been noted by Shimizu et al [35] that in the DC case neither the electroluminescence nor the degenerate region would form.

In some experiments it has been found there is a correlation between the onset of the detection of electroluminescence and the initiation of electrical treeing [28, 36]. A coincident threshold voltage for these would suggest linked origins and the electroluminescence being related to tree initiation in some form.

There is however some dispute in some claims regarding the electroluminescence threshold voltage and care must be taken that the detected threshold is the true threshold of electroluminescence and not simply the point at which sufficient electroluminescence is produced in order to become detectable. It was found by Champion et al [31] in their experimental setup that the limiting factor was their equipment and the variation in electroluminescence magnitude, which is strongly field dependent, caused the apparent voltage threshold. It must also be considered that while they may be linked, that does not indicate a causal relationship between them.

Champion et al [31] measured changes in luminescence magnitude over time in the lead up to tree initiation in an epoxy resin. Patterns in the light emissions were identified. These are shown in Figure 1-8. They interpreted this as mirroring different processes preceding and ultimately leading up to initiation, and suggested understanding the change in light emission with time may provide evidence as to the processes which underlie initiation.

Figure 1-8: Intensity changes of electroluminescence (showing intensity in different phase quadrants) [31]

Similarly work by Zhang et al [37] also found changes of electroluminescence magnitude with time and expanded upon this attempting to provide physical relevance to the electroluminescence. This identified two forms of electroluminescence occurring. One form was UV emission (300 - 400 nm wavelength) which was understood to be caused by hot electron collisions and which changed with time (and apparent material degradation). The second was termed yellow light emission (500 – 600 nm wavelength) and was understood to be caused by charge recombination. This did not seem to change with polymer or ageing time and was not believed to be involved in material degradation for this reason. This work supports previous findings that the electroluminescence is directly involved in causing tree initiation and provides a much needed physical relevance to it. It promotes the theory that tree initiation occurs as a consequence of direct collisions between electrons and the polymer molecules. To further this work however a more direct link between the types of light emission and the supposed physical sources would be ideal.

1.3.2.3. Charge Injection Charge injection is the transferal of charged particles, such as electrons and positive holes from an electrode into the insulation. It is believed to be a key element in the initiation process, and there is strong evidence linking it to the electroluminescence.

The role of charge injection within a dielectric is to provide energy with which to excite atoms/chemical bonds within the insulation. When these excited bonds relax to lower energy levels they emit photons which can be observed as electroluminescence [38]. It is this process which is commonly thought to be involved in the dielectric degradation, linking charge injection, electroluminescence and the tree initiation process.

Charge injection here is distinct from partial discharges due to the scale of the events. Charge injection can be considered to be individual or small number of electrons injected from the needle tip. Whilst partial discharges are considered to occur when electron avalanches are triggered causing extremely high numbers of charges to strike the polymer walls. Partial discharges require sufficient air gaps in order to occur whilst charge injection can occur over much smaller spaces.

Figure 1-9: Charge Absorbed by system (Experimental vs Theoretical) [40]

The evidence for charge injection in treeing experiments is well described by Tanaka [39] in which they describe a number of experiments which demonstrate its occurrence. One such method involves measuring the charge held by materials of known capacitance. When the charge held by the material exceeds that which would be calculated theoretically due to capacitive effects it indicates an alternative mechanism (charge injection) is active. Such a test was performed by Hadid et al [40] demonstrating the presence of charge injection in electrical treeing and can be seen in Figure 1-9. Another test performed by Tanaka and Greenwood [26] targeted pulses of UV light at the region of dielectric surrounding the needle tip after a large voltage had been applied. UV light gives energy to release the injected charges which have become trapped within the material, inducing a current flow which was associated with the voltage application.

Evidence Linking Charge Injection to Electroluminescence There is strong evidence linking charge injection to electroluminescence; as with charge injection, electroluminescence is primarily associated with AC voltages, not being identified under DC voltages [41], and with impulses and step pulses it is only observed on voltage rise and fall [42] indicating field changes are necessary for electroluminescence (as with charge injection). It has also been observed that electroluminescence has a phase dependency [31] and is most intense with a rising negative field; much like charge injection this would suggest a distinction not just in field strength but between the positive and negative charges and their effect upon the material.

The conclusion which follows is that electrical treeing initiation appears to be dependent upon charge injections, and in this process electroluminescence is also observed. There are however cases in which the applied field should be insufficient for charge injection to occur, yet treeing occurs. In these cases, in order for discharges to occur, there must be a source for initial electrons to trigger the electron avalanches. One such source is believed to be cosmic radiation, the lower probabilities

28 and inherent stochasticity of collisions is posited to explain the time delays and statistical variations in tree initiation times [43].

1.3.2.4. Degraded Region Chemical analysis performed on the area surrounding the needle tip following extended ageing has shown the polymer is chemically damaged over time when sufficient voltages are applied. This gives rise to the notion of a ‘degraded region’ which becomes more susceptible to channel formation over time. The chemical degradation in this region is inherently linked to the electroluminescence and charge injection processes with charge carrier interactions understood to be the cause of the chemical damage [12, 37].

The tests which have been performed on the degraded region to characterise the chemical ageing as well as the processes which are suggested to cause it are discussed further in Chapter 3 (AFM-IR). These will discuss the findings of methylene blue [13], FTIR [11] and Raman spectroscopy [11] testing in this region. These suggest a ~10 µm radius sphere of chemically damaged polymer surrounding the needle tip which builds prior to initiation. This review in Chapter 3 will include the limitations of these techniques and the questions and uncertainties which remain regarding this region and its connection to tree initiation.

1.3.3. Tree Initiation Mechanism Tree initiation is often considered to cover the stages until the formation of a channel in which discharges are able to occur [38]. This channel formation is often believed to occur virtually instantaneously, a channel of typically between 5 µm and 20 µm in length [39] forms where previously there had been solid polymer. Of course the measurements supporting this will always be limited by the time and spatial resolutions of the techniques used to image the polymer in the moments prior to visible tree formation. That this degradation happens so quickly limits the study which is able to be performed upon it.

Tests have been performed looking at discharges which occur both, during and following visible tree initiation. Positive current pulses are commonly reported as having occurred first following tree initiation, [23, 39, 44] reported observing pulse currents in polyethylene which preceded tree initiation and then changed in magnitude on tree initiation. This may suggest discharges and tree initiation are fundamentally linked, and the authors state this evidences that electron avalanches trigger electrical tree initiation. However identifying the moment of tree initiation optically, particularly in a material such as polyethylene in which it is difficult to make clear observations, is not simple. It cannot be ruled out in these cases that tree initiation precedes the occurrence of the

29 discharges, because initiation occurs first on a scale below that which they are able to see. This theory along with others is discussed further in Chapter 3.

1.3.4. Impulse Treeing This review has until now considered electrical treeing through steady state AC sources. These are the sources used in this project and represent a large proportion of the work done in electrical tree testing. However trees grown under impulses should also be noted. Transients involve quick, time dependent events, such as lightning strikes or switching impulses in which the voltage momentarily exceeds the rated voltage, sometimes by significant levels.

Such transients have been commonly proposed as mechanisms by which electrical trees could form in cables which lack obvious defects, such as those used in the laboratory. For this reason they have been well studied and much literature is available on them [45-49]. Key aspects which must be considered with the study of transient events are of course voltage magnitudes [45], but also rise and fall times [47] and polarity [46].

The differences in polarity effect noted by Sekii [46] were attributed to the differences in field experienced by the electron. In particular it was noted positive voltage transients resulted in a significantly larger tree growth. This was interpreted as occurring as the electron would gain more energy approaching the larger field of the point electrode than when approaching the lower field of the planar electrode. The dependence of tree initiation voltages with different rise times are understood to occur due to space charge effects. Space charge formation is the build-up of charges within the dielectric in response to the applied field; this typically has the effect of reducing the effective field around the needle tip. This means the effective field produced by the needle tip is smaller than that which would be anticipated without space charge effects [50]. In DC treeing, and for power frequency AC tests, the effective field applied is reduced. For pulses however, in which the rise times are often significantly faster than the time taken for space charges to form, the dielectric is exposed to the full field and initiation voltages are reduced [47]. It can be seen then impulses are particularly dangerous to insulation due to both their magnitudes, which may exceed rated fields and for their rise times. For these reasons they are often considered in the discussion of tree initiation mechanisms.

A similar process which should be considered is the grounding effect, in which a large voltage is suddenly grounded. This mimics fault events in cables and has been shown in laboratory testing to initiate electrical trees [49, 51]. Again this process is believed to have roots in the formation of space charges in the dielectric. As discussed before, these will often form near to the electrode in response

30 to the applied field. This reduces the effective applied electric field. However, when the voltage is suddenly withdrawn in a grounding event the charges are suddenly acting to create a significant electric field assuming the rate of voltage reduction exceeds the space charge dissipation rate. The large space charge field, coupled with the suddenly neutral electrode can cause an avalanche of charges towards the electrode triggering localised degradation [51].

1.4. Tree Growth

Immediately following tree initiation the channel will begin to grow and propagate through the insulation. This is typically accompanied by discharges within the newly formed channel. The process of this propagation is however not straight forward and is found to be dependent upon many properties both intrinsic to the material and dependent upon the applied stresses and ambient conditions. Changes to the applied field such as the voltage magnitude [14] and frequency [16] or changes made through alterations to the electrode configuration [15] will produce differently presenting trees. So too can dielectric based factors such as the use of different polymers or different treatments to the polymers [17], varying ambient temperatures or humidity [21], the inclusion of filler material [52], and mechanical strains [19].

There are many aspects of electrical trees which are found to change in these different conditions; growth speed, chemistry [53], partial discharge magnitude and location [54], and channel widths [55] among others are known to vary significantly. To truly understand the tree propagation would be to understand why treeing occurs in such a variety of ways. These aspects are discussed here, as well as some of the most prominent theories describing tree growth.

1.4.1. Tree Shape A tree will typically follow a very general pattern; that of a channel growing from a high voltage source to the ground with bifurcations occurring along the way. However, there is significant variation that can occur within this, as can be seen Figure 1-2.

In Figure 1-2 branched trees can be seen as having a few main channels, with a number of smaller channels branching off from them. In comparison to this, the bush trees cover a much larger volume with many channels and branches forming far more regularly. Although a number of different aspects can affect the formation of these trees they are most commonly considered as being voltage dependant with higher voltages causing bush trees to form and lower voltages leading to branch trees [14]. Though perhaps unintuitive, tree growth in bush type trees (at higher fields) is slower than branch types trees (lower fields). This occurring as a consequence of the greater density of channels in bush trees which slows their radial extension [56].

Bush and branched types, describe the large majority of tree samples seen during testing however many others are found to form. These include monkey-puzzle [57], which is similar to a branch tree except that microscopic channels form along the sides of these creating an almost hairy look to the channel, and bush-branch (or branch-bush) trees in which the tree first forms a bush type which then converts into a branch tree [58].

1.4.2. Bifurcation (Branching) That tree channels will bifurcate (and they always bifurcate) is not an immediately obvious phenomenon. In fact identifying why this occurs, providing a physical basis for it and being able to produce models which explain the observation of such a variety of tree structures has been described as one of the greatest challenges in the study of electrical treeing [38].

Bifurcation is found to occur in electrical treeing cases regardless of the material they are grown in or the voltage applied. Though there can be differences between the rates of bifurcation and structures produced in these cases, that bifurcation is ubiquitous across electrical treeing cases encourages the reading that it is an inherent part of the tree growth mechanism.

Trees often appear as fractals, a self-repeating pattern, due to the way they branch. The structure of a tree is often described by its fractal dimension, a measure of the space taken up by the tree in 2D projections. The more branching that occurs the greater the volume occupied. A bush tree will have a higher fractal dimension (2 < df < 3) than a branch tree (1 < df < 2) due to their higher branching rate and so greater channel density. Approaches to study and understand bifurcation in tree have often considered this fractal-type behaviour. Such a model is well described by Dissado [38]. In this paper he describes and discusses models which attempt to cover the “deterministic chaos” of electrical tree propagation. These models were effective in recreating the branching nature of electrical trees. Deterministic chaos here essentially means that while there is no inherent randomness involved in the process of tree growth very slight changes at any point can change the system as a whole. Models often work around this through the introduction of stochastic elements; however these often lack a physical basis which should be avoided wherever possible. The introduction of non-physical factors into such models means they cannot be effectively tested, and where required values can be added, changed or removed to suit results.

In order to provide the best opportunity to understand tree growth, bifurcation should not be treated as an element for which a quick fix can be applied, but should fall out as a natural consequence of the physical models utilised. One model which purports to so is described by Dodd [59] which he states has no stochastic elements involved. This model considered partial discharges

32 and electron avalanches with a dielectric which was divided into segments which each had a critical level of material damage necessary for tree growth into that segment. Tree growth, and bifurcation, occurred when a segment had experienced sufficient degradation for the vaporisation of the material in that segment. Stochasticity here was replaced by the complex interactions of many field strengths and discharge sizes causing damage which would build over time to produce the critical level of material degradation necessary in a segment for tree growth. This allows the seemingly random tree shape to be generated by deterministic means. This does however generate concerns regarding symmetry breaking. The trees in Dodd’s simulation were almost exclusively non-symmetric however the explanation for why this happens was not convincing. If symmetry breaking cannot be adequately explained by the model then it raises questions about potential hidden stochastic elements such as within the computing of the various elements considered. However, this should not overly detract from a model which does demonstrate how extremely slight changes in tree shape or in previous discharges can deterministically lead to seemingly random tree shapes. Ideally this model, which was two dimensional, would be extended to three dimensions for further testing against empirical evidence. This may be computationally prohibitive however.

1.4.3. Stages of Tree Growth Tree growth is not a single unchanging stage in the development of electrical trees; the properties of the channels forming will change along with the tree length. Due to the variations in trees seen across different materials, voltages and ambient conditions there is no singular way of describing them. Nonetheless they can be considered independently, looking to develop an overall understanding of tree growth.

A common picture of electrical treeing is of a tree forming, generating channels which then steadily continue to grow until it reaches a runaway stage of rapid growth as the tree approaches the ground conductor. In numerous cases however stages of treeing have been identified which do not conform to this picture. Properties such as tree chemistry [53, 60], partial discharge magnitudes and phase dependency [18, 55], channel widths [55], tree structure [18, 58] and growth speeds [38] are regularly found to be dependent upon the radial extent of the trees. Also identified are changes of channels with time; such as when, at certain periods of time during tree growth they will be seen to be radially extending, while later the extension may have halted but the channels are observed to be widening, darkening or forming new micro-channels [18]. Each time the properties of the trees change we are observing differences in the underlying process or processes which produce the electrical trees. This may be a temporary change, such as when channel growth halts only to restart

33 after a waiting period. Or it can indicate a more fundamental shift in the nature of tree growth has occurred.

Understanding how it is these different stages of tree growth, whether they are fundamentally different or represent smaller shifts in behaviour, link into the overall picture of tree growth is essential for understanding the underlying mechanisms which make-up electrical treeing. This is extremely relevant when we begin to consider how we take the results obtained in laboratory based experiments and interpret them for the cases of treeing in cables. Laboratory testing is performed with higher fields, smaller voltages, different electrodes and smaller insulation widths. The changes which occur over 2 mm of tree growth are significant and can fundamentally the mechanisms of tree growth. We must ask ourselves what we actually know about the tree growth in dielectrics with thicknesses an order of magnitude larger in cables.

1.4.4. Growth Mechanisms The most prominent theory describing the growth of electrical trees considers it as a product of high energy partial discharges within the channels. These discharges cause material damage to the surrounding walls and wear away the front of the channels causing a gradual extension of the channels as the polymer is eroded. An alternative proposed mechanism is that of fracturing; here strains produced due to either high gaseous pressure within the channels or strong electromechanical (or Maxwell) forces. These strains induce cracks within the insulation which are able to propagate through the polymer with time. With this model discharges would occur as a response to tree growth, not as the precursor to it.

Tree growth is typically accompanied by an associated chemical change [53, 60]; this is typically taken to occur as a product of partial discharges within the channels which cause molecular damage and material vaporisation. This can often be seen in the darker appearance of some channels, which become conductive due to the carbon deposits which have formed within them as a result [53]. Non- conducting channels [53] also have an associated chemical damage associated with partial discharges, however in this case carbonisation does not occur. There are however narrow channels which form, filamentary channels, which do not appear to have an associated partial discharge [20] and for which no associated chemical degradation has been reported. While some have proposed that this is the result of discharges below the detection limit [55] others have proposed that this is evidence of growth through fracturing [61].

The scale of tree growth makes it very difficult to test directly. Tree tips can be sub-micron in size (which creates diffraction limit problems), they occur within the bulk of a dielectric (which greatly

34 limits applicable testing options) and appear to occur over small scales and very quickly. This means while theories can be proposed, these are often limited to utilising indirect evidence or that which is limited by the resolutions to which we can observe tree growth.

1.4.5. Electromechanical Fracturing Electromechanical fracturing is a mechanism of tree growth which has been proposed by a number of authors [16, 61-66]. While not as popular as the partial discharge driven theories as tree growth it has remained relevant in such discussions.

High electrical fields are able to induce strong mechanical responses, the most relevant of which here are electrostatic forces (or Maxwell stresses), the repulsive and attractive forces between charges. It is suggested that as charges gather at the tips of channels under high fields, these experience repulsive forces between each other. This force is applied to the surrounding polymer, causing the material to become stressed. If sufficient force is applied this stress will cause the fracturing of bonds in the polymer, leading to further tree growth.

The electromechanical fracturing theory has remained popular as it has shown an ability to predict many aspects of electrical treeing [62]. Importantly it also explains in a most direct manner the strong dependency of tree growth upon external mechanical stresses [66, 67] which partial discharge based models cannot do so easily.

Mechanical fracturing induced tree growth or initiation dates at least back to 1974, suggested by Noto and Yoshimura [16] to in particular explain tree initiation. They calculated Maxwell stresses using the equation

2 P = (1/2)εEmax .

A stress (at 7 kV) of 23.8 kg/cm2 is obtained, which compared to a tensile strength of polyethylene of 90 kg/cm2 at 20oC. They postulated repeated mechanical stresses would induce fatigue leading to tree initiation. It is also possible (although perhaps not identified at this time) chemical degradation occurring due to charge injection could weaken the surrounding material sufficiently to allow the Maxwell stresses to induce tree initiation. Shimizu identified mechanical weakening was a consequence of the chemical degradation in a polyethylene sample [11].

Since the publishing of Noto and Yoshimura’s work numerous models have been considered which discuss electromechanical fracturing. Zeller [64] in particular further developed work in this region, introducing more formalised fracture mechanics. This work balances the electrostatic energy (Wes) which is released by a fracturing against the formation energy required (Wf), a combination of the

35 plastic deformation energy and surface energy. If the electrostatic energy is larger than the formation energy required then this will result in crack propagation and tree growth.

The electrostatic energy (Wes) here is ultimately determined by the density of charges at the tip of the channel which will depend crucially upon the applied voltage and the radius of the channels. Smaller radii will increase the charge density and with this the electrostatic energy. It may be worth reflecting at this point upon filamentary channels, discussed in Section 1.4.3, which are also commonly described as fine trees due to their small size. Their growth is commonly proposed to be driven by fracturing, due to a lack of observable discharges. The small size of these channels would make electromechanical fracturing within them more likely according to this theory.

Fothergill [61] considers a similar but still distinctly different theory, in this model the fields at the tips of electrodes are believed to be significantly higher due to differences in space charge models. The result of this is that the dominant mechanical strain is understood not to be the repulsive forces at the tip of the channel but the attraction between the channel tip and the counter electrode. This bears strong similarities with work by Stark and Garton [68] in which the attractive forces between electrodes were found to have produced mechanical breakdown in a dielectric. A potential issue with Fothergill’s model is although it predicts small radii channels (0.1 – 1 µm), which are observed in fine trees, these are anticipated to have large fields and hence to generate discharges. Tests upon fine trees in epoxy have not observed a related discharge, though one may argue the discharges are simply below the observable level. These limits are typically 1-5 pC but tests have been performed down to 0.5 pC [69].

Lewis et al [70] take a different approach, considering the Lippmann electromechanical equation which describes changes in interfacial tension under high fields. They propose that under high fields the dielectric at all points has a repulsive force in all directions orthogonal to the electric field (as illustrated in Figure 1-10).

Figure 1-10: Tensile stress orthogonal to the electric field. Taken from [70]

In bulk these forces will work against each other, but are unable to tear the material apart. However at cracks, or channels, these forces will be concentrated at sharp points, creating significant mechanical stresses which can further drive tree growth. The direction of growth in this case will be a consequence of the applied field and distortions of this due to space charge effects, the material properties which in a polymer such as polyethylene will be heterogeneous and the starting direction of cracks/channels in the polymer. Crine [62] takes such ideas further, proposing that the repulsive forces under high fields are able to cause degradation on molecular levels, breaking the ‘attraction bonds’ between them. This would then lead to microcavities forming within the dielectric which then further develop into larger fractures.

There are many theories which describe electrical tree growth as occurring through electromechanical forces. While such theories may be able to accurately predict many aspects of electrical tree growth they suffer for lacking direct evidence that fracturing does actually occur in tree growth. Partial discharges on the other hand are observed to occur in virtually all forms of electrical treeing in all materials. Whilst electromechanical fracturing lacks empirical evidence of its involvement in electrical treeing it will struggle to reach the pervasiveness of theories based upon discharge driven growth.

1.5. Aims and Objectives

The primary aim of this research is to understand the mechanisms which drive the formation of electrical trees. Considering in particular, whether more representative alternatives can be developed to the standard needle-plane configuration and the dielectric ageing which occurs before and along with electrical treeing. A more complete understanding of these topics will provide a better ability to translate findings in such tests to the conditions in which electrical treeing occurs within cables in practice. The objectives of this work are to investigate:

 Alternative electrode configurations for electrical tree growth which do not utilise the needle-plane configuration.  The chemical degradation which precedes electrical tree initiation, understanding where and when this forms. Applying new techniques to the study of this.  The mechanisms which are responsible for tree growth and formation in AC trees.  The changes in channel properties as trees propagate through the insulation.

1.6. Thesis Summary

This thesis contains 7 chapters:

 Chapter 1: A review of previous work on electrical treeing considering; the impacts of electrical trees upon cables, common electrode configurations used in electrical tree testing and concerns regarding these, tree initiation and the observables related to this, and tree growth and the properties of different types of trees.  Chapter 2: The methodology for producing needle-plane samples in this project is discussed. Three distinct sample designs without needle electrodes are then reported upon; fine treed, interfacial tracking aged and water saturated. These are tested to see if tree growth or localised degradation can be made to occur using them.  Chapter 3: The AFM-IR (atomic force microscopy with infrared spectroscopy) technique is discussed providing background information of previous tests from the literature and the sample preparation procedure used in this project for the AFM-IR technique. The AFM-IR provides very high resolution chemical analysis well beyond that of previously applied techniques.  Chapter 4: Results from tests using the AFM-IR technique on electrical treeing samples are reported. These cover epoxy samples which are in different stages of ageing. Pre-initiation, post-initiation but focused on degradation around the needle tip, and mature channels are tested.  Chapter 5: A discussion of the findings from this project. A model is proposed describing tree growth at different stages of ageing.  Chapter 6: The conclusions and findings of this work are given, providing reference to the results in the text.  Chapter 7: A consideration of topics which this work suggests should be further investigated.

2. Plane-Plane Sample Testing

Standard electrical tree growth experiments use the needle-plane electrode configuration; this is chosen as it produces reliable and consistent electrical tree growth on relatively short time frames. The standard needle-plane configuration is discussed in Section 1.2.2 of the Introduction. Cables however do not have reliable and consistent sharp points within them and evidence for trees forming in this manner is lacking. Furthermore it is known the electrode configuration can have significant effects on the growth of electrical trees [15, 17]. While recent work by Zhang et al [37] found the electroluminescence occurring in divergent and uniform fields differ.

The work described here, investigating alternative methods for tree growth which are more comparable with the conditions experienced within cables, is intended to understand the differences which may exist between the electrical trees produced in laboratory testing and those which form in the field. In addition this should generate a better understanding the conditions truly necessary for electrical tree initiation.

This chapter will discuss attempts to develop alternative mechanisms of electrical tree initiation and growth which are more comparable to conditions which may be experienced within a cable. These include reverse tree initiation testing, electrical tracking testing and water absorbance testing. The rationale for these methodologies is discussed in each section. They are all however based upon conditions under which electrical trees are known to form without an obvious sharp point being present.

2.1. Fine Tree Initiation Testing

2.1.1. Introduction This section will discuss the testing of planar samples containing fine or filamentary electrical trees. These are channels which commonly form within epoxy during treeing tests. Unlike more standard trees they do not have an observable partial discharge associated with their growth and crucially when they bridge the insulation between the conductors they do not immediately create a short circuit and breakdown channel. Instead a ‘reverse tree’ is formed from the planar electrode which grows towards the needle electrode to trigger breakdown. These channels are discussed in further detail in the following sections.

The aim of this work was to use these finer channels with planar electrodes to attempt to mimic and recreate the response of a channel forming from a planar electrode, and so better understand the conditions necessary for electrical tree growth.

2.1.2. Background Reverse treeing is the name which has been given to electrical tree growth which, as a response to a forward growing electrical tree, occurs in the opposite direction. These have been identified and discussed in [20, 55] and are illustrated in Figure 2-1 (circled).

The causes and mechanics of reverse tree growth are not yet clear. Within the epoxy samples tested at The University of Manchester it appears to grow as a response to fine tree growth. The fine trees themselves are unable to short circuit the insulation on bridging the insulation between electrodes.

2 mm

Figure 2-1: Grown in epoxy, fine tree growth leads to reverse trees forming (circled). Growing from the planar electrode towards the needle tip. Breakdown typically occurs when the reverse tree bridges the insulation.

While a definitive explanation has not yet been established for fine tree growth (nor does this project intend to do so) electromechanical stress based models for electrical tree growth have been proposed [61, 62, 64]. They grow from the high voltage, point electrode towards the grounded, planar electrode. The channels closest to the needle tip are often dark and thicker (suggesting carbonisation and partial discharges are active in this region). However, further from the needle tip, these channels become finer (around 500 nm in diameter). PD measurements have found no evidence of partial discharges (with sensitivity of 0.5 pC) in these stages of growth, suggesting either their growth is not a by-product of electrical discharges or the discharges are so small as to be below the lowest sensitivity of the detection equipment.

The mechanism by which a reverse tree is formed is not clear, though suggestions for it are discussed in [20]. It is observed in this paper that reverse trees may form when a fine tree channel is in contact with the planar surface, or one may form before the fine tree makes contact (from a breakdown channel that bridges from the ground plane to the nearest channel).

Of particular interest in this study is that reverse trees form from planar electrodes, these seemingly lack the divergent fields which are commonly believed to be necessary for tree formation. It is not believed reverse trees form from contaminants or imperfections upon the planar surface, these have never been observed and attempts to frustrate reverse tree growth through careful sample preparation have proven unsuccessful. This then raises the question of why and how they are able to form. It also provides a path of investigation, studying reverse tree with the aim of furthering our understanding of more general electrical tree initiation.

While reverse trees form from planar surfaces, they do so with a high field point electrode exposed to high voltages still within the insulation. This can be anticipated to continue to impact the system, such as through providing a source for charge injection. To better understand the mechanisms behind reverse tree formation it was decided to attempt to initiate a reverse tree using only planar electrodes. This removes the high field needle electrode which may still affect the reverse tree formation. The use of planar electrodes provides a uniform field in the absence of inhomogeneities or space charge in the dielectric. This electrode arrangement more accurately represents the process in cables and could be sufficient to provide an alternative testing regime for electrical treeing.

2.1.3. Methodology

2.1.3.1. Sample Preparation This section will describe the sample preparation used during this project from beginning to end. The fine trees are first grown using a more typical needle-plane configuration and then cut down to

41 produce the flat, planar samples. The epoxy curing process will be discussed first. Samples are then electrically aged to produce the fine trees and this will be described including the experimental configuration which is used. The processes by which the fine trees are sectioned and polished into flat samples will then be described. This will then be followed by the experiment configuration for planar samples including a brief overview of its development.

Epoxy Curing Process The process used here is to produce samples in the needle-plane configuration, necessary to produce fine trees. The needle electrodes and sample preparation process used here has been commonly used in Manchester electrical treeing tests for years.

An unfilled epoxy resin is chosen as the dielectric to be tested. The degradation of epoxies under different stresses has been widely tested, which makes them suitable for chemical analysis and comparisons. Epoxies are also regularly used in electrical treeing tests, which again allows for comparisons between the findings in literature and the trees and degradation observed in this project.

Epoxies are commonly used as the dielectric in electrical treeing tests, due to a combination of strong dielectric performance, with low dielectric losses and high breakdown strengths, and a relatively simple and consistent curing process. Another significant reason to use unfilled epoxies over other materials such as polyethylene is due to their optical transparency which enables more accurate imaging of the electrical trees.

The epoxy used here is the Araldite LY 5052 and the Aradur HY 5052, both produced by Huntsman. The epoxy is Novalak based; containing an ether diluent while the hardener contains a mixture of polyamines. This provides high thermal and chemical stability. It has a glass transition temperature of 120oC after post curing. All testing in this project is performed with epoxy in the glassy state.

The cured structure of the epoxy is not known precisely, however we can consider the structure of the constituent Araldite and Aradur based upon material data sheets and studies from literature [71, 72]. These are given in Figure 2-2 and allow us to understand aspects of its structure. This figure illustrates the complexity of epoxy chemistry and the related challenges involved in understanding its ageing processes.

Figure 2-2: The chemical structures of the molecules which make up LY 5052 epoxy resin and HY 5052 Hardener [72]

The Araldite and Aradur are mixed according to the stoichiometric ratio suggested by the manufacturer (100:38 by weight). The mixing of these materials creates an exothermic reaction and

43 curing will begin at room temperature meaning that the process of preparing the samples is time limited.

 The mixture is first stirred within a beaker by hand for 3 minutes and then by use of a magnetic stirrer for 7 minutes to ensure they are well mixed.  The beaker is then placed within a vacuum chamber for 30 minutes to remove air pockets formed during the mixing process. As discussed previously, voids within the dielectric are known to increase treeing susceptibility and will also reduce repeatability of testing.  After 30 minutes the epoxy is poured into acrylic cubes (25 mm sides) and needle electrodes are inserted into the mixture. The needles are then held with the tips 2 mm above the sample base, measured using a stopper. The needles are made from steel and manufactured by Ogura with 3 µm radius tips, shaft diameters of 1 mm and a tip taper radius of 30o.  The cubes are then placed back inside the vacuum chamber for 35 minutes before being removed and left to cure at room temperature and pressure for 24 hours. The cubes are removed from the vacuum chamber at this point as the vacuum allows large air pockets to form within the samples. As the epoxy cures it becomes more viscous and these will remain if left to cure for too long.  After curing the samples are then post-cured. They are held within an oven for 4 hours at 100oC, after 4 hours the oven is switched off and the samples are allowed to cool slowly within the oven. If removed too quickly then the rapid thermal changes could place strains upon the sample due to thermal contraction.  The sample is then kept within a desiccator at room temperature for at least a week prior to testing, removing water ingress. This waiting period important as samples tested immediately after post-curing are noted to behave differently to those which have been stored for a length of time. This is most likely due to mechanical stress formed during the curing stages, a belief shared by Hepburn [73]. Though alternative explanations include further curing occurring over time.  Planar electrode coatings can then be added to the samples to ensure better connections. A number of different materials were used in the course of this project, as well as attempts with uncoated samples.

Prior to any testing the samples are removed from their acrylic cubes and their surfaces are polished to improve the quality of the optical imaging. A prepared sample, prior to final polishing is shown in Figure 2-3. In total this process would take around 10 days before a sample is ready for electrical ageing.

25 mm

Figure 2-3: Cured epoxy sample in the needle-plane configuration (2 mm gap from needle tip to base).

Circuit Design The epoxy samples are now ready to be aged electrically. This is not to be considered here as part of the testing of the samples but is instead another stage of sample preparation. Nonetheless the mechanism is typical of that commonly used in other treeing tests. It is a well-used system which has been active for years of electrical treeing study at this university.

The apparatus used are described in the circuit diagram and picture in Figure 2-4 and Figure 2-5 respectively.

Figure 2-4: Circuit diagram for high voltage electrical ageing of epoxy samples.

HV Supply

Dummy Sample

Test Sample CCD MPD Camera 600

Figure 2-5: Picture of experimental setup (corresponding to circuit diagram), including labels.

The voltage supply comes from a 30 kV high-voltage amplifier, connected to the sample through a 10 MΩ resistor and copper piping. The voltage is measured at multiple points to allow for any potential fault insertion to be identified. Measurements are taken from the HV amplifier output at (1000:1 ratio) and through a 10000:1 voltage divider.

The voltage is then connected to both the test sample and a dummy sample which are connected via a balanced circuit to the ground. The dummy sample is similar to the needle tip electrode except that its needle has a rounded head (as shown in Figure 2-6), reducing the electric field created at the surface. The dummy sample is designed to be similar in terms of capacitance to the real sample while being partial discharge free. The discharges detected in each sample are collected in the balanced circuit which subtracts the dummy sample discharges (which should entirely come from sources prior to the samples) from the real sample discharge.

The comparison essentially allows any noise from the power supply (which will be equal for each sample) to be negated by the balanced circuit when measuring for partial discharges. This dummy sample has been used for thousands of hours and shows no evidence of electrical tree formation or other partial discharge sources having appeared. During testing it is observed as providing low noise partial discharge measurements.

a) b)

1 mm 1 mm

Figure 2-6: a) Schematic of test sample with sharp tip (3 µm radius), b) Schematic of dummy sample with rounded head (0.5 mm radius). Epoxy is not to scale, see Figure 2-3.

Figure 2-7: The UI (User Interface) for the OMICRON partial discharge detection system. Discharges are shown as dots on the graph on the left. Their magnitude is recorded and shown on the vertical axis. Their phase is shown on the horizontal axis in comparison to the green, voltage line.

To measure the partial discharge signal, the output from the balanced circuit is passed to the MPD 600 made by OMICRON. This allows the number, size and phase of discharges to be recorded (Figure 2-7).

The experimental setup is capable of providing very low noise signals, on the order of 0.3 pC. It is possible certain types of tree channel produce discharges of smaller sizes than this; our understanding is limited by technical capabilities in studying such small discharges. In particular it has not been clearly determined whether fine tree growth is associated with partial discharges or whether they are just too small to be detected on the systems used to study them. In this case studying their growth using partial discharges is not necessary as their growth can be tracked optically. A crucial part of understanding the size of discharges being produced is careful calibration of the equipment being used to measure them. To ensure this the MPD 600 is regularly calibrated by used of a known PD signal (from a CAL 542) attached across the sample.

A CCD camera is used to capture optical images of tree growth; this is facilitated using a lamp situated behind the sample producing a projection of the needle tip and tree channels which are captured by the CCD camera. This can be seen in Figure 2-8.

2 mm

Figure 2-8: Projection of electrical trees in epoxy, captured on CCD camera with a backlight.

It can be seen in Figure 2-8 the thicker darker channels are clearly visible using this set up, and we are also able to see the ‘fine’ channels. This does however demonstrate clearly that while visible, the fine channels are difficult to clearly distinguish here. The transparency of unfilled epoxies is crucial to be able to observe fine channels and without an accompanying partial discharge signal does raise the question on why they are observed in epoxy and not in polyethylene which is commonly studied. It could be they do not form in polyethylene or it may be they simply cannot be observed in the less transparent material.

Needle-Plane Electrical Ageing

Using the needle-plane experimental setup the samples are subjected to an AC voltage of 15 kVpk which is chosen as a voltage at which fine trees have been found to develop in large amounts. In general, larger voltages produce more of the darker, thicker trees while lower voltages will lead to fine trees. In cases in which electrical trees could not be initiated at 15 kVpk they would be subjected to a short period of increased voltage until tree initiation occurred. Upon initiation the voltage was then reduced to 15 kVpk for the channel growth stage. The process of increasing the voltage for initiation alone is not believed to impact the properties of the fine channel growth. However, it is an

49 area which should remain under review. There is evidence suggesting properties at the needle- dielectric interface can influence tree growth [74].

Tree growth can then be tracked electrically and optically, with the voltage removed when the fine tree growth has achieved the desired state. This process would take around half a day.

Sectioning of Samples Following electrical tree growth, the fine tree sections of the samples are then sectioned. Producing planar samples as in Figure 2-9 in which the dotted lines show cutting lines. This was done using a cutting machine, of which two were used. These were the Brilliant 200, which was manually operated, and the Brilliant 220, which was automatic. Both machines are manufactured by ATM. These work by spinning cut wheels at extremely fast speeds (~2000 rpm) allowing the material to be ground away and the cut wheels to effectively cut the sample into two pieces. The cut wheels were from MetPrep, Superfine Abrasive Cut Wheels, Type HNF 10 99 03. They were 150 mm wide and 0.45 mm thick. The thickness of the wheels had to be taken into account in choosing where to cut along the samples. When looking to produce a 2 mm section the 0.45 mm of material which would be ground away by the cut wheels must be accounted for.

The samples were held in place by clamping and precisely aligned using a manual alignment wheel. As the cut wheels were spun and cut into the material they were cooled using a cooling fluid which also acted as corrosion protection for the cutting machine. This was made of a water soluble oil (MetPrep Cool Advanced Cutting Fluid). If the rate of flow on the fluid was insufficient then this could cause overheating and burning of the samples. The speed of rotation of the wheels, and the rate of cutting could be set on the automatic cutting machine (Brilliant 220). On the manual (Brilliant 200) the rate of cutting can be altered by moving more slowly if required however the disc rotation speed is fixed. This was not typically impactful when cutting through epoxy, however if you were to cut through the steel needle then this would have to be accounted for and could lead to the shattering of the cutting wheels (or damaging of the samples) if not done correctly. The impact of the oil upon the testing is a potential concern however as it is water soluble and the samples are later exposed to water this was considered to be unlikely to remain within the sample. The greatest risk however is that it is able to penetrate further within exposed trees than the water is. Ideally in future testing a sample manufacturing method without this would be employed.

The exact cutting positions for different samples could be varied. As this work was quite exploratory a range of sample thicknesses, levels of fine tree growth and proximity to the needle tip were considered. The typical types of cut aimed for are shown in Figure 2-9.

Figure 2-9: Schematic of samples grown in needle-plane configuration. Dotted lines show different example areas for sectioning.

Polishing of Surfaces The surfaces of the samples once cut required smoothing. Whilst they were not overly rough the aim of this work was to demonstrate the effect of the use of a planar configuration. Ensuring field enhancing points were not introduced on the rough surface was crucial. The samples were smoothed using a polishing process with silicon carbide paper and with a diamond compound paste. For this a grinding polishing machine is used, an example of these is given in Figure 2-10. This consists of a wheel on which the polishing papers would sit; they would adhere to the wheel when a layer of water was on the surface. The wheel, and so the polishing paper, would then be spun, the speed of which could be set on the controls. The sample could then be polished; a constant flow of water is used, this acts as a lubricant and allows for the waste material to be removed.

The silicon carbide grinding papers come in a variety of grit sizes, with larger grit sizes indicating smaller diameter particles on the paper. When polishing this should be done from smallest to largest grit sizes, giving an increasingly smooth surface as the process continues. When switching grinding papers an effort is made to clean the sample with the water to prevent cross-contamination of the particles. The papers used have grit sizes ranging from 106 microns (P120) – 2.5 microns (P4000). To further polish the surface a diamond compound polishing paste is used.

Figure 2-10: OmegaPol grinding and polishing machine

For the diamond polishing pastes, similar grinding polishing machines are used with a polishing cloth adhered magnetically to the wheel. A monocrystalline diamond compound paste containing particles of a well-defined size is applied to the cloth. As with the grinding papers, multiple pastes are used containing different particle sizes, in this project particles sizes of 3 micron, 1 micron and 0.25 micron are used. By again spinning the wheel at speed the epoxy surface can be polished further. By this method a very smooth and uniform surface is made for the planar samples, providing confidence any discharges or degradation occurring is not as a result of field enhancement points. The sectioning and then polishing process takes a full day for a full day for a set of samples. This is followed by further desiccation to remove added water.

The total time required to prepare a sample to this stage is around two and a half weeks.

Planar Sample Testing Apparatus The experimental apparatus is described in Figure 2-11 and Figure 2-12.

12 mm

200 mm

Figure 2-11: Schematic of experiment design for planar samples.

Figure 2-12: Photo of experimental design showing planar samples within Perspex box and under silicone oil.

The experiment is performed within a Perspex box; this primarily acts as a container for the silicone oil (used to prevent flashovers occurring outside of the epoxy samples). A lid was created for the Perspex box, the lid has a dual role in the experiment. Firstly it helps to keep the oil clean and contaminant free; contaminants reduce the effectiveness of the oil as an insulator and can also pass

53 in front of the camera, blocking any imaging at worst or at least reducing the image quality. The lid also acts to hold the electrodes in place, the copper piping passes through precisely cut holes which limit the lateral movement of the electrodes.

The circuit diagram is identical to the point-plane tests (given in Figure 2-4), with the only difference being the replacement of the needle tip electrodes with flatter, planar electrodes. The HV supply is again delivered via copper piping (from the same HV amplifier) which connects the steel electrodes. The electrodes are designed to produce a uniform field through the epoxy sample (and an unaged dummy sample), going through to the brass ground electrode. These are connected to the balanced circuit and through this the MPD 600. The operation and measurement is the same as that described previously with point-plane treeing.

Optical Imaging Optical imaging is performed again by CCD camera with a light behind the samples. However, there are difficulties in imaging such thin samples when compared to the larger samples used in point plane testing with a needle electrode. These include having the top and bottom surfaces close together creating reflections which obscure the actual image. As well as having a wider area from which initiation may occur instead of a needle tip. This makes it more difficult to know where any degradation would occur and to focus the camera on this location.

Steps were taken to minimise these issues such as careful alignment and contrast/exposure adjustment of the camera and software however the imaging achieved of planar samples remains less clear than that of point-plane samples. It was also found to be extremely sample-dependent; some producing relatively clear images while in others the channels were barely distinguishable. Some examples of optical images taken from planar samples are given in Figure 2-13.

a) b) c)

2 mm 1.5 mm 2 mm

Figure 2-13: Three samples containing fine trees of different optical clarity. a) is taken using a lower magnification than b) and c). This provides a lower resolution but wider field of view. In b) fine channels can be seen. In c) due to the quality of the sample very little can be observed. The dashed lines indicate the epoxy-electrode interfaces.

After testing the different magnifications and settings, most images were taken at a higher resolution. It was felt the benefits of a wider field of view (Figure 2-13a) were not sufficient given the lack of resolution and inability to see fine trees at all.

Sample Coatings The samples were initially coated in aluminium giving the uniform fields through the samples. To coat the electrodes in aluminium an evaporative source was used, pictured below (Figure 2-14). This uses a thermal filament coil (a coiled wire capable of reaching and withstanding high temperatures as large currents pass through it) as a heat source and an aluminium wire is wrapped around it. The coating is performed in a vacuum (to aid the evaporation and produce contaminant free coatings) and is performed on both sides of the sample.

The first sample configuration tested is shown in Figure 2-15a. The full coating of the sample can be seen here. In this image the electrode has a flat face. This is later rounded to reduce risks of corona driven noise and sample damage as well flashover possibilities, as shown in Figure 2-15b.

Figure 2-14: Aluminium sputtering machine, works under vacuum with thermal source.

a) b)

2 cm 2 cm

Figure 2-15: a) Photo of a planar sample between a (steel) top electrode and a (brass) bottom electrode, b) Photo of the top electrode having been rounded at the edges to prevent discharges

Figure 2-16: Typical partial discharge signal from samples without electrical trees

Testing using the planar-plane configuration was found to produce significant partial discharge noise (an example of this is given in Figure 2-16), the source of which was initially unclear. Here noise reached around 300 pC, so discharges from the sample would be impossible to identify.

Following extensive testing and adaptations to the sample design the discharge noise is believed to have arisen from vibrations which occur during voltage application. These affect the contact between the top electrode and the top layer of the sample. Sample design changes made to address this include the use of weighted electrodes, clamping, changing the sample coating thickness, the area covered and coating material as well as removing the coating entirely. These had an impact upon the noise, however these were unable to reduce this noise to a level which would allow careful

56 study of discharges within the samples. For this reason it was decided to perform the testing noting discharge measurements could not be relied upon.

2.1.4. Fine Tree Testing Results Tests performed attempting to produce ‘reverse treeing’ through planar electrodes changed a number of factors but primarily focused upon the distance of the fine trees from the surface. Initially tests were performed using a large distance between the trees and the planar surface (roughly 0.5 mm). But with no evidence of tree growth or initiation this distance was reduced to increase the number of channels at or near the surface. Continued testing however found no positive indication of standard or reverse tree growth or initiation.

The first sample tested was approximately 1.8 mm in thickness. Dark channels are seen to grow approximately 1 mm into the sample. This was imaged in Figure 2-17.

The fine channels in Figure 2-17 can just be discerned. The channels however do not reach the top surface of the epoxy. The limited extent of the tree in this sample was helpful from an imaging perspective because a new channel growing to bridge such a gap from the opposite plane electrode would be immediately apparent. However it was considered that such a large gap between the fine tree and the plane electrode opposite was less likely to trigger tree growth than a smaller gap. For that reason a smaller gap was preferred in tests from this point on.

2 mm

Figure 2-17: Optical image showing electrical trees within a planar sample. Taken at a lower resolution the fine channels are not clear in this image. The dashed lines indicate the epoxy- electrode interfaces.

A sample was then tested in which a large number of channels had bridged most of the insulation. Taken using the higher resolution optics a slightly darker region of channels can be seen in Figure 2-18 towards the top of the sample. A small number of these had penetrated the top of the surface here (confirmed by optical microscopy) although there was no evidence of reverse treeing having occurred.

Further attempts were made using samples in which channels bridged the insulation gap, including through grinding and polishing the surface down to expose significant numbers of fine trees. This would however have the effect of ensuring the potential sites for tree initiation covered a wider area, making it even more difficult to track optically. At no point was there any evidence of new tree growth, of either the existing fine trees or a new reverse tree.

Other aspects tested for reverse tree growth included the orientation of the sample. Tests were performed with both the fine treed side and then the side with darker and thicker channels exposed to the high voltage electrode. This was to account for potential differences in charge injection and is illustrated in Figure 2-19.

a) b)

1.5 mm 1.5 mm

Figure 2-18: In a) Optical image in which the fine channels close to the surface can just be seen, one is pointed out by an arrow. In b) the image is altered to make the channels more visible, however they remain difficult to see clearly. The dashed lines indicate the epoxy-electrode interfaces.

a) b)

Figure 2-19: Schematic of different testing orientations. a) The darker channels are exposed to the high voltage electrode, with the fine trees closest to the ground. b) The fine trees are approaching the high voltage electrode.

In addition different thicknesses of sample were used. Generally decreasing in size as testing went on. Thinner samples were less clear when imaged, however they would experience higher fields.

Sizes ranged from 2 mm to 1.3 mm, and typical voltages of up to 10.6 kVRMS.

After a significant amount of testing, tests were stopped. It was considered that consistently decreasing the thickness (correspondingly increasing the field) and with longer ageing treeing may occur. However, the conditions of the testing were such that identifying tree growth or initiation either optically or through partial discharge monitoring was unlikely to be effective. Only in the event of major tree growth would these identified. Alternative testing designs were then considered.

The electric fields present within the samples during voltage application are inherently very difficult to calculate. The behaviour of the material is unknown; it will depend upon the precise chemical make-up of the channels, their size, density and locations.

In order to gain some understanding of how the fields may be affected we can make some assumptions and simplifications.

 Fine treeing is assumed to create changes in bulk but individual tree branches are not considered. This is not unlike in adding nanoparticles where bulk effects are considered to be far more important than the effects of individual particles.  The fine trees are assumed to form a region of altered material which extends a set distance from the electrodes. This is done to simplify calculations.  The effects of the fine trees are modelled in terms of permittivity changes. Other effects such as high conductivity, space charge accumulation and partial discharges are not considered here. This is not believed to be unreasonable as there is no evidence of these changing due to fine channels in the planar electrode configuration.

Figure 2-20: Model used to calculate the effect of fine trees on permittivity of dielectric

This model is illustrated in Figure 2-20.

The field in the undamaged region will be expected to follow the equation:

휀푉 푉 퐸 = = 휀 푑 + 휀 푑 휀 푑 + 푑 휀

Where E1 and E2 are the electric fields within the fine treed and undamaged regions respectively. If we assume a maximum voltage of 15 kV, an insulation thickness of 1.5 mm and a fine treed region of 1.3 mm then we obtain the equation:

15000 퐸 = 휀 0.0013 + 0.0002 휀

It is seen here the relative change in permittivity due to fine treeing is crucial in deciding the field present within the different regions. The effect of the fine trees on the permittivity is not yet known, however we can calculate the potential effects across a range of possibilities.

If there was no permittivity change (ε1 = ε2), this would mean a field in the undamaged region of:

15000 15000 푀푉 푘푉 퐸 = 휀 = = 10 = 10 0.0013 + 0.0002 0.0015 푚 푚푚 1∗ 휀

10 kV/mm is less than the breakdown strength of epoxy, 25-45 kV/mm, and well below the field created at a 3 µm needle tip which is approximately 800 kV/mm (before space charge effects are considered, which will be considerable when using a needle electrode). It is however comparable to fields which will be present in cables.

If there is a minor permittivity change (ε1 = 1.1*ε2), this would mean a field in the undamaged region of:

15000 15000 푀푉 푘푉 퐸 = 휀 = = 10.86 = 10.86 0.0013 + 0.0002 0.00138 푚 푚푚 1.1 ∗ 휀

If there is a significant permittivity change (ε1 = 2*ε2), this would mean a field in the undamaged region of:

15000 15000 푀푉 푘푉 퐸 = 휀 = = 17.65 = 17.65 0.0013 + 0.0002 0.00085 푚 푚푚 2∗ 휀

These fields are calculated without space charge effects which can be expected to reduce the effective field sizes. The formation of reverse trees in needle-plane samples appears unlikely to be due to bulk permittivity changes and instead a more localised phenomenon.

With no evidence of tree growth or initiation from samples utilising pre-stressed, fine-treed epoxy an alternative method was tested. This utilises interfacial tracks from which electrical trees have been observed to form.

2.2. Interfacial Tracking Tests

2.2.1. Introduction This section will discuss tests which built upon electrical tracking work performed by Bastidas within The University of Manchester [9, 75]. Interfacial tracking is when discharges and damage occur along the surface of a material, or in this case in the interface between two dielectrics. Interfacial tracking is of particular concern at joints, which are the weakest areas in a cable, vulnerable to contaminants and imperfect connections [76].

Bastidas’ tests [9] found trees were able to initiate from tracks formed between two interfaces. While the tracks came from a field enhancement point, the trees themselves did not, instead forming from the tracks. It was notable the trees would grow perpendicular to the field from the electrodes as illustrated in Figure 2-21 and imaged in Figure 2-22. It was decided to attempt to replicate this with planar electrodes connected to tracks which had already formed. The aim of this was to produce an experimental technique which did not rely upon field enhancements generated by metallic electrodes to test electrical treeing. This would also be very comparable to joints, a major weak point in cables.

Figure 2-21: Illustration of trees initiating from interfacial tracks.

Figure 2-22: Electrical tree growth perpendicular from tracks in epoxy at different pressures. a) 0 kPa, b) 20 kPa, c) 40 kPa, d) 60 kPa. Taken from [9].

This section will briefly discuss why tree growth is believed to occur from tracks and then the methodology of the tests before discussing the results obtained from them. Ultimately these tests

62 did not produce any evidence of this being a viable method for tree growth without an additional field enhancement point.

2.2.2. Background Tree growth from interfacial tracking was previously studied by Kobayashi et al [10]. They identified the trees tended to grow further into the dielectric with the higher permittivity, which is the property they assigned the most impact to. This was done by using multiple layers of material on each side of the interface as in Figure 2-23.

The EPR on each side is identical however the materials beyond these (Glass and EPR) would change the direction of the tree growth encouraging it to grow to the high permittivity (glass) side. Their work also suggested mechanical factors such as material brittleness, and stresses and strains may also affect tree growth from interfacial tracks. Bastidas [9] found tree growth was most likely when high external mechanical pressures are applied to the tracks. With a silicone rubber and epoxy interface it was found tree growth favoured growing into the epoxy. The mechanical pressure dependence of tree growth was believed to be due to the higher pressures creating a better interface contact and so inhibiting track growth.

There has not been a significant amount of work in the area of tree growth from tracks and the responsible mechanisms are not known. It is possible one of the reasons is that when discharges are occurring through them they are filled with many free electrons travelling with high energies. A significant question in electrical treeing is how starting electrons are able to be injected into the bulk to begin electron avalanches [38]. In the plasma and electron rich environment of the interfacial track this may occur along any sharp regions or voids between the interfaces.

5 mm

Electrode (Al foil 5 µm) Figure 2-23: Experimental design for electrical tracking tests showing the preference for trees to grow into higher permittivity materials. Adapted from [10].

2.2.3. Methodology The testing methodology for this test followed that from Bastidas’ experiments as described in [9, 77]. A 45 µm diameter wire electrode curled into a 1.2 mm diameter circle is used for field enhancement. This is placed between plaques of epoxy (5 mm thick) and silicone rubber (3 mm thick). The methodology for track growth is described in detail in [9] and is unchanged for this project. The sample configuration is illustrated in Figure 2-24, adapted from [9].

A sample in which tracks have already been formed is used. The wire electrode is removed, as the aim is to perform testing without field enhancement points. This was replaced with a steel planar electrode placed at the base of the plaques. This configuration is illustrated in Figure 2-25.

Figure 2-24: Schematic of interfacial tracking experiment. Adapted from [9].

Figure 2-25: Schematic of Planar Electrode Tracking testing configuration

In order to increase the field size within the tracks and trees, the plaques were cut down in length from their original 150 mm. First to 80 mm, then further to 50 mm then finally down to 34 mm (with a 30 mm air gap to the ground electrode). The size was eventually limited by the sample design and the fields which they were able to withstand. The ends were polished each time to give a planar surface free from rough areas or contaminants which could unintentionally enhance the field.

The samples were monitored by partial discharge detection (again using an OMICRON MPD 600) and through optical imaging. The optical imaging was able to be used in a well-lit environment to see if track/tree growth had occurred. In a darkened environment it could be used with a long exposure to detect light emissions due to partial discharge emissions. The samples had up to 50 kVRMS (at 50 Hz) applied.

2.2.4. Interfacial Tracking Test Results The testing ultimately was unable to produce any evidence of further growth in either the electrical trees or the interfacial tracks. Discharges were not optically observed within these areas and no partial discharges which could be tied to treeing or tracking growth were measured.

Long exposure images such as those in Figure 2-26 were taken (this is from a 30 second exposure at

45 kVRMS), showing the location of any detected discharges. In this image the discharges only occurred in the silicone glue surrounding the sample, demonstrating the need for this area to be strengthened but not indicative of discharges within the sample. Improving this removed these discharges in further testing.

20 mm

Figure 2-26: Optical image with 30 second exposure of discharges showing as a purplish light emission during tracking tests.

After 6 hours of ageing at up to 50 kVRMS without sign of discharges within the sample the samples were reduced in size to increase the fields present. The result of this was when reduced to 34 mm in thickness the interface was unable to withstand the fields and breakdown occurred through the interface. The damage of this is shown in Figure 2-27, with a blackened material across the area of the sample.

Ground End of Sample Darkened sections of sample show carbonisation during flashovers Treeing and tracking

25 mm

Wire Electrode was originally here

High Voltage End of Sample

Figure 2-27: Breakdown through the interface causes surface level carbonisation.

The blackened material was shown to be at the interface and did not extend into the bulk of the insulations and hence could be cleaned off when the materials were separated. However continued testing showed the experimental design was insufficient to withstand the high fields with flashovers occurring repeatedly. At no point were discharges found, either through the MPD 600 or optically, to have occurred within the tracks or trees. Examination of the samples also showed no sign of channel growth or change.

As further development of the experimental methodology was necessary in order to apply larger fields it was decided at that point to first pursue alternative sample designs. This seems to demonstrate that even in mature tracks, with significant levels of carbonisation, in order to achieve the discharges necessary for tree growth a field enhancement point such as a metal inclusion is necessary.

2.3. Water Absorption Testing

2.3.1. Introduction Having attempted to induce electrical trees using reverse trees, alternative mechanisms for tree growth were considered. Once such method was based upon the conditions experienced during water treeing and involves generating high fields within localised areas of epoxy. The idea is that by controlling the permittivity of the layers through water absorption the field is greatly enhanced in small areas of epoxy leading to localised breakdowns. This section will briefly discuss the background of water treeing and how these are understood to lead to electrical tree initiation. The methodology and the results of the tests will then be covered.

2.3.2. Background Water treeing is a phenomenon what has been studied in detail for decades. Since the early 1970s [78, 79], there have been countless papers written on the subject and this is not intended to become a detailed literature review into how and why they form. It will however provide a brief background on them as well as a review of how it is they are understood to initiate electrical trees.

Water trees form within dielectrics under high voltages upon the sustained exposure and absorption of water. Distinct from electrical treeing, water trees are commonly observed to form in insulation lacking obvious sharp, conductive points [80] as is generally necessitated in electrical tree growth. Though it may well be the slight imperfections are able to trigger growth. Unlike electrical trees which form in very distinct and well-defined channels, water trees form diffuse and ill-defined channels [5]. The study of them using chemical [81], optical [82] and electron microscopy methods [83] has found they are comprised of many individual water-filled micro-cavities which may be interconnected.

The micro-cavities aggregate to form water filled ‘channels’ and gradually grow in the direction of the electric field. These channels can cover wide areas in the dielectric. Being inherently diffuse they effectively form a water dense region within the insulation. Such a region will experience, as a bulk, changes in conductivity and permittivity. The result of this being that changes in the electric field are experienced both within that region and in surrounding regions.

When water trees bridge the insulation between conductors they do not necessarily induce breakdown themselves. They do however allow the initiation and formation of electrical trees [7, 84]. This has been a topic of significant study and debate within the field and has been observed in both cables and in laboratory-based testing. An example of this is shown in Figure 2-28.

Figure 2-28: Electrical trees forming in response to water trees forming within the insulation (stained with methylene blue). Taken from a) [8] and b) [85].

Similar to reverse treeing, water trees can form from planar electrodes. It is very distinct from the sharp points typically used in electrical treeing tests. It is not fully agreed how water treeing leads to electrical tree growth however a common theory is the water trees distort the electrical field within the insulation. This field distortion creates a sufficiently high and localised field for tree initiation to occur. Other factors which may be involved are the weakening of the dielectric through the chemical degradation caused by the water [7] as well as the water within the trees evaporating at high temperatures, creating both higher pressures and voids within the channels in which discharges may occur [84].

Studies performed upon the effect of water trees in polymers have found they increase the permittivity of the material in which they grow, these tests are discussed by Radu [85]. It is known in a capacitor, two dielectrics with different permittivities will experience different fields according to

E1 =Q/(ϵ1A) and E2=Q/(ϵ2A). The larger permittivity experiences a smaller field and vice versa. The effects have been modelled in significantly greater detail [85, 86]. Others have considered the effects of space charge in the water tree and the impact on the wider electric field [87, 88]. The model by Acedo is notable here for one model in which the water tree is modelled as multiple layers in a planar geometry in Figure 2-29.

Figure 2-29: Model for calculating electric fields within insulations in which water trees have formed. Taken from [86]

The model proposed by Acedo follows the same configuration as that used in the tests discussed in this project.

Considering Figure 2-28b) it appears unlikely electrical tree initiation occurs due to the bulk effects of the water trees alone, assuming the trees do not in fact extend further than observed in this image. Space charges could extend yet further into the bulk as well, forcing the effective field enhancement into a smaller space. Another aspect could be the mechanical strain of the water tree channels as they are heated, in this case the sample experienced a sharp increase in field of 35 kV/min, this would mean a rapid heating and boiling of the water. This would cause an increase in pressure in the sample directly above the point of initiation. A combination would likely be occurring of these factors and may be able to explain this tree initiation. The tree in Figure 2-28a) appears easier to explain and could be a consequence of the above factors as well as the water trees creating discharge paths in which PD may occur.

2.3.3. Methodology

2.3.3.1. Adjusting Water Absorbency of Epoxy The idea behind this experiment, recreating the suspected water treeing effect, was to have different sections of the epoxy sample absorbing different amounts of water. This will increase the field in controlled regions, allowing localised breakdown and then treeing to occur. It is very similar to the suspected mechanism driving increases in failures following extended outages and heavy rain discussed in the background.

Epoxy is a polymer which will absorb water and when this occurs it can be expected to produce bulk effects upon the permittivity and conductivity of the sample. If present in high enough quantities these effects can be significant.

The water absorbance can be controlled here through altering the epoxy-hardener ratio of the sample. This is known to change the water absorption properties of the epoxy, in general as the actual ratio differs from the stoichiometric ratio (the ratio at which epoxy and hardener most completely react) the amount of water absorbed at saturation increases [89, 90]. In particular this is found with increases in hardener ratio [89]. The recommended ratio from the manufacturer for the LY5052 and HY5052 resin and hardener (100:38) is taken to be equivalent to the stoichiometric ratio here.

To test the effect of different resin-hardener ratios, and to understand the levels of differentiation in water absorption which could be achieved by this method, a number of calibration tests were performed. Samples made using a range of different epoxy-hardener ratios were held in a heated water bath at 90oC for 2 days. This temperature was chosen deliberately as being below the 100oC post-curing temperature so intended not to cause any chemical change in the molecules, while also accounting for time concerns as hotter water is absorbed far more quickly. The samples were weighed using a mass balance (Precisa XB 120A) before, during and after being held in the water bath, the surface of each sample carefully dried before each weighing. Three samples for each epoxy-hardener ratio were used and the percentage weight gain recorded and averaged, the results of this are summarised by the table in Figure 2-30.

Figure 2-30: Graph of water absorption for epoxy samples against Hardener - Epoxy Resin ratio following 2 days of submersion in heated water

By the end of the two days the absorption in each sample had significantly slowed down, with only very slight changes in weight occurring. The samples appeared to be approaching saturation point. From this graph we are able to see how different epoxy-hardener ratios will affect the level of water absorption.

When the stoichiometric ratio is held the water absorption is found to be at its lowest of 2.1% (100:38 in this case). This finding makes sense as in this case we would anticipate the best and most complete reactions to have occurred, limiting free volume area for the water to absorb into. Halving the hardener mass in each sample increased the water absorption levels slightly to 2.4% as can be seen in Figure 2-30. Increasing the hardener levels had a far more pronounced impact. At a 100:76 ratio (double the stoichiometric ratio) the water absorption had more than doubled to 5.5%.

There are concerns on the effect different resin-hardener ratios could have upon the epoxy dielectric performance. Without in-depth study, it can be expected that incomplete bonding of the constituent molecules will lead to non-optimised performance. While this is acknowledged as a potential issue, no efforts at this stage are taken to mitigate them; instead feasibility studies are performed first to observe the outcome from this test. Another concern which was considered was the effect of water release during the tests. An equilibrium is reached between the epoxy and the outside environment, absorbing more water in humid conditions and less so in arid conditions. The water absorbed in the water bath will be released over time outside of this, including during testing. This could be prohibitive for longer term tests, depending upon the rate of water loss, and for experiments performed in silicone oil the contamination of this through water loss must be noted and treated if necessary. With these noted only shorter term tests are performed and the sample design for this is discussed below.

2.3.3.2. Sample Design The chosen sample design is given in Figure 2-31, here two layers of epoxy are to be used. The first layer (“A”), is an epoxy made using the stoichiometric ratio of epoxy and hardener (100:38), this can be seen to have a thin layer at the centre of the sample, the thickness of which can be controlled in sample preparation. A number of different thicknesses were chosen in sample preparation, 200 µm, 300 µm and 500 µm with the distances controlled by using stoppers of set sizes. Here smaller gaps will create larger fields. However this may also cause difficulties in observing degradation and breakdowns. The second layer (“B”) is the same epoxy made using a 100:76 ratio of epoxy and hardener. When saturated in water, such a sample will see a conductivity and permittivity in the second layer which is significantly higher than in the first layer, the voltage drop across the first layer will then be significantly increased by this.

B A 1.5 – 3 mm

0.2 – 0.5 mm

Figure 2-31: Schematic of water absorbency test. A represents the base layer with the lowest water absorption and so the highest fields. B represents the layer with the highest water absorption, experiencing the lowest fields. A breakdown channel and tree growth is illustrated at the point at which layer A is thinnest, this is the intended growth mechanism.

As the schematic in Figure 2-31 illustrates the increased voltage drop, and so higher localised field, is intended to produce a small breakdown channel across layer A, and from this to observe tree growth into layer B. It is also anticipated this could lead to degradation being observed in the interface between layers A and B.

2.3.3.3. Sample Preparation The first layer of epoxy is produced using the epoxy mixing and vacuuming processes as described in the methodology for reverse treeing samples in section 2.1.3.1. In this case, before being left to cure, the epoxy-hardener solution is poured into acrylic moulds and then a rounded mould is pressed into the top to produce the shape of section “A” in the schematic shown in Figure 2-31, the apparatus for this is shown in Figure 2-32, including three different moulds trialled using this system. The layer thickness is controlled by the use of stoppers of known thicknesses.

Figure 2-32: Three trialled moulds to create the depression within layer “A”.

After the first layer had cured for 24 hours, the rounded mould was removed and a second layer of epoxy solution was poured into the void created (“B”). This epoxy was made using the 100:76 ratio and otherwise followed the same mixing and degassing procedures. This was then left to cure for 24 hours (as standard) and following this the entire sample was then post-cured at the standard 100oC for 4 hours.

Concerns were considered around the quality of the interface and the impacts which curing and in particular post-curing could have upon these. The interface this was optically inspected, looking for void creation or cracking occurring due to any difference in thermal-mechanical properties between the two epoxy mixtures. There were no issues identified optically in any of the samples.

Such a sample is given in Figure 2-33, the interface cannot be clearly seen in the photograph but is indicated by the red arrow in a) and overlaid with a solid line in b).

Following post-curing the samples were then held within a water bath for a week to become water saturated, maximising the water concentration differences and the conductivity differences between the layers of epoxy. The samples were then ready for testing.

a) b)

2 cm

Figure 2-33: A finished sample. In a) Just visible (pointed at by the arrow) is the interface between the layers of epoxy. In b) this interface is highlighted by a black line.

2.3.3.4. High Voltage Testing The testing procedure for the water saturated samples largely follows that of the reverse treeing samples, which is discussed in detail in Section 2.1.3. Partial Discharges were tracked. However, as with previous cases, separating any discharges from the noise is an issue. A note is made in this section of the difficulties in obtaining clear optical images in these samples due to the thin width of the bottom (“A”) layer.

Figure 2-34 illustrates the difficulties in making clear optical observations on the degradation occurring in this region. Samples are imaged as shown in Figure 2-34b, due to the shadows created in Figure 2-34a making optical observation impossible. Nonetheless difficulties remain, in particular the size of the lower layer and the wide area it covers creates issues. Imaging was chosen to be done by focusing at the area at which the lower layer is at its thinnest.

a) b) Electrode Electrode

“A” Layer 1.5 mm

“B” Layer

Electrode

Figure 2-34: Optical imaging shows the difficulty in imaging samples. Within a) a shadow covers the areas of interest where the base layer is smallest. In b) the interface is visible however due to its small size difficult to see within. The dashed lines show the interface between the epoxy and the electrodes.

From these images it can also be seen the interface between these two epoxy resins is not smooth. This is anticipated to increase the chance of a breakdown occurring due to creating field enhancements within these regions. This must be considered in two ways. As the field is increased it can be expected this will make degradation more likely. This could be seen as compromising the test which would ideally involve no field alteration from the interface geometry and be perfect, uniform planar samples. However, in this case it is viewed that this is an initial test of a new sample configuration where finding the exact parameters necessary to induce breakdown or electrical treeing may be difficult. Having small field enhancements which increase the probability will be helpful in understanding the feasibility of initial tests. This is considered acceptable on the understanding that the impact of such roughness or defects is relatively small within the scope of the sample configuration.

2.3.4. Water Absorption Result

Following the application of voltage (25 kVpk, 50 Hz) it was observed a breakdown channel appeared in the thin layer, imaged in Figure 2-35. The gap here was 200 µm.

It is believed the dark channel, shown more clearly on the right-hand picture of Figure 2-35 was caused by the high fields generated, however the difficulty in observing it optically, and with noise in the PD measurements this has been difficult to investigate. When the channel appeared and whether it could have been a defect/contaminant present from before voltage application during sample preparation is not clear. This demonstrates the difficulty in finding the balance between a smaller gap, with increased fields, and larger gaps, with improved visibility. Attempts to repeat this with larger gaps (300 µm and 500 µm) to aid with optical imaging were not successful.

1.5 mm

0.25 mm

Figure 2-35: Optical Images of the water saturated samples, planar electrodes above and below (seen on left). Potential breakdown channel (magnified in right hand image)

a) b)

200 µm 200 µm

Following the appearance of the channel it was dried in an oven and a desiccator before the voltage was reapplied. This was intended to see if tree growth could be triggered from the channel however after a voltage of 20 kV was applied for 1.5 hours and 25 kVpk was applied for 5 hours there was no observed breakdown, or signs of tree growth or channel widening after this.

The channel was later imaged under optical microscope as shown in Figure 2-36 however this was unable to clearly discern any information about the breakdown.

To calculate the fields present in the epoxy a similar approach is made to that made in the fine treed section. The calculation will focus upon the closest point between the layers and assume a perfectly flat region, without the roughness which will be assumed to be negligible for the purpose of these calculations.

The same equation will be used as previously:

휀푉 푉 퐸 = = 휀 푑 + 휀 푑 휀 푑 + 푑 휀

In this case however the permittivities, ε1 and ε2, will be affected by their respective water absorptions. It is assumed at the point of testing they had not released a significant amount of water and were effectively saturated. Layer 1, at a ratio of 76:100, will be at approximately 5% by weight water. Layer 2, at a ratio of 38:100, will be approximately 2% by weight water.

To estimate the effect of water absorption upon the permittivity of the epoxy [91] is used which tracked permittivity changes at different water absorption levels for a neat epoxy. Their epoxy had a relative permittivity εr = 4.5 with zero water absorption. At 2% by weight water absorption, the

77 relative permittivity was 5, an increase of 11%. At 5% by weight water absorption, the relative permittivity was 7.5, an increase of 67%.

The applied voltage is taken to have a maximum of 25 kV, d2 is 200 µm and d1 is 1.3 mm; the conditions under which the breakdown channel formed. The field at this time can be calculated as being:

푉 25000 푀푉 퐸 = = = 11.6 = 11.6 푘푉/푚푚 휀 7.5 푚 푑 + 푑 0.0013 + 0.0002 휀 5

Notably the 11.6 kV/mm calculated field is below the typical breakdown strength of epoxy (25-45 kV/mm). This discrepancy could be explained in a number of ways. The differences in permittivity could be larger than anticipated, the surface roughness observed in Figure 2-35 could create minor field enhancements (notably the breakdown did not occur at the closest point which would give credence to this), or the interface between the layers could also create a space for water to have gathered and further influence the electric field.

A breakdown channel has been observed to form from water saturated samples, with different absorbencies producing significant differences between different regions in the epoxy. This is a localised breakdown created using plane-plane electrodes however it has not yet been shown to be reproducible, nor was electrical tree growth found to occur following this breakdown. This result may provide insight into the mechanism of tree initiation in cables however this remains to be established.

2.4. Conclusion

This chapter discussed experiments performed utilising samples subject to fields generated using planar electrodes in attempts to replicate observed phenomena of tree growth and initiation which do not require sharp points in experiments and in cable insulation. This replicated fine treeing (Section 2.1), interfacial tracking (Section 2.2) and water absorbance (Section 2.3) with only planar electrodes used for excitation. In these tests only one channel was found to have developed, in the water saturated sample. The fine treeing and interfacial tracking tests were unsuccessful. Unfortunately the saturated sample result could not be repeated, nor could further tree growth or initiation be found to occur in this case.

It is believed a channel was able to form in the water saturated sample (Section 2.3.4) as the product of two factors. Firstly the bulk permittivity changes created by such significant differences in water saturation level are likely to have created greater levels of field enhancement than those seen in the

78 fine treeing case. It is also probable the surface roughness on the interface between the samples also created some level of field enhancement and that this aided the channel’s formation.

The attempts to trigger tree initiation or growth through planar samples uncovered a number of issues which affect these experiments. Difficulties were found in imaging samples effectively and in identifying whether a sample was indeed ageing or not. While this may be a minor issue on tests which last for no more than a day, when performing tests over months or years if trying to replicate the conditions within cables it becomes a significant problem. The findings of this work are discussed in greater detail in the Discussion in Section 5.1.

Improved sample analysis techniques are necessary for tests such as these which may occur over significantly longer periods of time, in particular in providing the ability to discern ageing prior to electrical tree initiation. The next chapter provides details on such a technique which allows detailed chemical analysis to be performed on samples, reaching spatial of around 50 nm.

3. Atomic Force Microscopy with Infrared Spectroscopy (AFM-IR) - Technique

3.1. Introduction

This chapter covers the application of atomic force microscopy with infrared spectroscopy (AFM-IR) to electrical treeing for the first time. This is a state-of-the-art technique which can provide chemical information at resolutions of around 50 nm, well below the size of a tree channel. The necessity of such a technique has been shown in the previous chapter however it also has great applicability in the study of electrical treeing more generally which is provided here. It is hoped the understanding of electrical treeing developed here can later be applied to the development of planar sample testing techniques.

This chapter provides a brief background in the chemical study of electrical trees. Primarily this looks at previously applied chemical analysis techniques and discusses what they found but also what their limitations are. A description is then given of the AFM-IR technique, explaining what it is and how it works. This is followed by the methodology which was used, primarily focussing upon the sample preparation methods applied such that samples were suitable for testing with the AFM-IR.

3.1.1. Background information on chemistry in electrical treeing

3.1.1.1. Degraded region formation The incubation period, including the stochastic nature of tree initiation was previously discussed in the literature review. During this time period, when voltages applied are sufficient (which can be determined by monitoring electroluminescence [36]), then a degraded region is formed around the needle tip. This region is an area of chemically damaged material, although the mechanisms by which this form remain disputed.

The link between electroluminescence, chemical degradation and tree initiation are well established. This has been done through threshold testing [36], free radical monitoring [12], low temperature and oxygen deprivation testing [13, 24, 92, 93] and comparisons between the size and locations of electroluminescent regions and degraded regions [35].

The most commonly proposed mechanisms are the hot electron theory and photo degradation through charge recombination.

In the hot electron theory, charges are injected from the needle tip into the interface between it and the polymer. These electrons are then accelerated by the applied electrical field, gaining energy until

80 they eventually collide with a molecule within the polymer, transferring the energy to it. If the energy gained by the electron is sufficient then a bond will break within the polymer, creating free radicals and causing chemical reactions to occur within the material. These reactions weaken the material over time. When the degradation products relax (release energy) they can do so through a radiative emission [33, 42], in which photons are emitted which we observe as electroluminescence.

Charge recombination is also proposed to occur due to charge injection. Suggesting that following charge injection, charges become trapped within trapping sites in the polymer. They can remain stuck there for long periods of time until it recombines with a charge of opposite polarity. This charge recombination gives off a large amount of energy which can cause damage to the immediate surroundings similar to the hot electron theory. Charge recombination will also produce a light emission, which we detect as electroluminescence [33, 42].

There has been significant testing and analysis performed to identify the cause of the chemical degradation. Shimizu et al [33] performed energy calculations indicating the electroluminescence could not be the dominant cause of degradation. This utilised anticipated chemical damage and the associated bond energies, as well as anticipated number and energy of photons based upon experimental evidence. Bamji et al [32] suggest hot electrons cannot explain electroluminescence above the threshold voltages and therefore cannot be responsible for tree initiation. While Zhang et al [30] analysed electroluminescence patterns to determine that both hot electron interactions and charge recombination occur and are involved in tree initiation. It is important to understand the mechanisms which are involved in this period. However, this project does not focus in detail upon the cause of this chemical degradation, instead focussing on characterising the degradation which has occurred.

3.1.1.2. Degraded region chemical characterisation The degraded region is a very chemically active region during the pre-initiation stages of electrical ageing. Tests have shown carbonyl (C=O) groups and carbon-carbon double bonds (C=C) form under extended ageing within polyethylene. These tests have often used oxygen deprivation techniques to allow extended ageing without inducing electrical treeing. Techniques such as cooling to 77 K [92] effectively reduce oxygen mobility to zero and performing tests under vacuum or under nitrogen [13] increase the time taken for initiation to occur.

Methylene blue dyeing [35], FTIR [24] and Raman spectroscopy [24] have each been used in such tests and repeatedly confirmed the presence of these groups in the initiation process. The chemical products identified, as well as the dependence on oxygen has meant a chain scission and

81 autoxidation process is commonly understood to occur in forming the degraded region. This begins with a hydrogen abstraction (in which a hydrogen atom is lost from a carbon it was attached to), creating a free radical R* which will be extremely reactive with surrounding molecules. In an oxygenated (O2 rich) environment this can then form a peroxide (ROO*) and then a ROOH group by attacking other chains, which creates more free radicals (R*). This mechanism continues to damage chains producing C=O and C=C groups, known as chain scission, while the excess free radicals it produces continue this process (autoxidation). This process was described by Shimizu and Laurent [34] and is drawn in Figure 3-1.

The limitation of the prior work has been the resolutions which the techniques used have been able to provide. FTIR has at best a resolution of around 15 µm (which exceeds the radius of the degraded region) and Raman, which although said to have a resolution of around 2µm, has been shown to not necessarily perform to this level [53]. While methylene blue can have better resolutions (up to the optical diffraction limit) it is unable to deliver a true chemical characterisation. Other techniques which have been used in chemical studies of treeing (Electron spin resonance (ESR) and X-ray photoelectron spectroscope (XPS) [12]) similarly have limitations in the types of information collected which mean they are unable to provide a complete chemical analysis.

Figure 3-1: Autoxidation and chain scission reaction, taken from [34]

A result of the limited spatial resolution of the techniques previously used is that assumptions have been made about the degraded region being chemically homogeneous [32] which are not backed by evidence. There is no understanding of how the chemistry may change at different locations around the needle tip and whether degradation occurs across the entire region or is concentrated in particular locations. When developing models, particularly those which calculate energies based upon types of chemical damage occurring across a region, such information is vital.

3.1.1.3. Initiation Mechanism There are a number of mechanisms which have been proposed to be responsible for tree initiation. Distinct from the incubation period this focuses upon the moment of initiation, how it is we go from a solid block of polymer to a polymer with a gaseous channel within it.

Arguably the most commonly proposed mechanism [11, 34, 38, 26] is one in which an electron avalanche occurs within the polymer. These happen due to injected charges from the needle tip and a high electric field which accelerates the charge to give sufficient energy to breakdown the air in the interface and then damage the polymer. This is commonly believed to follow successive chain scissions and autoxidation reactions within the polymer which have produced small voids close to the needle tip (Figure 3-2).

Small voids have been evidenced in the degraded region prior to initiation through SEM imaging [11, 94] while TEM imaging in the same work showed damage to the spherulite structure of the polyethylene.

Figure 3-2: Damage to polymer bonds due to hot electron collisions in oxygen rich environment leading to tree initiation [11]

Another mechanism which has been commonly proposed is electromechanical fracturing [16, 61-65, 70]. This theory suggests the electromechanical forces on the polymer created by the changing fields are responsible for tree initiation. There is significant mechanical stress stored within the insulation and when this exceeds the withstand capabilities of the material it leads to cracks forming through the material. These cracks may continue to propagate through the dielectric with the fields at the tips sufficient to drive further growth.

Such theories can make several predications which are well held up against experimental data, and such cases are well described by Crine [62]. However the theory has until now lacked any direct evidence which indicates mechanical fracturing does occur during tree growth or initiation. On the other hand electron avalanches (partial discharges) are known to occur; evidenced through optical, chemical, electrical and acoustic means.

3.1.1.4. Tree growth chemical study The study of chemical degradation caused by electrical tree growth is rather limited. The resolution limitations of the tools which have been available means that it has been difficult to discern more than broad information on what is happening around the trees.

Vaughan et al [53, 95] used Raman spectroscopy in the study of tree growth within conducting and non-conducting polyethylene trees. These tests identified carbon deposits within the channels of a tree grown at 30oC, rendering it effectively conducting. This carbonisation was sufficient to prevent discharges from occurring within the body of the channels, limited to the channel tips (as shown in an example image in Figure 3-3).

a) b)

2mm

Figure 3-3: Conducting Polyethylene tree in a) the light emission from PD is visible at tree tips. In b) the same tree is visible. Circled in each image is the location of the discharges. (Image courtesy of Dr Hualong Zheng).

a) a)

2mm

Figure 3-4: Non-conducting polyethylene tree, in a) light emission occurs along the length of the channels. In b) the tree is visible (Image courtesy of Dr Hualong Zheng).

Within the non-conducting tree (which was grown at 20oC), there were no corresponding carbon deposits to increase the conductivity of the channels. The result of which is that discharges instead occur along the full length of the channel (as shown in example image in Figure 3-4).

Their work however found difficulties achieving the desired high resolutions suggested by Raman spectroscopy and while they could not provide an accurate measure of their realised resolution it was above 2 µm.

Tests by Hu et al [60] utilised micro-infrared spectroscopy to study electrical trees grown in polyethylene. These tests identified C=O groups as in tree initiation studies, but also hydroxyls (C- OH) and ether (C-O-C) groups, it was also found that the level to which they were found changed in different locations around the tree. This was believed to be due to differing oxygen levels in around the sample.

The C=O groups were believed to have occurred via the same chain scission process as discussed previously, while the C-OH and C-O-C were suspected to have developed as a later stage of degradation. A suggested process for this was a Norrish Type 2 reaction (Figure 3-5).

Figure 3-5: Norrish Type 2 reaction

The limited resolution of the technique (~15 µm) again prevented detailed analysis of individual trees however. With this limitation the study and analysis were limited to changes around the tree as a whole.

3.1.1.5. Comparison of polyethylene and epoxy The bulk of chemical testing performed on treeing has been done upon polyethylene. While both polyethylene and epoxy are strong dielectrics they are very different chemically. Polyethylene is chemically a very simple material composed of an extended carbon chain with attached hydrogens (Figure 3-6).

Epoxy is a more complex material, defined by containing the epoxide group (Figure 3-7 and visible in Figure 2-2) of two carbon atoms connected to a single oxygen atom. This group is typically involved during the curing process when it reacts to form a crosslinked network within the epoxy.

The exact structure of the cured epoxy will be dependent upon the resin and hardener used to produce it. There are many forms of epoxy in used each with different constituents and different properties. The type of epoxy used in this project is the LY 5052 from Huntsman and the constituent molecules are shown in Figure 2-2. Here we can see there are both aromatic and nitrogen groups present. The more complex chemistry present in the epoxy, compared to polyethylene, means the spectra will be more involved in epoxy samples and the additional groups present creates the possibility of reactions occurring which are not typically found in polyethylene.

3.1.1.6. Typical spectra A typical spectrum is given in Figure 3-8. It is also annotated at peaks and regions which are commonly found in this project.

Figure 3-6: Polyethylene molecular chain

Figure 3-7: Epoxide group

N-H and C=C bonds Carbonyl Region Fingerprint Region

Aromatic Peak

Figure 3-8: Example spectra (from Sample AFM(1)-A2(1)), annotated with key spectral regions

 The carbonyl (C=O) region is typically measured from around 1650 cm-1 to around 1800 cm-1. All carbonyls absorb in this region, while few other groups do making it a very useful measure for carbonyl presence in a material.  At around 1620 cm-1 is a peak which is regularly found identified, this could correspond to NH and C=C groups, both of which could be expected in aged epoxy samples making it difficult to distinguish between the two.  The 1500 cm-1 aromatic peak is a strong peak which is often used to normalise measurements. The aromatics are relatively stable and unlikely to change during ageing making them ideal for this. There is also very little else which absorbs at this wavelength.  The fingerprint region is the name given to the IR spectrum from around 1450 cm-1 to 400 cm-1. In this region many bonds and groups are active and will absorb. This makes discerning particular bonds here more difficult and less certain. The bonds active in this region include C-O-C, C-O, O- H, C-N. Attempts are made to distinguish these groups in this report however these cannot be certain.

A table of relevant assignments is given in Table 3-1, however it should be noted these assignments cannot be precise as the exact make-up of the chains they are part of will impact the absorption wavenumbers.

Wavenumber (cm-1) Functional Group 1760 C=O (Carboxylic Acid) 1726 C=O (Esters) 1708 C=O (Ketones) 1670 C=N (Imine) 1680 - 1610 C=C Bonds 1650 - 1580 N-H Bonds (Amine) 1500 Aromatics 1360 - 1300 C-N (Amine) 1248 C-O 1056 O-H Table 3-1: Relevant peak assignments [96]

3.1.2. Atomic Force Microscopy with Infrared Spectroscopy – How it works Atomic Force Microscopy with Infrared Spectroscopy (AFM-IR) is a technique developed in 1999 by Hammiche et al [97] which combines two common analytical techniques to produce the ability to study the chemical nature of a material with spatial resolutions on the scale of 50 nm. This spatial resolution is attained while maintaining a high spectral resolution, which here is shown to allow a new level of insight into the processes underpinning electrical treeing.

AFM-IR utilises Atomic Force Microscopy (AFM) which is a type of scanning probe microscopy which allows surface profiles to be measured with spatial resolutions as low as a nanometre. This goes well below the diffraction limit of optical microscopy and is on the same scale, in some cases exceeding that achieved with scanning electron microscopy (SEM). As illustrated in Figure 3-9, this technique works by use of an AFM probe which has a tip on the order of a few nanometres in radius, attached to a cantilever micrometres in thickness, which is in contact with the surface and pulled across it. A reference laser is incident upon the top of the cantilever and reflected towards a photodetector. Any probe movements due to the topography of the surface are then detected using this system as variations in the light incident upon the photodetector.

The AFM technique can be developed to perform many different functions; including measuring the mechanical properties of the material [98] and even performing atom manipulation [99], however the AFM application here is limited to the imaging of samples in this technique.

The AFM-IR combines AFM with infrared spectroscopy, making the AFM-IR analogous and comparable to other infrared spectrographic techniques such as Fourier transform infrared spectroscopy (FTIR). In AFM-IR a tuneable, pulsed infrared laser is incident upon the surface of the sample such that when the correct wavelength is used, it is absorbed by the constituent molecules and functional groups leading to a thermal expansion. This process is illustrated in Figure 3-10.

Figure 3-9: Schematic illustrating the functioning of the AFM part of the AFM-IR (Image courtesy of Dr Suzanne Morsch)

Figure 3-10: Schematic illustrating the operation of the AFM-IR

Figure 3-11: The deflection reading taking from monitoring the position of the AFM probe can be Fourier transformed to give the amplitude of deflection.

The thermal expansion of the sample is a sharp but short-lived response to the laser illumination, this expansion causes the probe in contact with the surface above to receive a ‘kick’, which causes the cantilever on which the probe is held to vibrate in response. The intensity of this kick corresponds to the size of the expansion, which will be dependent upon the level of photon absorption and the number of the pertinent functional groups present within the material. In essence a higher concentration of a particular chemical bond will lead to a larger expansion and so a larger kick.

Such a kick is shown in Figure 3-11, labelled as the deflection of the cantilever as measured using the photodetector. This deflection can then be Fourier transformed to determine the amplitude of the kick. The effect of this is that this amplitude measured using the AFM-IR is indirectly a measure of the infrared absorption at the chosen wavelength of light.

By changing the wavelength of the light a spectrum of absorption can be built up at a particular, very well defined (~50 nm resolution) location. An alternative use of the AFM-IR is to use it as an absorption mapping tool, by holding the wavelength of the light stable while moving the AFM probe across the surface a map of the absorption over the region can be developed. These tools are both used in this project and examples are given in Figure 3-12.

Figure 3-12 looks at the chemical degradation occurring around the needle tip following electrical ageing. The images on the left give the spectra of the degradation multiple locations around the needle tip as indicated in the map; closer regions represented by green, while red represents regions further from the needle tip. The image on the right is a map of the absorption at 1726 cm-1, which here has been assigned to Ester C=O bonds, indicating precisely where they have and have not formed. These results are given here for demonstrative purposes and will be discussed in detail in the results and discussion chapters.

a) b) -1 1726 cm

Ester C=O bonds

Figure 3-12: In a) an example spectra is shown above a surface profile map showing the locations the spectra were taken from, b) shows an absorbance map of the same region at 1726 cm-1 where red areas show higher absorbance. The map is 30 x 30 µm.

The AFM-IR should be however only be considered as a qualitative measure of the number of degradation particles within a system. Due to the nature of the technique variations in material structure will affect the size of the kick and expansion making quantitative assessments of the exact numbers of a particular functional group within a sample impossible using the AFM-IR technique. Similar disparities can be produced through subsurface materials inducing different readings at different depths. The infrared absorption measured using AFM-IR has however been shown to correlate with that observed using FTIR [100, 101].

The AFM-IR has gained extensive use as a chemical characterisation tool [100-102] including studies into epoxy [103] and electrical tracking more specifically [77]. Three papers have been written on the AFM-IR work detailed in this project. Two conference papers, one presented at ICD 2018 [104] and the other at EIC 2019 [105] and one journal which has been accepted awaiting publishing [106].

3.2. AFM-IR Chemical Analysis Testing Methodology

This section will discuss the methodology applied to sample testing using the AFM-IR chemical analysis technique and then discuss the electrical ageing of the samples. The polishing, grinding method used to prepare samples for the AFM-IR is then described followed by the specifics of how the AFM-IR is used to analyse each sample.

The point-plane configuration is used to grow electrical trees for the AFM-IR testing. The epoxy is made using the standard mixing/degassing/curing process as described in Section 2.1.3.1. The

91 epoxy-hardener ratio used is the manufacturer recommended stoichiometric ratio, producing the best quality of dielectric and samples which can be readily compared to those typically used in electrical treeing tests. Electrical ageing is carried out using the point-plane configuration and circuitry as described by Section 2.1.3.1. Taking around 10 days to prepare.

3.2.1. Samples Here the samples aged and tested in this experiment are discussed. These cover two separate stages (described here as the First Testing Set and the Second Testing Set); the first set covers a range of stages of tree formation and growth, the second set exclusively considers chemical ageing prior to tree initiation.

3.2.1.1. First Testing Set The first testing set consists of five batches of five samples each, 25 samples in total. Due to time restrictions only a small number of these were tested using the AFM-IR; the samples used for testing are noted in the results section. Each batch of samples is prepared using the same method, aiming to produce consistency across the set.

The samples prepared can be divided into four different groups: Control (unaged), Aged (pre- initiation), Initiation (voltage removed shortly after initiation), Growth (voltage removed as tree approaches planar electrode).

Each batch of five produces: 1 control, 1 initiation, 1 growth and 2 aged samples. With more ‘Aged’ samples produced due to the inherent stochasticity of electrical tree initiation increasing the probability of unintended tree initiation during the electrical ageing. Dividing each batch into different groups minimises the impacts of inherent variation between different samples.

Labelling Each sample is uniquely labelled to identify batch and grouping following the pattern given:

AFM Test Set Number – Group Letter Batch Number (Number)

- AFM: Identifies that the samples were produced for testing by AFM-IR - Test Set Number: The first number is used to identify which testing set the samples belong to. In this project this will be either a 1 or a 2 for the first or second testing sets. - Group Letter: The letter here identifies the type of ageing applied to the sample with C (for control), A (for aged), I (for initiation), G (for growth) being used in the first testing set. - Batch Number: The batch number allows the batch the sample was produced from to be identified.

- (Number): This number is specifically used for aged samples in the first testing set. As 2 aged samples are produced in each batch it allows them to be distinguished, and are each labelled with a (1) or a (2).

Examples of this written out are: AFM(1)-G1 & AFM(1)-A2(1).

Control The Control Samples are produced using the same method as every other sample, excepting that no electrical ageing is applied prior to AFM-IR testing. The intention for this is to ensure that any chemical degradation observed is as a product of the electrical ageing alone and does not occur due to any other mechanism.

Aged The Aged samples are intended to investigate degradation which occurs around the needle tip prior to tree initiation. The ideal test would study the chemical degradation immediately prior to initiation however there is no method by which the proximity to tree initiation can be accurately determined. Due to the stochasticity of electrical tree initiation this was identified as the most difficult of the groups to test consistently. Ageing must be long enough for sufficient chemical degradation to build up, but not so long as to cause tree initiation to occur.

In these samples the voltage is raised to 15 kVpk and were run for set times after which the voltage would be removed. While this voltage is arguably quite high, it was chosen as such to ensure chemically degradation would occur even at the cost of large numbers of samples initiating prematurely. Tests were stopped if a tree was suspected to have been initiated. To track for initiation samples were monitored using optical and partial discharge monitoring techniques. Trees were found to have initiated in 6 of the 8 samples suitable for testing (not discarded for defects or contamination) leaving 2 aged but uninitiated samples from this set.

Initiation

Initiation samples are samples in which a voltage of 12 kVpk is applied for 10 minutes and then slowly increased until tree initiation occurs. The voltage is removed immediately following evidence of tree initiation. These samples allow the study of the degradation occurring at the needle tip, as well as within and around the tree channel during the earliest stages of tree growth. To determine tree initiation optical imaging is used. Due to time and equipment constraints, partial discharge monitoring could not be used for these samples.

Growth Growth samples are samples in which more mature channels can be studied. The power of the AFM- IR with its high resolution analysis allows for precise localisation of the chemical degradation within, around and ahead of tree channels to identify where the degradation occurs, and what drives channel growth.

In these samples a voltage of 25 kVpk is applied to initiate the channels and for initial tree growth. As the channels grow towards the planar electrode this voltage is dropped to 12.5 kVpk, this is understood to encourage wider channels to form. This is done due to concerns regarding the width of epoxy channels, which are typically finer than those formed in treeing in polyethylene, and the ability to accurately cross-section them. The voltage is removed as the channels approach the planar electrode, before they are at risk of a breakdown channel forming. Such samples take longer to grow, around half a day per sample.

Voltage (RMS) Comments AFM(1)-G1 25 kVpk --> 12.5 kVpk Grown to approximately 1.6 mm AFM(1)-G2 25 kVpk --> 12.5 kVpk Grown to approximately 1.6 mm AFM(1)-G3 25 kVpk --> 12.5 kVpk Grown to approximately 1.6 mm AFM(1)-G4 Unaged AFM(1)-G5 Unaged AFM(1)-I1 12 kVpk --> 14.63 kVpk Tree appears branch-like and approximately 30 µm in length AFM(1)-I2 12 kVpk --> 14.07 kVpk Tree appears bushy though only ~15 µm extent from the needle tip AFM(1)-I3 12 kVpk --> 15.98 kVpk Tree bushy appearance. 60 µm in length. AFM(1)-I4 Unaged AFM(1)-I5 Unaged AFM(1)-A1(1) 13 kVpk Large PD immediately. Sample later retested – 100 pulses between 1 and 2.3 pC found in 10 minutes testing AFM(1)-A1(2) 15 kVpk PD after 7 minutes. Possible tree initiation AFM(1)-A2(1) 15 kVpk PD occurring. No optical evidence of treeing AFM(1)-A2(2) 15 kVpk Some PD after 1 minute. Tree later identified. AFM(1)-A3(1) 6.3 kVpk Large PD, defect identified on needle tip. AFM(1)-A3(2) N/A Contamination. Sample discarded AFM(1)-A4(1) 15 kVpk Tree found to be initiated. AFM(1)-A4(2) 15 kVpk Tree initiation after 1-2 minutes AFM(1)-A5(1) 15 kVpk Tree initiation after 3 minutes AFM(1)-A5(2) 15 kVpk No optical sign of treeing. Large PD. Table 3-2: First Testing Set – Sample ageing voltages and comments

Optically identifying tree initiation can be difficult when the channels produced are small. Samples A1(1), A1(2), A2(1), A4(1) and A5(2) were selected for further study under microscope. A1(2) was found to have only just undergone initiation, a bifurcation having just occurred – this sample was

94 also chosen to be analysed using the AFM-IR for the study of initiation and early stage tree growth. A4(1) and A1(1) were also identified to have initiated and are not further studied.

Samples A2(1) and A5(2) were found not to have initiated and so were selected as AFM-IR candidates.

3.2.1.2. Second Testing Set The second test set was focused upon pre-initiation degradation, to identify the chemical degradation occurring prior to tree formation and to observe how this changes with time under voltage. The intended result of this investigation was to identify the condition of the insulation necessary for tree initiation to occur.

The second test set was composed of 2 batches with 6 samples each, giving a total of 12 samples. These were manufactured using the sample preparation method as previously described in section 2.1.3.1. As this test set focused on the degradation preceding initiation each sample was intended to be aged for a set period of time, with the voltage removed prior to initiation. Of course the inherent stochasticity of tree initiation creates difficulties with this as it is not possible to predict when a tree will form. This is not a significant issue for samples with shorter ageing periods however those which are intended to be aged for longer times are at risk of initiating too early.

The ideal sample for testing would be one in which the voltage was removed immediately prior to initiation, giving maximum information on the chemical changes and chemical state necessary for initiation. The test was designed around maximising the probability of attaining such samples by pushing the ageing times while acknowledging this would reduce the total number of samples which did not initiate. In any case, the limited AFM-IR time prevented the testing of large numbers of samples, requiring that those tested maximise the information taken from them.

Labelling The labelling for the second AFM-IR testing set follows the same pattern as the first testing set except that as every sample produced was “Aged” the group letter was omitted for simplicity. They followed the pattern AFM(2)-#, where # represents the sample number (1-6 were from batch 1, 7-12 from batch 2).

Ageing The electrical ageing histories of the samples are provided in Table 3-3, showing the voltages and times they were aged for, as well as whether they initiated or not.

Voltage (RMS) Ageing Time Initiated (Yes/No) AFM(2)-1 9 kV 1 min Yes AFM(2)-2 9 kV 3 min Yes AFM(2)-3 8 kV 38 min Yes AFM(2)-4 8 kV 1 min Discarded AFM(2)-5 8 kV 20 min No AFM(2)-6 N/A N/A Discarded AFM(2)-7 8 kV 71 min Yes AFM(2)-8 8 kV 55 min Yes AFM(2)-9 8 kV 42 min Yes AFM(2)-10 8 kV 40 min No AFM(2)-11 8 kV 19 min Yes AFM(2)-12 N/A N/A Discarded Table 3-3: Second AFM-IR Testing Set – Sample ageing voltages and comments

As can be seen in samples AFM(2)-1 and AFM(2)-2, which were the first to be tested, these were initially ran at 9 kVRMS but initiation was found to be a near immediate process at this voltage making it unsuited for the suggested tests. Further tests were performed at 8 kVRMS.

Samples AFM(2)-5 and AFM(2)-10, aged for 20 min and 40 min respectively, did not initiate electrical trees and were used for further testing under AFM-IR, representing different stages of pre-initiation ageing. This alongside the aged, initiated and control samples from the first testing set provides for a range of ageing levels. Ideally more samples would be tested to build a more complete picture however limits on time using the use of the AFM-IR restricts the number of samples which can be studied using this technique. Each sample taking a full day on the machine to be fully characterised and this time being restricted by heavy demand and technical issues which meant it was out of order for most of 2019.

3.2.2. Preparation for AFM-IR This section discusses the preparation of samples for the AFM-IR, including the polishing and grinding technique used in combination with optical imaging to ensure the areas chosen for study were exposed while maintaining a high surface quality.

Work using the AFM-IR to study chemical degradation during surface tracking was previously performed in Manchester. The preparation process for the study of electrical treeing differs greatly from surface tracking however and is significantly more involved because:

 Surface tracking tests are inherently surface effects, occurring at the interface between epoxy and silicone rubber. They can be exposed simply by separating those layers, and studied effectively assuming the layers were well produced and not overly rough.

 Electrical treeing is a process which occurs within the bulk of the material; meaning the needle tip or channel, micrometres in thickness, must be brought to the surface. While doing so, the sample surface must be kept extremely smooth for the AFM-IR to work effectively as surface roughness compromises the testing. This must be done without overly impacting the channels, and inducing any chemical change not due to the electrical ageing.

There is no defined limit on surface roughness or how close to the surface the channels must be in order to be studied. The AFM-IR requires effective contact with the surface to produce consistent results, the effectiveness of the contact will be reduced by changes in the surface topography. It can be treated as a signal-to-noise issue, the rougher the surface the greater the noise. The ability of the AFM-IR to study chemistry below the surface is material dependent but can even vary within the same sample, dependent upon properties such as the crystallinity of the material well as the strength of the thermal-expansion response for different chemical bonds. Therefore, no clearly defined limit can be made regarding the depth at which chemical products will be detected. However experience in these tests suggests channels must be at the surface or within a number of micrometres of it, to be studied.

Identifying Areas of Interest and Cutting down Sample The first stages in preparing samples for the AFM-IR are to identify the areas which are to be studied and then to cut the samples down to size. This is first done using optical imaging which is a non- invasive and non-destructive method, but still allows high resolution observations on the channels. X-Ray Computed Tomography (XCT) [107] was considered at first; this has been used to develop 3D models of electrical trees (whereas optical imaging is limited to 2D projections) and could identify where degradation has occurred, in comparison to where our models predict it to occur. Unfortunately XCT utilises powerful, low-wavelength, photons (X-rays) in order to study the channels these were identified as discolouring the samples when used, which is assumed to be oxidation or similar chemical change occurring. Using this technique would compromise the study and affect the results.

The samples must be cut down to size for the AFM-IR, to approximately 1 cm3. This can be done using a hand saw however care must be taken to minimise cracking which can form in the glassy state epoxy. The needle electrode should also be removed at this stage.

Polishing and Grinding Process To bring the channels or degraded region being studied to the surface a polishing and grinding technique is used as it allows a high degree of control and accuracy to be realised in approaching the

97 channels. It is combined with optical imaging which allows the position of the channels relative to the surface to be tracked consistently.

Silicon Carbide grinding papers are used as the first stages of the grinding-polishing process. Larger, coarser grit sizes are used at first when the channels are further from the surface, moving to finer grit sizes as they are approached. The finest silicon carbide grinding paper used is European P-Grade P4000 with a typical particle diameter of 2.5 µm. A grinding and polishing machine can be used with this in these stages, however should be done with care as over grinding is a risk and the final stages are done by hand to ensure the areas of interest are not removed in the process.

The final approaches are performed using diamond compound polishing paste, consisting of suspended polycrystalline diamond particles. Three sizes of diamond paste are used in this project; 3 µm, 1 µm and 0.25 µm. As with the grinding papers these are used in decreasing sizes and were entirely polished by hand due to the risk of over polishing and removing the channels. Particular care is taken at this stage to minimise the creation of large scratches in the samples during polishing which impact the AFM-IR measurements, these can be made if too much force is applied during the process or contaminants are allowed into the system. This is a particularly time intensive process given the precision and care required in preparing each sample. This took at least 1 day per sample to achieve a result suitable for the AFM-IR. The most precise requiring a week’s worth of polishing to reach the desired location. It is noted however that greater experience in the process has decreased the time required.

The polishing and grinding process is tracked throughout by optical imaging techniques; a microscope is used to identify the position of the channels below the surface and how much further the sample needs to be polished. Both lighting from below the sample (“bottom-lit” - providing a projection of the trees) and lighting from above the sample (“top-lit” - providing a clear image of the surface) are used here to provide a complete picture of the polishing process; these are illustrated in Figure 3-13.

a) b)

25 µm 25 µm

Figure 3-13: Optical images of an electrical tree grown in epoxy in Sample AFM(1)-G1: a) lit from above, b) lit from below. Circled are two points at which the channels are exposed.

The top-lit sample in Figure 3-13a) provides an image of the surface including two apparent holes in the surface which are circled; a white reflective material is also visible below the surface to the left of this point. The bottom-lit sample from the same location is given in Figure 3-13b) which provides a projection of the channel below the surface. The holes in the circled area can be seen to correspond to tree tips which have reached the surface while the white reflective material is confirmed as coming from within a channel. This level of control is useful in the preparation of any sample, but in particular when looking to study degradation occurring within, around and in front of channels. To accurately study them it is necessary that the surface be parallel to the channel.

Other Potential Preparation Methods Other techniques were considered in cross-sectioning the samples however were not viewed as providing the same control as the grinding-polishing technique.

Microtoming is a technique in which the surface can be cut away in small, extremely precise slices to approach the sample. Varieties of this exist in which the slices themselves, hundreds on nanometres in thickness, can be retained and studied under the AFM-IR. Microtoming is a technique which has previously been used for studying electrical trees and water trees [108]. However here, with the requirements of sectioning channels with precise angles and the size of channels in this epoxy being 1-2 µm [107] (or lower with fine trees [20]), it was felt that the control necessary could not be achieved. In particular combining the optical imaging with the microtoming was not seen as practical.

Serial block-face scanning electron microscopy (SBFSEM) was also considered, there is experience using this technique in Manchester [109] which combines microtoming with layer-by-layer SEM to build 3-dimensional models of the electrical trees. This could be used in finding the precise points at which to stop microtoming a channel once the tree has become exposed. However, this was not pursued due to concerns such a technique would struggle to discern between the tip of a channel being exposed, and a channel which extends further into the epoxy below the surface. There are potential ways this technique could be integrated into the current preparation method and these are discussed in the future work, Chapter 7.

Cryo-fracturing or cryomictromy at low temperatures has been used in different studying techniques [110] however was not pursued in this case, due to concerns over the precision of the technique and the observed success of the grinding-polishing method. This does remain an option worth investigating however particularly in the case of polyethylene which, being less rigid than a glassy- state epoxy, may not be able to be polished in the same manner.

3.2.3. AFM-IR Operation and Data Analysis This section will discuss the use of the AFM-IR, following on from the detail provided in the background section it will discuss the specifics of its application and the system used here. It will then discuss the data processing which is applied to each sample, which is done to reduce the effects of noise and to ensure samples are well normalised.

3.2.3.1. NanoIR2 System The AFM-IR instrument used is the NanoIR2 manufactured by Anasys Instruments. This uses top- down illumination from a pulsed, tuneable laser which operates across the IR range from 500 cm-1 to 4000 cm-1, this is done with a 4 cm-1 spectral resolution. The laser source is an optical parametric oscillator with an in focus spot size of 30 µm, the high resolution of the AFM-IR technique coming from the AFM probe, not the IR illumination. The AFM probe used is made from silicon nitride with gold coating. The power of the laser varied significantly based on sample and ambient conditions. It was determined on a sample by sample basis and was typically between 2 and 16 %.

In measuring each spectrum, to improve the accuracy of the analysis, a number of measurements are performed at each wavelength. These absorption measurements are then averaged together to give a more accurate data point for this wavelength and location. For spectra 1024 co-averages were taken for each wavenumber of the spectrum. Similarly for the spectral maps at fixed wavelengths, the absorption shown at each point is the product of a number of readings averaged together. These typically used 16 co-averages for each point on the map with a scan rate of between 0.08 and 0.1 Hz, although in some cases 32 co-averages were taken for more confident readings. The points per map was dependent on the size of the map but a standard number of points was 300 points per 300 lines for a 30 x 30 µm map, although this would sometimes be increased.

3.2.3.2. Use of the NanoIR2 A sample, prepared as described above, is placed within the AFM-IR, the NanoIR2 is equipped with an optical microscope allowing the chosen sample location to be found. The AFM mapping function can then be performed, allowing a surface profile to be generated. Such a surface profile is given in Figure 3-14 (bottom). Using this profile, locations can be identified from which to take a spectrum, as shown in Figure 3-14 (top).

The AFM-IR analysis of a single sample will take approximately a day. As it is an expensive and in high-demand piece of equipment time on it can be restricted giving limited sample numbers. It is important then to carefully prepare samples to achieve a high quality surface and to select samples carefully to maximise the findings from the limited time.

100

Figure 3-14: Top: Example spectra taken by the AFM-IR from different locations on Sample AFM(1)-G1. Bottom: The locations from which these spectra are taken are shown on a surface profile image generated by the AFM-IR. (The locations are chosen to show distinct chemical signatures: the blue marker is chosen to be close to a branching point, the green marker is close to the subsurface channel and the red marker is a large distance from the channel).

These spectra can then be used to give immediate insights into the types of degradation products being formed and where these are forming. This can then be used to inform the wavelengths selected for the spectral mapping, as well as through applying knowledge of the material and likely degradation mechanism in this selection process. The spectra and maps then go through post- processing and data analysis as described in the following section.

3.2.3.3. Data Processing This section describes the post-processing applied to the spectra and maps output through the AFM- IR. These processes such as normalisation and smoothing are intended to ensure that the information taken from them is more informative and easier to draw accurate conclusions from. However these should be clearly stated, such that the data can be properly understood and any artifacts inserted into the results can be explained without compromising the results.

First the spectra will be discussed. As stated earlier, each spectrum produced is the product of a number of readings which are co-averaged to minimise noise. This reduces the effect of any individual fluctuations which can occur due to effects such as variations in the structure near the surface (which increase or decrease the observed impact of the thermal expansion) or the surface topography which can affect the contact with the AFM probe. The effect of normalisation is shown in Figure 3-15.

101

a) b)

Figure 3-15: a) Unnormalised spectra and the locations they are taken from, b) The same spectra having been normalised against the 1500 cm-1 peak. Taken from Sample AFM(1)-I1.

Figure 3-15a), on the left is the unnormalised spectra, the Figure 3-15b) on the right is the same spectra having been normalised against the 1500 cm-1 peak. This peak is consistently chosen for normalisation throughout the project as it corresponds to the aromatic group within the epoxy. The aromatic is a very stable bond [111], making it unlikely to change from location to location. The 1500 cm-1 peak also has the advantage that it does not overlap with other absorption bands. The use of this bond is standard for normalisation however it does carry the assumption that the number of aromatic groups is homogeneous across the sample.

There is a clear difference between the unnormalised and normalised spectra. This is clearest in the red and dark green groups which see significant variation across the IR range in the first image, yet when normalised this variation largely disappears. The light green spectra (which are the most degraded) however still have visually different spectra, indicative of different functional groups having formed during ageing. Spectra will often be grouped by colour in this project (typically based upon location).

A second process which the spectra undergo during post-processing is smoothing. This is done to minimise the effects of noise. The Savitzky-Golay filter is used and parameters (polynomial order - 7, side points – 4pt) are chosen with the advice of Dr Suzanne Morsch, an academic with significant experience in studying epoxy using the AFM-IR. This is a very widely applied process in the analysis of IR spectra using convolution with the method of least squares designed to allow the identification of spectra in samples whilst reducing the effect of noise [112].

The spectral maps are also subject to a number of post-processing procedures. As with the spectra, and as mentioned before, each map given in this thesis will be the product of a number of maps averaged together to reduce noise effects. These maps will occasionally be given in this thesis

102 however typically maps will either be Flattened or a Ratio will be taken against a second map, in a process similar to normalisation.

Maps without Flattening or without a ratio are not typically used here, flattening can remove line artifacts, produced for example by the AFM tip collecting debris and changing the effective height along the surface or through surfaces not being flat. While every effort is made to ensure that in sample preparation and in placing it within the AFM-IR that the sample is flat, this will never be perfect. The Flatten tool allows each line to be viewed as if though they were at the same height.

The Ratio technique is the most commonly used to study samples in this project. Here, like normalisation, maps at wavelengths likely to experience changes after electrical ageing are normalised against a map at a different wavelength (here 1500 cm-1 and 1504 cm-1 are commonly used due to corresponding to the aromatic peak) which is unlikely to change with ageing. AFM-IR readings at different locations on a sample can differ for many reasons besides different concentrations of the active functional groups. In particular, changes in topography and the material structure can appear as different levels of photon absorption. Applying a Ratio against a map in which we expect all fluctuations in absorption to be due to changes in topography and the material structure allows removal of such impacts.

The final aspects of post-processing which must be considered are the Contrast and Offset settings. These change the colours used to produce the maps showing the absorbance and will typically be consistent for images looking at the same sample. In some cases these will be changed to allow them to focus on particular aspects of the chemical degradation, in these cases this will be made clear.

3.2.3.4. Surface Profile Analysis Surface roughness is a crucial aspect of the quality of measurements taken using the AFM-IR, with rough surfaces producing more noise and artifacts in the results, so care was taken in preparing samples to ensure they have smooth surfaces. The surface roughness of samples was measured using the AFM which has a profile analysis tool which is shown below in Figure 3-16.

103

a) b)

Figure 3-16: The surface roughness of two lines across Sample AFM(2)-10 are shown in the graph in a) (the y-axis of which goes from +40 nm to – 60 nm centred on the average height. The lines tracked across the samples are shown in b).

These graphs are not shown for each sample, however the surfaces roughness is referred to at points to give explanations for samples having clearer or less clear spectra and maps. It is also intended that this description gives an insight into the surface quality necessary to achieve clean signals using the AFM-IR.

3.2.4. Conclusion This section has presented the AFM-IR technique and provided a sample preparation methodology utilising polishing and grinding to expose channels at the surface of samples while achieving a smooth surface. The next chapter will present the application of this technique and the results obtained in doing so, providing a chemical analysis of both tree initiation and tree growth.

104

4. Atomic Force Microscopy with Infrared Spectroscopy (AFM-IR) - Testing and Results

This chapter provides details on testing and the results from the AFM-IR. These are provided in two sections; the first section looks at the initiation stages of electrical treeing (both pre-initiation and post-initiation samples are considered independently), the second section looks at mature channels. In both sections, it is seen that observing both the type of chemical ageing occurring, as well as the location of this ageing is crucial to understanding the mechanisms which underlie electrical treeing.

4.1. AFM-IR Tree Initiation Testing Results

This section will consider tree initiation, looking primarily for degradation occurring around the needle tip (the degraded region) as well as the initial channel formation.

4.1.1. Pre-initiation Tests The study of tree initiation can be considered to cover the pre-initiation degradation, in which chemical degradation occurs around the needle tip and becomes more severe over time, through to tree formation. Samples looking at chemical degradation prior to tree initiation are considered first here. Looking at the chemical changes induced, how these change with voltage application time and how these change spatially. The first sample (Section 4.1.1.1) shows how the degradation of the polymer differs when in direct contact with the needle tip, compared to regions in bulk. The second sample is in the earliest stages of degradation, in which methyl groups have formed preceding carbonyls (Section 4.1.1.2). Another sample (Section 4.1.1.3) shows different groups forming in different locations.

4.1.1.1. AFM(1)-A2(1) – Aged but Uninitiated Sample The first sample considered in this project, was one of the first tested using the AFM-IR. It has been aged for 10 minutes at 15 kVpk and there is no visual evidence of tree formation having occurred. The area surrounding the needle tip has been exposed perpendicular to the direction of the needle tip, creating a cross-section as illustrated in Figure 4-1. It is worth remembering at this point that the needle has already been removed prior to this and the needle void is what is cross sectioned.

105

Metallic Needle

Epoxy Resin

3 µm radius

Figure 4-1: A Schematic of the void created within the sample following the removal of the needle tip. The dashed line represents the cross-section created through polishing.

a) b)

Figure 4-2: Optical images from Sample AFM(1)-A2(1) taken a) during polishing, before exposing the void from the needle tip and using back-lighting, b) after polishing the void is exposed and the sample is front-lit emphasising the surface condition.

During and following polishing the samples are optically imaged in Figure 4-2.

Figure 4-2a) is a projection of the needle tip void (the needle has been removed) through lighting from below the sample. Here, the location of the needle tip can be seen and there is no evidence of tree formation. Figure 4-2b) on the right is a top-lit image which shows the surface around the needle tip after polishing. The location of the tip of the needle is clearly exposed however there is significant surface roughness from the polishing process, particularly around the needle void which will be seen to create noise in the AFM-IR measurements.

106

Figure 4-3: Top: Spectra for Sample AFM(1)-A2(1). Bottom: AFM-IR height profile showing the location of the spectra

Spectra are taken from a number of locations around the sample and these are given in Figure 4-3. The spectra are colour coded according to the location from which they were taken. Due to the high precision of the AFM-IR technique, with around a 50 nm resolution, the location of each of these spectra is very well defined and is indicated in the surface profile, also given in Figure 4-3. The spectra are normalised against the 1504 cm-1 peak, corresponding to the aromatic groups within the epoxy. These are relatively stable and expected to not experience significant degradation in most cases. Normalisation against this peak is a common technique [77, 113].

The red spectra (representing regions closest to the needle tip) are distinct from the other spectra, showing increases in a number of locations including a small increase across the carbonyl region (1650-1800 cm-1) and from 1400- 1200 cm-1 within the fingerprint region. There is also a singular spectrum which shows an increase from around 1150-950 cm-1.

Spectra from locations within the void created by the needle tip were not included in this figure due to the extent of degradation they experienced making them impossible to normalise effectively (this is discussed later) meaning any comparison with the other spectra would be unreliable.

To further understand these spectra and the potential chemical bonds indicated by the spectra, maps are taken at different wavenumbers mapping the level of absorption in the region around the needle tip. These are given in Figure 4-4.

107

It should be noted not all of these maps use the same contrast settings. These settings have been adjusted where necessary to ensure the variations in absorption are visible in the maps.

Degradation can be seen around the needle tip, with what appears to be C=C or NH bonds being formed as shown in Figure 4-4g) and h). However, it is difficult to pick these with confidence due to the roughness of the surface across the bulk (with variations of around 500 nm); in particular around the needle tip (with a variation of over 1500 nm). This surface roughness was due to the polishing process used. Determining how far the chemical damage extends from the needle tip is particularly difficult here. It was clear from the testing of this sample that the surface must be prepared with a higher level of precision to achieve a high level of chemical sensitivity. This was followed through by modifications to the polishing procedure which in later samples will be seen to have improved surface quality and a consequentially improved accuracy of the chemical testing.

a) b) c)

d) e) f)

g) h) i)

j) k)

Figure 4-4: AFM-IR absorption maps for Sample AFM(1)-A2(1); a) Surface Profile, b) 1056 cm-1, c) 1132 cm-1, d) 1248 cm-1, e) 1288 cm-1, f) 1448 cm-1, g) 1604 cm-1, h) 1656 cm-1, i) 1708 cm-1, j) 1726 cm-1, k) 1742 cm-1

108

The spectra taken from within the needle tip void demonstrate an extremely high level of damage to the epoxy molecular structure as shown in Figure 4-5. Here, even in comparison to a point very close to the needle tip (within a few micrometres), the spectra show totally different responses. The spectra from within the needle tip void, given in black, have a number of troughs at key wavelengths indicative of the loss of the corresponding molecular bonds. Most important of these in terms of the use of AFM-IR is the 1504 cm-1 trough, which correlates to aromatic groups and in undamaged epoxy produces a distinct peak (as can be seen in the spectrum shown in red). These are commonly used to normalise spectra due to being a relatively stable group, and unlikely to see significant variations in most cases. Unfortunately, that the aromatics are being broken down within the needle tip means that the 1504 cm-1 peak cannot be used for normalisation of spectra taken from such severely degraded regions. As a consequence of this, the areas of epoxy in direct contact with the needle tip are not typically analysed in this project however this is evidence of extreme degradation in these locations.

It is apparent very high levels of degradation have occurred in these regions in contact with the needle tip. Although not studied in depth in this project it is believed to be likely that carbon-carbon bonds have formed in greater numbers (carbonisation) here and this could be tested using Raman spectroscopy. The dielectric in direct contact with the needle tip is found to be chemically very distinct from the bulk material. This can only be distinguished through high resolution analysis such as the AFM-IR.

Figure 4-5: Top: Spectra for Sample AFM(1)-A2(1) from within the needle tip void. Bottom: AFM-IR height profile showing the location of the spectra

109

4.1.1.2. AFM(2)-5 – Earliest stages of pre-initiation degradation

Sample AFM(2)-5 was aged for 20 minutes at 8 kVRMS. Again there is no evidence of tree formation as seen in the optical images or in PD measurements. It appears to provide an insight into an early stage of pre-initiation chemical degradation which differs from those previously identified. However this should be taken with caution and requires further samples and testing before reproducibility can be established.

The sample was again polished down to expose the area in which the needle tip was located. However due to the improvements in polishing technique (force applied, amount of water used, polishing direction relative to the needle tip void) the surface is significantly smoother both around the needle tip void and across the surface of the material. This gives us significantly cleaner AFM-IR measurements than obtained from the first sample reported, Sample 2 AFM(1)-A2(1).

Figure 4-6: Top: Spectra for Sample AFM(2)-5. Bottom: AFM-IR height profile showing the location of the spectra

110

In Figure 4-6 the spectra taken from locations closest to the needle tip are identified in red (except for one location which is given in black due to presenting a particularly distinct signal). Spectra from further out (~8 µm) are coloured blue and those even further out are given in green.

All the spectra are largely similar to each other in the carbonyl region of the IR spectrum (1800 - 1600 cm-1) indicating there is no carbonyl formation in this sample. And while significant variation occurs there is no consistent pattern in the fingerprint region of the IR spectra (1500 - 400 cm-1) to suggest a change in chemistry closer to the needle tip. This is except for the black spectrum, which with a very strong peak at 1450 cm-1 and an additional strong peak at 1372 cm-1, could be indicative of larger concentrations of methyl groups (CH3) or CH2 groups in this region. Other smaller peaks are identified in this black spectrum, however due to the inherent noise of the fingerprint region these cannot be confidently identified as having increased more than other spectra. Before concluding upon this further confirmation is taken using the spectral mapping tool of the AFM-IR.

a) b) c)

d) e) f)

g) h) i)

Figure 4-7: AFM-IR Absorption maps for Sample AFM(2)-5 a) Surface Profile, b) 1056 cm-1,c) 1248 cm-1, d) 1360 cm-1, e) 1448 cm-1, f) 1604 cm-1, g) 1656 cm-1, h) 1726 cm-1, i) 1742 cm-1

111

The infrared absorption is shown in the spectral maps taken by the AFM-IR and given in Figure 4-7. Figure 4-7a) is a surface profile from the AFM and demonstrates greatly improved surface quality, with a measured surface roughness of less than 100 nm, the needle tip void can be seen in the bottom right of the map. As before, absorption maps were taken at a number of wavenumbers selected for being typically relevant for analysis of the degradation of polymers including epoxy. Selected maps are given here.

Immediately notable is the increase in absorption along the bottom of the maps in Figure 4-7b), d), e) and f) (wavenumbers 1056 cm-1, 1360 cm-1, 1448 cm-1, and 1604 cm-1). This increase in absorption corresponds very well with the black spectrum identified in Figure 4-6 which saw a particularly distinct increase around 1450 cm-1, now supported by these maps. The fingerprint region is complicated, and many IR absorption peaks will overlap with each other in this region. This makes identifying the functional groups responsible for a particular peak very difficult. As proposed

-1 previously, the 1448 cm map is believed to correspond to CH2 or CH3 groups. Notably there is also an increase at 1360 cm-1 (Figure 4-7c)) however this is less definitive than for the other wavenumbers. This could be explained by the peak being at 1375 cm-1 as identified in the spectra (Figure 4-6) meaning that at 1360 cm-1 there is only a slight increase. 1056 cm-1 could be assigned to -CO groups, however it has also been assigned to hydroxyl groups (-OH) [77] in tracking analysis tests. While 1604 cm-1 is likely indicative of alkene groups (C=C), aromatics are also absorbed at this wavenumber. However, as the maps are normalised against the 1500 cm-1 map, this is cannot be explained by aromatic groups. It should be noted that other than the 1450 cm-1 peak none of these increases are well supported in the spectra making discerning the cause of them by spectral analysis more difficult.

Also consistent with the spectra from Figure 4-6 is there appears to be no carbonyl formations identified in these maps. Figure 4-7g), h) & i) cover the most common wavenumbers of absorption due to carbonyl groups (C=O) yet identify no change across the region tested. This appears to indicate that although we are currently seeing chemical damage and change due to the application of high electrical fields at the needle tip, we have not yet reached a stage at which carbonyls have begun to form in this sample. Our understanding on pre-initiation degradation would anticipate that carbonyl formation would occur with the first stages of chemical degradation. Indeed carbonyl formation is repeatedly used as the primary indicator of whether chemical degradation has occurred in a number of papers [13, 114].

These results raise the question of how certain we are that chemical degradation has occurred. And if it has, then what chemical change has been seen and why does this precede carbonyl formation?

112

For the first of these questions, the maps are consistent across a number of wavenumbers showing an increase in absorption in the same location. Yet the wavenumbers corresponding to the carbonyls consistently identify no change in absorption in that location, suggesting the change identified here is chemical in nature. This is further supported by the spectra, in particular the spectra highlighted in black. Here a large peak at the 1450 cm-1 matches the increase in absorption in the maps, both in wavenumber and in location. Further evidence for a chemical change distinctive from carbonyl formation is now discussed in Sample AFM(2)-10.

4.1.1.3. AFM(2)-10 Multiple areas of pre-initiation degradation identified

Sample AFM(2)-10 is another pre-initiation sample, aged for 40 minutes at 8 kVRMS. This is twice as long as the sample AFM(2)-5 which was previously discussed. We can expect, and will see that this sample has experienced greater levels of chemical degradation. The similarities and differences of these will be observed and discussed with the aim of identifying how degradation changes over time prior to tree initiation. This sample was studied in a paper which has been submitted to ICD 2020, “A High-Resolution Study of Chemical Aging Prior to Electrical Tree Growth” and which is provided in Appendix B-4 (Chapter 9).

Through PD and optical imaging there is no evidence of tree initiation having occurred within this sample. And again the sample is polished down to the needle tip void prior to AFM-IR testing.

Figure 4-8: Top: Spectra for Sample AFM(2)-10. Bottom: AFM-IR height profile showing the location of the spectra

113

Spectra are taken from around the needle tip (Figure 4-8), with the areas closest to the needle tip given in red, areas, areas further out given in blue and those furthest away (around 10 µm) given in green. Black has been used to highlight a number of spectra which are taken from chemically significant locations (which is discussed in the chemical maps section). The spectra in these locations show somewhat high absorbance around the 1450 cm-1 peak, however it is not as significant as that found in the previous sample AFM(2)-5. Most significant are the high absorptions at 1250 cm-1 and from around 1140 cm-1 to 1010 cm-1 as well as a low level of absorption from 1430 – 1270 cm-1.

The spectral maps, Figure 4-9, show that the chemical degradation can be split into three different groups in this sample. These are distinct in location, demonstrating no obvious relation to each other as illustrated in Figure 4-10. In each case the degradation was within 10 µm of the needle tip.

a) b) c)

d) e) f)

g) h) i)

j) k) l)

Figure 4-9: AFM-IR Absorption Maps for sample AFM(2)-10 a) Surface Profile, b) 1056 cm-1, c) 1132 cm-1, d) 1248 cm-1, e) 1288 cm-1, f) 1360 cm-1, g) 1448 cm-1, h) 1604 cm-1, i) 1656 cm-1, j) 1702 cm-1, k) 1726 cm-1, l) 1742 cm-1

114

Figure 4-10: Schematic illustrating the location of three distinct groups of degradation identified in Sample AFM(2)-10

Case 1, on the left, indicates the locations in which carbonyls and alkenes were found to have formed around the needle tip. The wavenumbers for these were 1656 cm-1, 1702 cm-1, 1726 cm-1 and 1742 cm-1 (Figure 4-9i)-l)). The chemistry for how these could form is well covered in the literature, and is discussed in the background for this chapter (Section 3.1.1).

Case 2 captures other forms of chemical change more difficult to assign to specific functional groups and covers the wavenumbers 1056 cm-1, 1132 cm-1, 1248 cm-1 and 1288 cm-1. Potential assignments for these wavenumbers include 1056 cm-1 (-CO in an alcohol and less commonly –OH groups), 1132 cm-1 (which could be indicative again of –CO groups forming this time in an ether), 1248 cm-1 (again – CO in an ether). 1288 cm-1 could be assigned to –CO groups in aromatic esters, however this is not likely given the lack of C=O groups in same region making alkyl aryl ethers a more probable assignment here.

Case 3 is the shape of the region in which the 1448 cm-1 (Figure 4-9g)) absorption was increased, likely indicative of CH2 or CH3 formations. Whether this is in some way related to Case 1 or 2 or indeed neither is not clear based upon the size, shape and location of the degradation region.

Analysis of the location in which full spectra were taken reveals that due to unfortunate choices of location it is difficult to resolve Case 2 from Case 1 in any of the spectra, as illustrated in Figure 4-11 below. Each spectrum covered in Case 2 degradation was also within the region of Case 1 degradation. Due to limited time on the AFM-IR (compounded by the equipment being broken for most of 2019) it was not possible to retest the sample.

115

a) b)

Figure 4-11: AFM-IR absorption map for sample AFM(2)-10 showing absorption at a) 1248 cm-1 and b) 1726 cm-1

Figure 4-11a) on the left (taken at 1248 cm-1) shows highlighted in black the only three spectra to sit within the area of main degradation for Case 2. Figure 4-11b) on the right (taken at 1726 cm-1) demonstrates that those same three spectra were each within the area of degradation for Case 1. This complicates the already tenuous assigning process and means the spectra may create distorted results.

Case 2 is suspected to be indicative of ether formations (C-O-C). These are able to form due to reactions between two alcohol/hydroxyl (-OH) groups or between a hydroxyl group and an unreacted epoxide group. In particular Alkyl-Aryl-Ethers appear likely based upon the spectra and absorbance maps. These are ethers which contain alkane and aromatic (aryl) functional groups within their chain. These could be explained in a number of ways:

1) Incomplete reactions during the curing and post-curing stages being activated under higher fields/energy input from the electrons. It is unclear why this would induce further reactions which did not occur during the curing and to post-curing stages of sample preparation. However, it somewhat fits the results as there are ethers identified within the epoxy and etherification is a known reaction within epoxy during the curing process [115]. However, alkyl-aryl-ethers might not be expected to form within this epoxy as there is no obvious route for them to form. This is also not something seen before in other samples in this project, raising questions on why this is occurring here and not elsewhere. 2) Imperfect mixing, we have a limited amount of time to mix samples before they start to harden and if some level of grouping occurred you could have a higher level of alkyl-aryl ethers there for some reason. This seems unlikely as variations like this have not been observed elsewhere.

116

3) Chemical degradation occurring, affecting –OH bonds and producing ethers from them. The main problem with this is there is no OH bond attached to the aromatics; raising the question of how the C-O-Aromatic groups form.

It is impossible currently to discern with confidence why high levels of ethers appeared here, and whether this is ‘degradation’ due to the applied electric field or of a further curing occurring due to it. Alternatively this could simply be related to the sample preparation/curing and unrelated to the electrical ageing. As little can be determined on why these have occurred, nor their influence on the electrical treeing process they are noted here but not discussed in detail.

Case 3: Seemingly distinct from Case 2, and seemingly not explained by further curing/epoxification reactions. These results are similar to those identified in the previous sample AFM(2)-5 which was aged for a shorter period of time. A region was found to have an increase in absorption at the 1448 cm-1 wavenumber. Though this can correspond to aromatics, it is considered unlikely as the absorption at 1500 cm-1, which these maps are normalised against, also corresponds to aromatics so any changes should be minimal. Most likely this can be explained by Methyl (CH3) groups having again formed in this region. This region is close to and overlaps with that of the carbonyl groups (Case 1), but is significantly larger. This is consistent with methyl groups forming prior to carbonyl groups, and would explain why they cover a larger region. The close and overlapping regions could suggest the processes are related. This relationship could be as involved as one functional group directly reacting to form the other, or as weak as this being the region of highest field or charge injection promoting both in this region.

117

4.1.2. Post-initiation Tests This section will now move on to discuss analysis of the tree channels after initiation has occurred. It will cover two samples, one of which was imaged at two different spatial planes (effectively giving three case studies). In the second of these (Section 4.1.2.2) the channel is studied having had the voltage removed immediately following initiation, it will show chemical degradation does not occur during tree formation. In the most aged of these cases (Section 4.1.2.3) a typical spherical region of degradation around the needle tip is identified. Again consider the impact of increased periods under high fields around the needle tip, with very different chemical signatures identified in these cases. This section will also question what conditions are necessary for electrical tree initiation, including providing evidence suggesting tree initiation is not a discharge driven event and an alternative mechanism is likely to be active during tree initiation.

4.1.2.1. AFM(1)-A1(2) – Earliest stages of growth (1st Examination) The first sample considered is one in which the voltage was removed immediately following initiation. A tree of around 20 µm length, reaching about 10 µm from the needle tip was formed and had just undergone bifurcation. The tree is imaged in Figure 4-12.

Partial Discharge Analysis The tree was aged at 15 kV for 7 minutes before optical evidence of tree initiation was seen and the voltage was immediately removed. There was very little partial discharge activity associated with the tree initiation or growth as shown in Figure 4-13 with 112 pulses above 500 fC recorded, the largest of which was around 1.5 pC and the large majority being below 1 pC. Due to the difficulties in confidently identifying when a tree has initiated it is not possible to determine whether these discharges occur during or after tree initiation, but they appeared to occur around the time of tree initiation. Before chemical analysis was performed it was calculated whether a tree this size could be formed by conversion of the energy associated with these discharges. The results of this are not conclusive, however suggest discharges are less likely to be responsible for the channel formation. This work is provided in Appendix A.

20 µm

Figure 4-12: Optical Image of a tree formed in Sample AFM(1)-A1(2), approximately 15 µm in length.

118

Figure 4-13: The partial discharges measured associated with AFM(1)-A1(2)

Chemical Analysis To further study the initiation process the AFM-IR can be applied to study any chemical changes which have occurred within the channel and the surrounding areas. The tree in Sample A1(2) (1st Time) is studied at two different locations approaching the channel providing a fuller picture of the chemistry around the needle tip.

The sample is polished down to approach the needle tip void. In the study it is polished down to approach the channel however the channel remains well below the surface. Unfortunately it is not possible to determine exactly how far below the surface. This situation is illustrated in Figure 4-14.

The dotted line in Figure 4-14 represents the surface the sample is polished down to, approaching but not reaching the channel. This surface is then analysed using the AFM-IR, first using spectra around the needle tip void (Figure 4-15).

119

Figure 4-14: Figure: Schematic illustrating polishing cross-section in relation to needle tip void and tree channel.

Figure 4-15: Top: Spectra for Sample AFM(1)-A1(2). Bottom: AFM-IR height profile showing the location of the spectra

120

The spectra have been divided into different coloured groups. Light blue spectra are from locations close to the needle void but on the left-hand side as we see it. Dark blue spectra are also from locations close to the void, but from the right-hand side. These spectra may actually be above the void in some cases, here the thickness of the material and the sensitivity of the AFM-IR create uncertainty what exactly is measured in these spectra. These may or may not include the interface itself, as shown in Figure 4-5 such measurements can be difficult to interpret. Red spectra are taken from areas further from the exposed area (~ 5 µm in this cross-section which will mean a greater distance from the needle tip) and black spectra from even further out (~ 9 µm). The area to the left- hand side has been studied more carefully as this is the side the tree initiated from.

Immediately apparent is the increase in absorption in the 1745 cm-1 peak for the light blue spectra. This is certainly indicative of carbonyl (C=O) formation here, though the exact functional group is more difficult to determine. We have previously assigned this peak to aldehyde formations (-CHO) and while this peak fits the typical aldehyde output well, aldehydes are relatively unstable. They will oxidise rapidly to produce carboxylic acids (-COOH) meaning they would not be anticipated to form in such concentrations. The peak could then be assigned, at least in part, to carboxylic acid. However, this would be a relatively high wavenumber for them to absorb (typical peak is 1760 cm-1 although small shifts can happen depending on the precise molecule). Esters (-COOR) may also absorb around this wavenumber however the lack of an accompanying peak in the 1210 – 1160 cm-1 range (or even nearby) for C-O bonds would suggest this is less likely. In conclusion, a tentative assignment is made of the 1745 cm-1 being derived from either aldehydes or carboxylic acids (or more likely a mixture of the two).

There are no other consistent and distinct indications of chemical change within the light blue spectra, and the red and black coloured spectra, situated further from the needle tip show no obvious chemical change. The darker blue spectra (the right-hand side of the needle tip void) however show an increase in absorption from 1400 – 1000 cm-1. Particular peaks are identified at 1248 cm-1 and from 1130 – 1100 cm-1. These peaks have been assigned to the formation of ethers (C- O-C), as identified in the pre-initiation samples (which would produce absorption at 1248 cm-1), and hydroxyl groups (OH), which can produce absorptions between 1200 cm-1 and 1100 cm-1 depending upon the exact molecule.

Spectral maps are used to further investigate sample AFM(1)-A1(2), however there were issues encountered in using the AFM-IR on this sample. Lines are formed across the maps, making interpreting them difficult, and are suspected to be related to power fluctuations in the IR laser during operation. The sample was retested due to this, however this effect appears in both cases.

121

The impact of this is demonstrated in Figure 4-16 and discussed below; following this an attempt is made to discern information from the maps however the confidence in these is more limited than in other cases.

The maps in Figure 4-16 are given to show the difficulty in analysing this sample effectively. Horizontal lines can be seen to form in each ratio image; these are believed to form due to power fluctuations in the laser. These are clearly seen to occur in the unaltered maps in Figure 4-16d) and e). The only exception is Figure 4-16c) in which the image is ‘flattened’, a process designed to remove line artifacts from the AFM-IR process. These maps however cannot be confidently normalised against the 1500 cm-1 map meaning surface effects cannot be eliminated.

The strong horizontal lines from the ratio process, shown in Figure 4-16b), are impossible to remove however attempts are made to mitigate them through the selection of contrast settings. This sample was tested on two occasions due to this effect, however it is believed the issues which produced the laser power fluctuations were present in both cases.

Selected ratio maps (each taken from the same set of tests) are given in Figure 4-17.

a) b) c)

d) e)

Figure 4-16: AFM-IR Absorption Maps for Sample AFM(1)-A1(2) a) Surface Profile, b) 1604 cm-1 ratio against 1500 cm-1, c) 1604 cm-1 flattened, d) 1604 cm-1 original image, e) 1500 cm-1 original image (increased contrast settings).

122

a) b) c) d)

e) f) g) h)

i) j)

Figure 4-17: Sample AFM(1)-A1(2) AFM-IR Absorption Maps: a) Surface Profile, b) 1132 cm-1, c) 1248 cm-1, d) 1360 cm-1, e) 1448 cm-1, f) 1604 cm-1, g) 1656 cm-1, h) 1702 cm-1, i) 1726 cm-1, j) 1742 cm-1

It is important to note because of the point at which the cross-section was taken (Figure 4-14) the areas closest to the exposed area in these maps are likely directly above the needle tip void and may well be capturing chemical degradation from the interface which has not necessarily penetrated into the bulk. As in previous cases, carbonyl areas are identified as having formed, albeit only close to the exposed region. These are clear in Figure 4-17i) and j) at 1726 cm-1 and 1742 cm-1 respectively. These are consistent with the spectra (Figure 4-15) in which the light blue markers closest to the exposed region showed a large increase at 1745 cm-1 which was not reflected in the other markers. The other map which showed a clear increase (as in many maps the signal was unclear) was at 1248 cm-1 (Figure 4-17c)), the degradation here being similar in location to the carbonyls. Notably there was very little response at 1656 cm-1 and 1702 cm-1 (Figure 4-17g) and h)) which suggests carbon-carbon double bonds had not formed in any significant numbers at this point.

4.1.2.2. AFM(1)-A1(2) – Earliest stages of growth (2nd Examination) The sample AFM(1)-A1(2), was further polished to reach the initiated channel and ensure the chemical degradation within and surrounding the channel was captured in the AFM-IR chemical testing. This tree was discussed in the journal paper "Chemical Analysis of Tree Growth in Epoxy Resin Using AFM-IR Spectroscopy” which has been accepted for publication and can be found in Appendix B-1 (Chapter 9).

123

Figure 4-18: Schematic illustrating polishing cross-section in relation to needle tip void and tree channel in Sample AFM(1)-A1(2).

a) b)

10 µm

Figure 4-19: Sample AFM(1)-A1(2), cross-section as in Figure 4-18. a) Top and Bottom Lit and b) AFM-IR Surface Profile. Circled is the point at which the channel reaches the surface in each image. Also visible in the optical image is the rest of the channel beneath the surface.

The optical image in Figure 4-19a) shows the proximity of the initiated tree to the surface. The full length of the channel can be seen immediately below the surface and is exposed at the tip (circled). The AFM-IR surface profile in Figure 4-19b), shows the exposed sections of the channel, the tip is again circled where exposed and a second exposed area close to the needle tip can be seen. There is no question on the proximity of the channel to the surface in this case, as it is clearly visible

immediately below the surface and is exposed at points.

124

Figure 4-20: Spectra for Sample AFM(1)-A1(2). Bottom: AFM-IR height profile showing the location of the spectra. Circled spectra are above or near the channel.

Spectra were taken around the needle tip and around the exposed areas of the channel (Figure 4-20). Locations marked with a green marker are closest to the needle tip, blue are from an area slightly further away and red even further from the needle tip. Spectra (circled) are also taken at areas from around and above the channel. o Measurements did not indicate any degradation associated with the channel itself, spectra taken from close to the needle tip and from further from the needle tip were entirely in line with other spectra around the channels. This indicates there was no degradation occurring here and suggests the initiation process is not driven by discharges. This finding is consistent with an electromechanical fracturing process of initiation. This is expanded upon in the spectral maps section and the discussion. o The region around the needle tip is also of great interest here (green marked spectra), it appears to differ from the previous cases notably in the measured carbonyl concentration surrounding the needle tip. Previous cases of extended electrical ageing have shown larger numbers of carbonyls surrounding the electrode, both within this thesis and in literature [11]. These results however show quite the opposite, an increase in carbonyls was identified in spectra further from the needle.

125

a) b) c) d)

Figure 4-21: Sample AFM(1)-A1(2). a) Surface Profile. Absorption amplitude at wavelengths b) 1650 cm-1, c) 1726 cm-1, d) 1752 cm-1. Circled in each image is the location at which the channel has been exposed at the surface.

The spectral maps give results which are largely in agreement with those identified using the spectra. There is again no evidence of any degradation products forming in association with the channel and this is highlighted in Figure 4-21.

Circled in each map is the exposed section of the channel, showing no increase in degradation products at this point. This was true for every wavelength examined. It can also be seen the channel extends far beyond the degraded region (particularly clear in Figure 4-21b)) indicating that while this region may have been relevant in the tree formation, it is not directly the limiting factor in how far it is able to extend from the needle tip.

The nature of the degraded region is also called into question by sample AFM(1)-A1(2). Here strong evidence of degradation surrounding the needle tip is found in the 1650 cm-1 and 1132 cm-1 wavenumbers, which we have assigned to the C=C and CO bonds. However no degradation in the carbonyl regions is identified around the needle tip; in fact the degradation appears to be reduced as evidenced in the 1726 cm-1 and 1752 cm-1 maps. This not only contradicts previous findings, it is physically unintuitive, with degradation appearing to increase away from the energy source. This was further investigated using spectral maps which had not been ratioed against the 1504 cm-1 (aromatic) peak in Figure 4-22.

The underlying assumption behind using the ratio method to normalise the data, is the aromatic bonds are largely stable so will not change significantly between different regions. If changes were occurring in the concentration of aromatics then this could lead to skewed results elsewhere. In Figure 4-22, a number of relevant flattened maps are examined. It should be understood when viewing these maps there may be more room for error and variation due to surface topography and material structure which is not present in ratioed maps.

126

a) b) c) d)

-1 -1 -1 -1 1504 cm 1726 cm 1752 cm 1248 cm

Figure 4-22: AFM-IR flattened absorption maps for sample AFM(1)-A1(2). Absorption wavenumber used for each map are a) 1504 cm-1, b) 1726 cm-1, c) 1752 cm-1, d) 1248 cm-1.

Looking first at Figure 4-22a) (1504 cm-1) which is assigned to aromatic groups, there does appears to be an increase in concentration along one edge of the needle tip electrode suggesting they are formed during ageing, the mechanism by which this may occur is not understood at this point. This change will have knock-on effects upon the ratio images for other wavenumbers. Looking at the carbonyls in Figure 4-22b) and c) (1726 cm-1 and 1752 cm-1), there appears to be very little change in absorption noted across the region. Any variations observed do not appear to be more than noise and variations due to changes in surface topography (such as at the drop into the void where the needle was once held). This suggests that the normalised spectra and ratio images were misleading in identifying a drop in carbonyl bonds around the needle tip, however they yet again give the result of no carbonyl products having formed close to the needle tip, meaning this conclusion is unchanged. Degradation is again seen at other wavenumbers such as 1248 cm-1 confirming their presence.

While it is possible the lack of carbonyls observed is due to an almost total removal of oxygen from the system, this seems unlikely. Firstly it has not been observed in previous samples in this project and the large scale removal of oxygen from a system would be difficult even if tests were performed under vacuum. Secondly this sample was studied at a different location around the needle tip earlier as covered earlier in this chapter, there carbonyls were identified as having formed suggesting this effect was not total.

Alternative explanations include that a previously unobserved type of degradation occurred here (and oxidation did not), while oxidation might happen it is in reduced concentration, or that carbonyl products did form but have since reacted away. This could be due to the structure or the contact with the needle tip electrode, and that the by-products produced are able to escape to other locations around the electrode. It is hard to explain in a region in which much chemical degradation can be seen to have occurred and indeed where channel formation eventually occurred. On this evidence, the direct relationship assumed between the tree initiation and carbonyl formation, for

127 which there is not insignificant evidence [12, 13, 24, 34], cannot yet be considered as proven. Instead it is clearly evidenced that tree initiation can occur in the absence of carbonyls.

These results do not however show that channel formation occurred in a chemically inactive region. There is significant chemical activity identified here. With aromatics most notably being observed as forming, for which no explanation can be provided at present. Also forming in great numbers are believed to be C-O and C=C bonds, as seen in 1248 cm-1 and 1650 cm-1 maps respectively.

It is shown then tree initiation can occur in the absence of carbonyl groups, a long-established relationship. Care should however be taken in drawing definite conclusion on the nature of tree initiation based upon a single sample. This is a clear demonstration of the capability of the AFM-IR in this area and one which demands further study.

4.1.2.3. AFM(1)-I1- Initiated Tree – Approximately 30 µm in length

Sample AFM(1)-I1 contained a short tree of simple structure, after a voltage of 12 kVpk was applied for 10 minutes, followed by slow voltage increase until initiation occurred at 14.63 kVpk after a total of 19 minutes of ageing. The sample was then cut and polished down to expose the location of the needle electrode and the region around it was studied using the AFM-IR. This is the sample with the longest time of exposure to voltage following initiation, in which the degraded region is examined. It was also the subject of discussion in a conference paper for ICD in 2018 [104] (Appendix B-2 (Chapter 9)). With a more mature tree formed it can be seen how the degradation continues after tree initiation. It is found here this takes on a more familiar picture. Particularly regarding the shape of the degraded region which appears to take the spherical shape familiar from previous experiments reported in the literature [11, 13]. The spectra from the region immediately surrounding the channel are given below (Figure 4-24), represented in light green. Spectra from further out are given in dark green, and spectra from even further out are given in red.

The sample, including the channel formed from the electrode is shown optically in Figure 4-23 below, the dotted line is representative of the cross-section taken of the needle tip. This channel, measuring around 30 µm in length and having bifurcated multiple times, is not captured by the AFM- IR analysis.

128

10 µm

Figure 4-23: Optical image of Sample AFM(1)-I1. Dashed line represents approximate cross- section.

Figure 4-24: AFM-IR spectra for Sample AFM(1)-I1 corresponding to tip locations indicated by markers on the contact mode height image (right) of epoxy resin following the initiation of an electrical tree from a needle tip. Spectra are normalized to 1600 cm-1 peak. Approximate needle position shown by white dashed line.

129

Compared to previous samples the spectra closest to the needle tip show a particularly strong carbonyl absorption in the 1800 cm-1 to 1600 cm-1 region (this could also include C=C absorption). This absorption is not occurring at a single peak (previously we have seen 1745 cm-1 peaks) but fairly evenly across the carbonyl region. It also shows a distinct drop in the amplitude of absorption from the areas closest to the needle tip (light green), to those further out (dark green) and to those even further out (red). Other regions of the IR spectrum showed change in absorption (such as the ~1300 cm-1 to 1420 cm-1 region in which absorption is higher in the light green spectra, however this lacks clear individual peaks making it difficult to discern the types of degradation occurring.

The absorption maps (Figure 4-25) correlate well with the carbonyl regions of the spectra and allow an estimate of a size of the degraded region to be made. Here the region reaches approximately 8 µm from the needle tip and can be seen in Figure 4-25g) and h). The shape of the degraded region is also notably circular in this 2D image. This matches well with the methylene blue tests performed in the literature but is somewhat at odds with the degraded regions observed in samples in this thesis which have been far patchier in terms of where the degradation has occurred. It cannot be determined at this point whether this is sample dependent (due to the epoxy, the needle tip or the interface) or an effect of the increased ageing time.

a) b) c) d)

OH bonds C-O bonds Nitrates/nitrites

e) f) g) h)

Aromatic groups Alkene C=C bonds Ketone C=O bonds Ester C=O bonds

Figure 4-25: AFM-IR images of the surface around the needle tip in AFM(1)-I1: a) surface profile. IR Absorption amplitude at wavenumbers: b) 1056 cm-1, c) 1248 cm-1, d) 1360 cm-1, e) 1448 cm-1, f) 1640 cm-1, g) 1708 cm-1, h) 1726 cm-1. Each map is ratioed against 1500 cm-1 (corresponding to an aromatic peak). The images represent a 30 µm x 30 µm surface.

130

The absorption in other wavelengths is notable for its absence, compared with the spectra which appeared to indicate a localised increase in absorption at 1450 cm-1. There is very little change in absorption observed in the 1056 cm-1, 1248 cm-1, 1360 cm-1 and 1448 cm-1 maps.

131

4.2. AFM-IR Tree Growth Testing Results

This section will discuss the chemical signatures of trees grown in epoxy. It will look at the chemical signals occurring within and outside of channels at different distances from the needle tip. This section will demonstrate there are multiple chemical signatures which can be identified with trees. Specifically; one in which no chemical degradation was noted (Section 4.2.1), two in which degradation products were only identified within the channel (Section 4.2.2 and 4.2.3) and one in which degradation products were measured both inside and beyond the channel walls (Section 4.2.4).

4.2.1. AFM(1)-A1(2) – Earliest Stages of Tree Growth The first channel discussed in this thesis is the newly initiated channel from sample AFM(1)-A1(2) which has been discussed in the section on Tree initiation. This section will discuss this sample explicitly from the view of the formed channel. Previous analysis covered in detail in section 4.1.2.2 will not be repeated.

The channel is optically imaged again in Figure 4-26.

As discussed previously there was no chemical degradation observed within or around the channel, this fact supported the proposition that the channel does not form due to partial discharges. It is suspected the channel is instead a product of electromechanical fracturing. It is not yet clear however whether fracturing is a mechanism which drives further tree growth beyond the initiation stage.

Although there is little in the way of chemical signature from channel formation, we can still use the AFM-IR to study the channel. Using the AFM’s surface topography we are able to determine the channel is approximately 400 nm in width. This can be seen in Figure 4-27, in which both sections of the channels (measured before going below the surface) were measured at 377 nm.

10 µm

Figure 4-26: Sample AFM(1)-A1(2). A tree grown in epoxy in which the voltage was removed following initiation. Optically imaged (top and bottom-lit)

132

Figure 4-27: Surface profiles taken on the AFM-IR, the size of the channels is measured in two locations at which the channel is exposed on the surface.

With the measurement from Figure 4-27 we can suggest the tree size upon initiation is no more than 400 nm, allowing for sample-to-sample variation. It is possible though the channel widths immediately upon initiation were actually smaller and this channel was only identifiable after widening.

A second element of the channel which can be considered is that there are raised sections visible when the channel is immediately below the surface, these are highlighted in Figure 4-28 in which a profile analysis was performed. They are then given to scale in Figure 4-29.

133

Figure 4-28: a) The height profile from two lines across sample AFM(1)-A1(2). These lines are shown on the AFM-IR surface profile in b). Circled in both are regions above buried channels in which an increase in height is observed.

134

Figure 4-29: Surface profiles for the channels from Figure 4-28 in Sample AFM(1)-A1(2) given to scale.

In these cases there are clear increases in height, of around 100 nm and 75 nm, where the channel would be expected to be below the surface. This effect will also be later identified in Sample AFM(1)- G2 and a potential explanation provided in the discussion.

4.2.2. AFM(1)-G1 – 100 µm from the needle tip

A channel is now examined from sample AFM(1)-G1, which was subjected to 17.7 kVRMS to encourage initial tree formation and growth before the voltage was dropped to 8.85 kVRMS for further growth. The voltage was removed as the tree approached the planar electrode to prevent sample breakdown occurring. The tree length was approximately 1600 µm, with the section of the channel examined being around 100 µm from the needle tip. This channel was studied in a Conference paper presented in EIC 2019, “High Resolution Chemical Analysis of Electrical Trees through AFM-IR Spectroscopy”. Awaiting published and provided in Appendix B-3 (Chapter 9).

The sample AFM1-(G1) was polished down to cross-section the chosen channel, exposing a point at which two channels converge, forming a single channel. This section was chosen primarily for being a wider section (possibly due to the interaction of the channels combining), this was the first attempt to cross-section a channel, fully exposing the material within the channel.

Optical images are taken during the polishing and grinding process to track its progress. The following images show different aspects of the channel. The first, Figure 4-30, is both top-lit and bottom-lit to show the channel below the surface in relation to the void from the needle electrode.

135

100 µm

Figure 4-30: An optical image (front-lit) of Sample AFM1-G1.

Note the needle void seen in Figure 4-30 does not include the void from the tip of the electrode which has been polished away at this point, meaning the channel will have been closer to the needle electrode than it appears in this image.

The next optical images, given in Figure 4-31 show the exposed section of the channel in more detail. The first image shows the channel structure below the surface, the second image shows the channel reaching the surface and that it is exposed: the exposed area of the channel is circled in both images. The top-lit image also allows observation of the channel beneath the surface as a light, reflective material can be observed within the channels.

a) b)

50 µm 25 µm

Figure 4-31: A polished surface with an exposed tree channel: a) Optical Image (bottom-lit); sub- surface channel visible. b) Optical Image (top-lit); reflective material visible within channels below the surface. Circled are the locations at which the channel is exposed at the surface. [106]

136

Figure 4-32: Spectra for Sample AFM(1)-G1 from within the needle tip void. Inset: AFM-IR height profile showing the location of the spectra [106]

The spectra taken from sample AFM(1)-G1 are given in Figure 4-32. These are divided into three regions of degradation: Red spectra correspond to those taken closest to the exposed section of the channel, the blue spectra are those taken further from the channel and green spectra are from areas directly above a buried channel. The spectra indicate there are differences between these regions. The 1800 – 1620 cm-1 and 1150 – 1000 cm-1 spectral regions show increases in absorbance in the green spectra (above buried channels) compared to both the red spectra and blue spectra. In particular a broad peak is identified at 1740 cm-1. The 1800 – 1620 cm-1 region suggests the presence of C=O bonds. There is no evidence that this degradation is present outside of the channel walls.

Spectral maps taken from the exposed region of the channel are given in Figure 4-33. Here red indicates greater IR absorbance. Figure 4-33d) shows the 1726 cm-1 absorbance across the sample, which is indicative of the presence of carbonyl groups. This map shows increased density of carbonyl groups above the buried channel, although no such increase is identified within the exposed section of the channel. In all of the Figure 4-33 maps a lower absorption is identified within the exposed section of the channel; this could be due to the degradation products being removed during the polishing process or due to the AFM probe not being able to make effective contact within the channel void. A larger absorbance was noted above the buried channels at all wavenumbers within the carbonyl region (Figure 4-33c)).

137

a) b) c) d)

e) f) g)

Figure 4-33: AFM-IR images of the surface around the channel in AFM(1)-G1: a) surface profile. IR Absorption amplitude at wavenumbers: b) 1664 cm-1, c) 1702 cm-1, d) 1726 cm-1, e) 1248 cm-1, f) 1604 cm-1, g) 3300 cm-1. Each map is ratioed against 1504 cm-1 (corresponding to an aromatic peak). The images represent a 30 µm x 15 µm surface. [106]

15 µm

Figure 4-34: AFM(1)-G1, optically imaged in the background with the 1702 cm-1 chemical map overlaid. Red areas indicate greater absorption at this wavelength. [106]

Figure 4-34 shows an overlay of the 1702 cm-1 absorption map from Figure 4-33c) over an optical image of the channel identifying where the degradation products are present. This confirms they are contained within the unexposed sections of the channels and they have not formed or dispersed beyond the channel walls. The absorption map of 1664 cm-1 (Figure 4-33b) follows the same pattern, absorption being identified above the buried channels. 1664 cm-1 is tentatively assigned to C=C bond formation, which are understood to be common in electrical treeing tests in polyethylene, however imines (C=N) are also a possibility. Other wavenumbers follow similar, albeit weaker, trends (1604 cm-1 in Figure 4-33f) and 1248 cm-1 in Figure 4-33e)). 1604 cm-1 absorbance can be indicative of C=C bonds forming, which would match with the 1664 cm-1, as well as amines (N-H). It was not possible in this project to conclusively distinguish between these.

4.2.3. AFM(1)-G2 (Channel 1 – 100 µm from the needle tip) A channel is now examined from sample AFM(1)-G2, the preparation and study of this sample is very similar to that from AFM(1)-G1 above. The degradation observed within and around these channels is also very similar. This sample was subjected to an initial voltage of 17.7 kVRMS to encourage tree formation and early growth before the voltage was dropped to 8.85 kVRMS for further growth. The voltage was removed as the tree approached the planar electrode to prevent sample breakdown

138 occurring. The tree length was approximately 1600 µm, with the section of the channel examined being around 100 µm from the needle tip, again very similar to the AFM1-G1 channel discussed in the previous case.

Sample AFM(1)-G2 was cross-sectioned to fully expose a section of a channel, in this case a 20 µm length of the channel is exposed, which then bifurcates with one of the resultant channels situated above the cross-sectioned surface (meaning it has been polished away) and the other channel now below the surface. The optical imaging of the sample is shown in Figure 4-35.

Figure 4-35a) is a bottom-lit, optical image which shows the structure of the channel below the surface. The branched nature of the tree can be seen from this as well as the growth of the channel from the needle tip. The section of the tree which is following the bifurcation of the channel can also be seen here in the top right of the image. This appears to be a region in which a void or defect already exists based upon the shape and size of the channel here. Figure 4-35b) is a top-lit optical image, the exposed section of the channel. The extent of the exposed channel (~20 µm in length) can be seen clearly here, as can the bifurcation of the channel and the partially exposed section of the channel in the top right of this image. Again a light, reflective material can be seen from below the surface coming from within the unexposed sections of the channel.

The region around the channel is again spectrally analysed at a number of locations to understand what chemical products are forming in different locations.

b) a) 25 µm

100 µm

Figure 4-35: Sample AFM(1)-G2 with an exposed channel. a) Optical Image (bottom-lit); channel structure visible below surface with exposed section of channel circled. b) Optical Image (top-lit); channel exposed at the surface. [106]

139

Figure 4-36: Selected AFM-IR spectra of a channel and surrounding areas from Sample AFM(1)-G2 (normalized to 1504 cm-1 peak). Inset: AFM-IR height profile displaying location of spectra. [106]

In Figure 4-36 blue spectra are taken from points within the exposed section of the main channel, green spectra are taken from locations immediately outside of this section and red spectra are taken from locations even further away from the channel. Black and purple spectra are taken from taken from sections of the tree channel which are unexposed, below the surface, with black being more central within the channel/void region and purple spectra being from closer to the edge of the void.

The most striking response from the spectral analysis in Figure 4-36 is the large peak within the carbonyls (~1740 cm-1) for the purple and particularly the black spectra. This is tentatively assigned to the aldehyde group, however there are uncertainties about this interpretation as the aldehydes are a relatively unstable group so would not necessarily be expected to form in such quantities. Other groups can exist in this region but do not fit the spectral profile so cleanly, including carboxylic acids and esters. This result is striking in comparison to the previous sample from AFM1-G1 which showed a broader increase across the carbonyl region instead of a large peak. Another aspect which can noted from this is the similarity in red, blue and green spectra indicating that again within the exposed sections of the channel the degraded material has been lost and little extends into the bulk of the material. This is further examined using the spectral maps in Figure 4-37.

140

a) b) c)

d) e) f)

Figure 4-37: Sample AFM(1)-G2 a) Surface Profile. Absorption amplitude at wavelengths: b) 1604 cm-1 c) 1656 cm-1 d) 1702 cm-1 e) 1726 cm-1 f) 1742 cm-1. [106]

25 µm

Figure 4-38: Background: Optical Image of a tree channel with an overlaid AFM-IR absorption map from Sample AFM(1)-G2 (1742 cm-1) [106].

The spectral maps in Figure 4-37 clearly show a region of absorption on the right hand side for a range of wavenumbers. In Figure 4-38 one of these absorption maps, from Figure 4-37f) (1742 cm-1), is overlaid upon an optical image, this allows us to identify where the degradation has occurred around the channel. The increase in IR absorption measured here can be seen to correspond very closely to the regions containing buried, unexposed channels. This can be seen by comparison with Figure 4-35b) (optical image) and Figure 4-37a) (AFM surface profile). As with the AFM1-G1 sample, exposed sections of the channel show no increase in absorption at any wavenumber.

It can be seen that the absorption occurring within the exposed sections of the channel from sample AFM(1)-G2 is similar to the surrounding areas. This is actually in contrast to the previous example in which there was a small decrease in absorption within the exposed sections. Lower absorption within the channel would indicate either a chemical change (such as carbonisation) or that there is a poor contact between the AFM probe and the surface. In this case however the channel and the

141 surrounding area exhibit very similar IR responses to the extent that it is hard to distinguish the channel in the maps in Figure 4-37. This suggests the exposed channel and surrounding areas are chemically very similar. It is likely the material present within the unexposed section of channel is lost during the polishing and grinding procedure. The material remaining is that which is not damaged during tree growth. This channel allows us to identify this as it has a large width-depth ratio; this allows the AFM probe to make an effective and more consistent contact with the channel surface. This is in contrast to previous cases in which a reduced signal observed from within a channel can be explained by a poor contact with the AFM probe. The lower width-depth ratios prevent the probe from effectively accessing the channel wall surface. In future study this factor should be carefully considered before reaching conclusions (such as carbonisation having occurred on the channel walls).

Chemical analysis has indicated the chemistry from the unexposed section of the channel is similar to the unexposed section of the previous channel. Figure 4-37b) (1604 cm-1), shows an increase in absorption in this region and is assigned to C=C bonds forming within the channels, although they could instead be indicative of primary amine (N-H) bonds. Figure 4-37c) (1656 cm-1) is difficult to assign confidently and could be due to C=C, C=O or C=N (imine) bonds, which are all absorbent at that wavenumber. Aromatics, although absorbent at 1656 cm-1, have been discounted as the maps are normalised against an aromatic band (1504 cm-1) meaning variation in absorption seen is unlikely to be due to a change in numbers of aromatic groups. Large absorption changes are identified within the buried channel region in Figure 4-37d), e) and f) which are all likely to be due to carbonyl groups. These are consistently identified to form in mature tree channels. The 1742 cm-1 absorption map shows the most well-defined region of absorption and in considering the spectra from Figure 4-36 this matches the peak which was identified at this wavenumber. This peak is tentatively assigned to the aldehyde group, which best match this IR response.

4.2.4. AFM(1)-G2 (Channel 2 – 500 µm from the needle tip) The channel is taken from the same sample and tree as the previous channel (AFM1-G2). The section of the tree studied in this case was ~500 µm from the needle tip in a tree 1600 µm long. This is a channel which can be expected to have formed under different conditions to the previous cases (which have been formed 20 µm from the needle tip (in a young channel) or 100 µm from the needle tip (in a more mature channel)).

The locations of these channels within the full sample AFM(1)-G2 tree can be seen using the composite optical image in Figure 4-39. The previously studied channel is denoted as Channel 1 here, the second channel denoted as Channel 2.

142

Channel 2 – Bulk Chemical Degradation

500 µm Channel 1 – No Bulk Chemical Degradation

Figure 4-39: Composite optical images from sample AFM(1)-G2. Circled are the channels studied in this project.

Sample AFM(1)-G2 was prepared as with the previous samples by use of the polishing and grinding technique, however in this case the channel itself was not cross-sectioned. Here the bulk of the channel is left slightly below the surface; close enough for the degradation products to be observed using the AFM-IR but not fully exposing the channel and allowing material to escape. This is shown in the optical images in Figure 4-40.

a) b)

20 µm

Figure 4-40: A sub-surface channel from Sample AFM(1)-G2. a) Optical Image (top-lit); channel exposed at the surface. b) Optical Image (bottom-lit); channel structure visible below surface. Circled are points at the surface at which the channels have become exposed. [106]

143

Figure 4-40a) is a top-lit image, showing the surface of the epoxy, it can be seen in this image there is no fully exposed channel though circled are two small parts of the channel which have become exposed to the surface. Comparison with Figure 4-40b), a bottom-lit projection showing the tree structure beneath the surface shows these correspond to small channels which have branched off from the main channel. The top-lit image also shows a light, reflective material from within the channel.

The channel is approximately 3 µm in width and also has a very fine channel branching off from it. This channel is just visible in Figure 4-40b), branching downwards off the channel as we see it in this image, roughly in the centre. The channel width here is hard to measure from this image but certainly sub-micron.

There is a distinctive increase in the 1800-1600 cm-1 range of the spectra (Figure 4-41) for the markers on top of or close to the channel (red and blue) in comparison to the spectra taken from further out (green). This suggests carbonyl groups will have formed within the channels and that they may be distributed beyond the channel walls in this case. It is notable in the blue spectrum, taken from what is understood to be the centre of the channel and close to a branching point, that there is a significant peak at 1740 cm-1. This suggests a particular C=O group has formed in high levels in this location, potentially an aldehyde. Although, as discussed in previous samples this could indicate exotic forms of other carbonyl groups. There is also an increase in absorption noted around the channels (particularly in the blue spectrum) in the 1200-1000 cm-1 region.

Figure 4-41: Sample AFM(1)-G2 with a sub-surface channel. Selected AFM-IR spectra of channel and surrounding areas (normalized to 1504 cm-1 peak). Inset: 50 x 50 µm AFM-IR height profile displaying location of spectra. [106]

144

a) b) c) d)

e) f) g) h)

Figure 4-42: AFM-IR images of the surface of AFM(1) – G2 (channel 2). a) Surface profile. IR absorption amplitude at wavenumbers: b) 1248 cm-1, c) 1604 cm-1, d) 1640 cm-1, e) 1656 cm-1, f) 1708 cm-1, g) 1724 cm-1 h) 1742 cm-1. Each is ratioed against 1504 cm-1 (aromatic peak). The images represent a 30 µm x 30 µm surface [106]

Applying the spectral AFM-IR maps to region immediately above the channel we are able to see some interesting chemistry which is very distinct from those in the channels studied previously. The maps for a number of typically active wavelengths are given in Figure 4-42.

Again significant carbonyl formations are identified as forming due to the electrical tree channels with significant absorption occurring at 1708 cm-1 (Figure 4-42f)), 1724 cm-1 (Figure 4-42g)) and 1742 cm-1 (Figure 4-42h)). Absorption was also detected at 1640 cm-1 (Figure 4-42d)) and 1656 cm-1 (Figure 4-42e)); these wavenumbers are harder to assign and could indicate the presence of carbon- carbon double bonds or imines (C=N bonds). To identify where degradation has occurred images overlaying the AFM-IR maps upon the optical images are given in Figure 4-43.

a) b) c) d)

30 µm

Figure 4-43: AFM(1) – G2 (channel 2) background images show the channel structure below the epoxy surface. Overlaid images use the AFM-IR chemical mapping at different wavelengths and contrast settings. a) Optical Image b) 1742 cm-1 and set to show the extent of the carbonyls c) 1742 cm-1 and set to show the highest concentrations of carbonyls d) 1656 cm-1 and set to show the highest concentrations of alkene bonds. [106]

145

In these figures it can be seen degradation does not follow the same patterns as the previous cases. The AFM-IR maps in these images have been contrast-adjusted to extract different aspects of the chemical information. The map in Figure 4-43b) is adjusted to show the full extent to which carbonyls have formed/dissipated within the sample; while the maps in Figure 4-43c) and d) have been adjusted to isolate the areas of highest absorption.

In Figure 4-43b) we can see the absorption extends well beyond the channel walls, by as far as 5 µm. This differs greatly from the previous cases examined in which the degradation appeared to be strictly contained within the channels. This is considered in greater detail in the Discussion (Section 5.3.4). Figure 4-43c), contrast adjusted to show the areas of highest concentrations of carbonyl groups, shows these are at the branching points. This can be seen to occur at both branching points in this map. Figure 4-43d) shows a similar formation for 1656 cm-1 with the absorption increasing independently around both branching points. However, comparing the maps side-by-side, does suggest the carbonyls are more concentrated at the branching points than the groups which absorb at 1656 cm-1 (C=C or C=N groups).

Also notable in these maps is the small channel visible in the lower left of Figure 4-43b) and Figure 4-42a). The surface profile is shown again in Figure 4-44 with the channel circled. A profile analysis is performed in Figure 4-45 of both an unexposed and an exposed section of the channel.

Figure 4-44: Sample AFM1-G2 surface profile captured using AFM-IR. Circled is a small tree channel raised above the surface.

146

Figure 4-45: The height profile of two sections of a small channel from Sample AFM(1)-G2. The red line shows the profile across an unexposed section of channel, the blue line captures a partially exposed section of the channel. a) captures the location the profiles are taken from, b) the height profiles.

147

Figure 4-46: To scale height profiles from Sample AFM(1)-G2. Considers the channels examined in Figure 4-45.

These channels are shown to scale in Figure 4-46.

The small channel from AFM(1)-G2 is around 500-700 nm in width and, as can be seen in Figure 4-43b), carbonyls have formed within the channel suggesting partial discharges are able to occur within a channel this size at least to a distance of around 10 µm.

Importantly the channel is detected by the AFM probe, which here indicates the channel rises above the surface (it appears lighter in the map than the surrounding region, indicating it is raised). This means that in polishing the epoxy surface the channel has survived when the surrounding (undamaged) areas of the epoxy were polished away. This was previously observed in the ‘young’ channel in sample AFM(1)-A1(2). This channel is distinct in that the channel is observed over a significant length, with a 5 µm stretch of channel raised above the surface. Profile analysis suggests this rise is roughly 60 nm in height above the rest of the surface. The mechanism by which this is understood to occur is explained in the discussion.

4.3. Conclusion

This chapter has discussed the application of a novel chemical analysis technique, AFM-IR. This technique, with a resolution of 50 nm, has never been applied to electrical treeing before. This has covered a range of stages of degradation from pre-initiation samples (Section 4.1.1), post-initiation (Section 4.1.2) to sample with mature channels (4.2).

A polishing and grinding technique was used to expose the electrical trees in a controlled manner while also producing the smooth surface necessary for the AFM-IR to work effectively. There were some issues identified with this however, primarily that when the channels become exposed the chemically active material within is easily lost. This issue can be overcome by ensuring the channels being studied remain below the surface (as in Section 4.2.4) and alternative preparation techniques have been suggested.

148

The AFM-IR technique has allowed many striking findings to be made regarding the nature of electrical treeing. Possibly the most striking is that of the lack of any chemical degradation identified within the 20 µm channel in 4.1.2.2, capturing the earliest stages of tree growth. Other findings regarding the initiation of electrical treeing included that the degraded region around the needle tip is not homogenous; in fact multiple different chemical groupings were found to have formed independently and in different locations (Section 4.1.1.3).

For tree growth a significant difference was found in the nature of chemical degradation surrounding the channels depending upon the distances of the channels from the needle tip (Sections 4.2.2, 4.2.3, and 4.2.4). Also found was an increase in IR absorbance around the branching points in one of the samples at 500 µm from the needle tip (Section 4.2.4).

The impact and outcomes from this work are discussed in detail in the next chapter. The promising nature of these results strongly suggests AFM-IR be applied more widely in the study of electrical treeing. This should be done both to confirm the findings of this work and to the study of electrical treeing more generally, potential applications of this are discussed in the Future Work chapter (Chapter 7).

149

5. Discussion

This chapter is a discussion of the results obtained through this thesis; covering the plane-plane samples and what can be learnt from the testing performed in Chapter 2, and an in breakdown of the findings from the AFM-IR testing on tree initiation and tree growth (Chapter 4). A model for electrical tree growth, developed based upon the results in this thesis and work within the literature, is also proposed.

5.1. Plane-Plane Sample Configurations

This section will provide a discussion on the results and findings of the tests for electrical treeing using planar electrodes. It will briefly discuss the results as well as what these may indicate in terms of findings as well as how experiments could be performed in the future. Three experimental designs were tested; fine-tree pre-stressed samples, tracking samples and water saturated samples. Of these only water saturated samples showed any evidence of triggering electrical ageing or degradation in the absence of a metallic field enhancing electrode.

5.1.1. Sample Design Issues There were serious issues which affected the ability to monitor the samples both optically and through partial discharges. While some optical monitoring was possible, partial discharges were not used in result analysis due to noise issues. Resolving this issue would be a major aspect of any future work. It is likely this issue was due to vibrations between the top electrode and the sample, stemming from the amplifier. Attempts were made to stiffen the contact and prevent these vibrations however this could not be done sufficiently well. An alternative approach which could be taken in future tests could be the use of a wire connection, fixed to the top of the sample. This would be more flexible and may then absorb the vibrations. A conductive coating would still provide a good planar electrode and uniform field.

5.1.2. Water Saturated Samples Water saturated samples were found to be the most promising of the testing techniques, generating a localised breakdown within one of the layers. This may have been more effective than the other techniques as they created a significant field variation at a concentrated point, although the surface roughness between the interfaces may also have been impactful. In this respect they are quite different from the other tests which attempted to exploit a developed weak point but did not necessarily develop high fields or current flows within them. An initial conclusion, albeit a tentative one, from this work would be for planar designs, attempting to alter the physical conditions to

150 create localised electrical stresses, is more effective than attempting to exploit specific weak points which do not inherently affect the electrical field.

Continued ageing after the localised breakdown did not lead to tree initiation. Whether it was causing material ageing is not something which was determined however. Tree initiation will take time before it occurs and the issue with a test such as this is that there is no way of knowing if it was ageing the insulation or how long it would take for initiation to occur.

It is questionable whether evaporating the water within the sample was the correct decision. Though this has the effect of placing additional stress upon the layer of material being tested it could also limit discharges from occurring within the breakdown channel and so actually limit ageing. With effective partial discharge monitoring these aspects could be studied. It is also possible the breakdown channel was immediately back-filled with oil after forming – a consequence of growing trees under oil from a planar surface rather than an embedded needle. This is something which has been known to reduce reverse tree growth [20] and could prevent the discharges necessary for tree growth from occurring.

5.1.3. Tracking Samples Tracking sample tests showed no continued tree growth under voltage. The voltage was raised and the sample shortened to increase the field until flashover occurred through the interface. Following repeated attempts at this the technique was not considered any further to allow focus on other promising techniques. It is possible the lack of continued tree growth with this testing technique is due to the lack of initial electrons present in the high field regions. The interfacial tracks initially grown with the wire electrodes were filled with plasma and free electrons at high energies. It is not clear how these were then able to trigger tree growth into the bulk of the dielectrics, and whether this is more equivalent to needle tip initiation, tree branching or something entirely distinct. However without such a flow of electrons, or field enhancement generated by the wire electrode the trees were unable to grow. Further attempts to recreate this experiment may look to focus upon this aspect, achieving a flow of plasma within the tracks in a way that allows controlled tree initiation and growth from the interfacial tracks (or a similar feature).

5.1.4. Reverse Tree Samples In the reverse treeing tests there were a number of choices to make regarding the sample properties. Those which affected the monitoring, sample thickness for example, have been discussed previously. Another was in deciding whether or not to polish the sample down to expose more fine channels, not knowing whether this would increase or decrease the likelihood of initiation occurring.

151

On one hand having more trees exposed appears to give more chances for initiation to occur. However, it is not known why reverse trees form, and whether fine trees in their final stages could be fundamentally different from channels closer to the needle tip. Fine trees typically darken as they approach the ground conductor and often pause a short distance away before a tree will bridge the gap. Removing these ‘final stages’ of fine tree growth in the interests of exposing more fine trees may actually prove counter-productive in achieving reverse tree growth.

It may be that reverse trees, though they initiate from a plane electrode, require a source of charge in order to initiate. Fothergill [61] and Dissado [38] have previously discussed the effects of space charge within discharge free channels. If space charge accumulation can occur even without discharges then it is arguable this extends completely through the fine channels, potentially driving their growth, but crucially here being required for reverse tree initiation. An alternative effect which could be considered is that of the lubricant oil as discussed previously. If this were able to fill the fine channels and then was not removed by the water it would be expected to influence the tests, as would water were it to become trapped within the fine channels and become difficult to remove even with desiccation. It is likely either the oil or water would become trapped within the sample in these tests and as such these will affect their fields and breakdown properties. The precise impact of these is hard to determine due to the unclear nature of tree initiation growth with fine and reverse channels however in order to avoid this future sample designs should look to alternative sample preparation techniques to negate this issue.

5.1.5. Improved Sample Analysis Needed A significant issue with these accelerated planar testing techniques is we are unable to tell if initiation would have eventually occurred given sufficient time. It could simply be that more time was required, though this could mean another hour or another decade. Treeing from a needle-plane configuration electrode will typically be preceded by an incubation period; it is not an immediate occurrence and indeed in well-made samples may not occur at all without significantly higher than typical fields being applied. Tree breakdown within cables occur over decades. It is likely any of these techniques will also be accompanied by an associated lag time prior to tree initiation.

In order to design and refine a technique which does not include a metallic field enhancing feature, and which allows for tree initiation on suitable timeframes for lab-based testing (typically meaning hours to weeks at most) some indication of whether it is working is required before initiation. This can be achieved without immediate evidence of treeing, if there is evidence the technique being used is causing a measurable form of ageing.

152

If tree initiation cannot be found to have occurred using a technique, then an analysis for chemical changes with time should be considered as a first stage indication of accumulated damage. Such chemical ageing is found prior to tree initiation in needle-plane configurations [13]. Applying chemical analysis to the study of the initiation and growth of tree channels is a technique which has greatly aided the understanding of electrical treeing including FTIR and Raman spectroscopy [18, 53, 60]. A new technique with a high resolution and level of sensitivity has been applied to electrical treeing. The high resolution of this technique allows new insights to be gained into the process of electrical treeing building upon the work which has come before. Though not performed in this project it is envisioned that applied correctly it would be able to capture the early stages of pre- initiation degradation occurring in techniques such as these.

5.1.6. Application of AFM-IR to plane-plane electrode configurations The tests in this thesis have unfortunately not demonstrated a testing methodology which could replace the needle-plane electrode configuration nor provided clear evidence of the conditions which are necessary in cables to initiate and grow electrical trees. The water absorbent sample does however provide a potential insight into this, the localised breakdown occurring due to conditions which may well be replicable in cables exposed to water.

There are a number of issues with plane-plane samples. One of which is the difficulty in detecting aging prior to tree initiation in long-term testing. The AFM-IR technique has been proposed here as a tool capable of identifying the earliest signs of aging through finding signature changes in the chemistry of the dielectric. Such cases of pre-initiation degradation were have been discussed in this project. Sample AFM(1)-A1(2) (section 4.1.2.2.) raised questions on the link between chemical degradation and tree initiation, finding tree initiation does not require significant carbonyl formations. Whilst the possibility of initiation being unrelated to chemical damage should not be ignored, all available evidence is the two at least occur in the same regions, meaning chemical damage can be seen as an indicator of potential initiation sites.

It is necessary when using this technique to have a known area to study as the AFM-IR does not cover a wide area or volume, and to have a chemically clean region prior to testing. This means that of the three configurations tested only the water absorbent configuration is suitable for testing with the AFM-IR.

In the water absorbency samples, described in Section 2.3, we have a clear location which we would be able to study. The breakdown location is very clearly defined meaning we would know where to study. This would be most applicable in the area around and in front of the breakdown channel.

153

Although the sample was tested at high voltage for several hours, it did not initiate a tree. This does not mean ageing was not occurring, just that this was not sufficient to initiate a tree. However, without a chemical analysis there is no way for us to determine whether any ageing was occurring.

The reverse treeing samples would be more difficult to study. However if it is assumed reverse trees are most likely to form around fine trees then these areas could be studied for chemical ageing. The first step however should be to study fine and reverse trees grown in needle-plane configurations to identify what chemical ageing, if any, occurs prior to initiation in these cases. If it was not feasible to identify degradation using the AFM-IR then FTIR could be used to study a wider area for degradation. The FTIR due to its wider area of testing is not as sensitive as the AFM-IR, nor will it be as informative. It will though be useful for identifying chemical change when the precise location of this change is not known.

Tracking samples would be the most difficult to test chemically. The methodology used in this project involved the use of tracks which had already experienced significant levels of discharges and chemical damage. Identifying chemical degradation would not necessarily be indicative of new tree growth. It is likely in tests looking for early stage degradation (low level chemical changes) that tracking would not be suitable.

5.2. Discussion on Tree Initiation Findings

There is a lot to unpick from the results seen in Section 4; the chemical analysis of the degraded region in samples pre and post-initiation. This will be done in this section topic by topic. In each case an explanation of the effects observed will be given referring directly to the evidence. An analysis will be performed explaining what the conclusions are from this work as well as the confidence with which these can be drawn. Finally context will be provided for the field as a whole to explain where these findings fit within the wider picture and what their relevance is.

The topics covered in this section will be:

 The degradation processes occurring prior to initiation; based upon results it is believed multiple processes were active prior to initiation. A consideration of the degradation products observed and the chemistry of the epoxy are used to determine the reactions which may be responsible for them.  The patchy nature of degradation, far from the spheres often assumed and considered in papers [11, 24, 33, 34] we have seen that the degradation occurs in very localised areas. This

154

will impact upon our understanding for how pre-initiation degradation occurs around the needle tip.  The degradation necessary for tree initiation to occur. Carbonyl formation has often been taken as the indicator of degradation leading to tree initiation. However, here we have seen cases in which carbonyls appear largely absent initiate a tree indicating other conditions can lead to tree formation. The findings regarding this are discussed in depth.  The mechanisms responsible for tree initiation. The findings of this work strongly suggest electrical trees can form by non-chemical and non-discharge related mechanisms, and electromechanical fracturing is postulated here as an alternative mechanism. This is a finding with a potentially very significant impact upon our understanding of tree initiation and growth.

These topics will be covered here relating specifically to the findings of this section, a further discussion will occur at the end of the thesis to tie different aspects of the project and the literature together. This will include relating work done here to the planar electrode testing and the overall aim of better understanding how trees in power cables can be formed.

5.2.1. Degradation processes occurring prior to initiation A number of different chemical signatures have been identified. The first processes which will be focussed upon are those identified in sample AFM(2)-10 described here as Cases 1, 2 and 3 (Figure 4-10). Here these will be discussed as carbonyl group formation, etherification and methyl group formation respectively.

Looking first at carbonyl formation; that they are found to have formed is not a surprise. These have been consistently identified within polymers and specifically within electrical trees prior to and leading up to tree initiation. This has been done previously using methylene blue [13] and FTIR [11] on multiple occasions by different authors. The findings here support that carbonyls do in fact form when the polymer is subject to a high, divergent field. There was also no evidence to suggest the mechanisms proposed for carbonyl formation (via chain scission and alkoxy radicals [34, 60]) are not correct.

Methyl groups (-CH3) are believed to have formed prior to initiation, indeed prior to any other chemical products. These were identified in samples AFM(2)-5 and AFM(2)-10. In the first of these

-1 CH3 groups were only identified due to the 1450 cm peak. There is no strong evidence for how the

CH3 groups were able to form. This theory suggests the epoxy produces small pockets of hydrogen during curing and these react with the epoxy during the early stages of treeing. The hydrogen is

155 consumed preferentially to oxygen which prevents oxidation from occurring initially (AFM(2)-5). After the hydrogen has been used up, oxidation will then be able to occur and this is seen in AFM(2)- 10. Here methyl groups have still formed however carbonyl groups have begun to form, it is no surprise to see a significant overlap of areas for the methyl and carbonyl groups if they are related in origin.

The other group identified in AFM(2)-10 is believed to be ether (C-O-C), generating peaks around 1250 cm-1 and a general absorption increase from 1140 to 1010 cm-1. As discussed in the Results (Section 4.1.1.3), this increase in absorption could be explained in a number of ways and there is not enough evidence to be conclusive about how these formed. Looking at other samples however, and in particular the more mature, initiated samples we can see evidence that the level of ether formation changes on a sample-by-sample basis. Within sample AFM(1)-A1(2) when the tree is partially exposed, the spectra there largely show a relatively high level of absorption from 1140 to cm-1 to 1010 cm-1 (Section 4.1.2.2) which could be indicative of higher numbers of CO bonds relative to other bonds in these areas similar to AFM(2)-10. Compared to AFM(1)-I1 however, here there is a relatively low level of absorption in this spectral region. It appears from this to be dependent upon the sample or even exact location within samples. This would support the conclusion that this is the product of differences stemming from sample preparation. However this should not be taken as proven; there are many aspects which could induce changes in reactions from sample to sample and more evidence would need to be collected on this to determine when etherification does and does not occur. It is not clear yet how such a process would affect the time taken for tree initiation to occur or the overall dielectric strength of the epoxy.

We have found then that while carbonyls do form as expected, other processes are also believed to be ongoing which have not been widely considered by the electrical treeing community. Namely methyl groups, which form prior to carbonyls, and etherification which appears to occur but on a sample dependent basis. It is suspected methyl groups may not have been observed before as they appear to be a relatively weak effect, and occur in a limited area. Chemical analysis techniques which have lower resolving capabilities, like FTIR, may not be able to identify such a small volume of change, particularly when testing will focus upon more degraded, carbonyl rich samples. An alternative explanation is that their formation is very material dependent and while they have been identified in this epoxy they may not occur in other epoxies or polymers.

Understanding chemical degradation prior to initiation is very important for understanding how tree initiation occurs. Theories are developed and selected based upon precise chemical pathways and when and where reactions occur. If there are reactions which occur along with carbonyl formation,

156 whether these directly link to carbonyl formations or not, then these can impact the rate of initiation. This is particularly true when reactions are suspected to precede carbonyl formations.

5.2.2. Distribution of degradation around the needle tip The ‘degraded region’ of chemical degradation around the needle tip is often considered as a spherical damaged area around 10 µm in radius. However the evidence which we have found using the AFM-IR in this project is that this is not necessarily the case. While a roughly 10 µm limit for degradation appears largely reasonable as no degradation was found exceeding this range. The chemical damage within this area is actually very localised. In only one out of 6 samples was a uniform, circular area of damage identified. In each of the other samples the degradation was patchy, with damage occurring in particular areas and even different types of chemical change occurring in different locations.

The finding of very localised degradations does then raise the question of why previous tests have tended to find these to be spherical regions of damage and not also identified that the degradation can be localised. This is likely a combination of low resolution (FTIR and Raman spectroscopy) as well as low sensitivity (methylene blue) testing equipment, and a tendency to test heavily degraded samples. Many tests performed on this topic have used methods which actually increased the level of chemical damage allowed prior to tree initiation such as performing tests at low temperatures or in the absence of oxygen. The result of this is that the samples being tested were much more degraded. With the samples tested in this project, the sample with the longest exposure AFM(1)-I1 was also the sample in which degradation was most evenly distributed. It is likely increased exposure times produce more even distributions of chemical damage. The non-even distribution may have previously been observed in tests but not fully identified. In Shimizu’s tests [11] certain samples were noted as having a “rather vague” deteriorated region when tested with methylene blue. It is not clear what was meant by this however as these samples were aged in air (while use of vacuum produced a more distinct degraded region) it is likely these were effectively less degraded samples. “Rather vague” in this context could mean there was no distinct spherical region of degradation as was typical in other cases.

Patchy regions of chemical degradation are consistently observed in the samples and are entirely reasonable. We know polymers are never entirely uniform on the micron scale, and different regions will have different properties and experience degradation slightly differently. An example of this is between lamellae or spherulites in polyethylene [22]. Other factors include exact needle shapes and air gaps at the needle-polymer interface. The air gap is an aspect of electrical treeing testing which is under increased scrutiny with a number of tests demonstrating the effect of the air gap upon

157 initiation and tree growth [27]. It can be expected that any air gap or imperfections in the needle shape will also affect the distribution of the field and so charge injection and then chemical degradation.

Understanding how chemical degradation occurs around the needle tip is crucial to developing theories on the mechanisms responsible for tree initiation when it occurs. This is discussed in more depth in the next section. The distribution of chemical products here however also allows us to identify that multiple processes are ongoing within the degraded region and to distinguish between them. This takes us closer to understand what is actually happening prior to tree initiation. Previously theoretical work and calculations have been performed to determine between different mechanisms which could cause this chemical damage. However if these methods assume entirely spherical regions [33] in which a single reaction has occurred, it can now be seen that these are assumptions which cannot be relied upon at this point.

5.2.3. Degradation necessary for tree initiation to occur It is interesting that tree initiation does not always occur at the needle tip, it would be natural to assume initiation would occur at the highest field, the point of highest stress. Repeatedly however it has been observed that this is not the case and it can occur away from the needle tip. We can assume from this that the process is not necessarily as simple as high fields leading directly to tree initiation. There have been attempts [25] to explain why initiation does not always occur from the point of the needle tip (where the field should be highest).

The findings from this project may allow the variation in tree initiation location to be explained. Chemical degradation has been found to vary significantly around the needle tip and is commonly understood to directly lead to tree initiation. This variation is believed to be due to chemical and structural inhomogeneity during sample fabrication or uneven ageing. This section discusses the chemical degradation which is present in the locations in which a channel has been able to form.

AFM(1)-A1(2) is a sample in which the degradation has been studied around the area of the channel initiation. What is immediately apparent is that this is not an area in which significant numbers of carbonyl groups have formed. Whether any carbonyl groups had formed there is difficult to say conclusively, even though the sample was tested twice. In the first case (above the channel) some carbonyls had formed. This is shown in Figure 5-1 in both a map showing absorption at 1742 cm-1 and spectra from around the needle tip in which increases at 1742 cm-1 are noted in the region around which the channel initiated. However there is no reason to believe the carbonyls identified

158 were present around the channel. The carbonyls here were found to extend ~4 µm from the needle tip.

In the second case (immediately below the channel) there were no carbonyls identified in the area of initiation at all (Figure 5-2), meaning they either had not formed or were entirely reacted away. Though it should be noted this did not capture the locations closest to the needle tip with the tree only imaged starting at 2.5 µm from the tip. The tree was above the surface before this and so the chemistry is not captured using the AFM-IR.

The lack of absorption around the needle tip (as in Figure 5-2) means the carbonyls, if they were present, did not extend more than 2.5 µm from the needle tip at the point of initiation. If the process in which carbonyls were formed was responsible for tree initiation then this occurred only within this range. The 10 µm degraded region of carbonyls often cited is not necessary for tree initiation. Other degradation products are observed more prominently in this region 10 µm from the needle tip. These are seen in Figure 5-3 in which absorption at 1132 cm-1 and 1650 cm-1 is increased in a region extending 6 µm from the needle tip. Smaller increases were also noted at 1056 cm-1, 1248 cm-1 and 1448 cm-1.

Figure 5-1: a) 1742 cm-1 first testing of Sample AFM(1)-A1(2) and b) Spectra isolating the highest carbonyls

Figure 5-2: 1752 cm-1 second testing of Sample AFM(1)-A1(2)

159

a) b)

Figure 5-3: Sample AFM(1)-A1(2) IR absorbance at a) 1132 cm-1 and b) 1650 cm-1.

The range of wavenumbers at which absorption has been observed here does not isolate a single process occurring as being responsible for initiation. There are a number of chemical processes active here, based upon previous assignments (Etherification, oxidation, chain scission), and any of these, or a combination of these, could be responsible for the occurrence of the tree initiation in this case. Further case studies and a consideration of the energies necessary to induce the proposed reactions will be necessary to understand what the processes are which make the tree initiation more likely.

5.2.4. Mechanisms occurring during tree initiation Partial discharges are typically considered to be the mechanism by which tree growth occurs and they have been built into many theories explaining tree initiation [22, 34]. However the results from this work are incompatible with such a mechanism. Sample AFM(1)-2(1) looked at a young tree, with the voltage removed immediately after initiation. Looking within the channel found no evidence of chemical degradation. Partial discharges by their very nature (high energy electrons and UV emission) induce significant chemical degradation in polymers. Also commonly described as an electron avalanche, it is proposed they grow trees through breaking the bonds of the host material, effectively vaporising the material within the channels. This will of course be expected to leave an overwhelming chemical signature. It has been observed in papers in the literature [53, 60] and will be clearly observed in using the AFM-IR (in particular a channel of similar size in a mature tree is shown to reveal chemical damage in Sample AFM(1)-G2 (Figure 4-43). However here, with no chemical degradation identified within the channel it can only be assumed the process of channel formation is not related to discharges or to any chemically damaging process.

There are other mechanisms which have been proposed to provide the mechanism by which tree formation actually happens. These include Joule Heating [118] in which electrical current flows more heavily in more conductive channels, such as the amorphous regions between lamellae. The heat created through this would melt the surrounding material and form the gaseous electrical tree channels. However this too would be expected to leave a chemical signal. The percolation of deep

160 traps mechanism proposed by Wu and Dissado [119] similarly would induce chemical change when the channel is formed.

The opinion of this author is in the case of an epoxy in the glassy state the process which is responsible for tree initiation is electromechanical fracturing. Here the material surrounding the needle tip is subject to stresses and strains imposed by the electromechanical forces induced by the strong electric fields. When the material is at full mechanical strength it is able to withstand the applied forces. However the material is weakened over time [11], with the chemical degradation due to charge injection discussed previously. This degradation occurs all around the needle tip, although different areas may age in different ways and at different rates as seen in sample AFM(2)-10. At some point an area is sufficiently weakened that the Maxwell forces applied to the surrounding material are able to induce a crack to develop. Crack formation has been well discussed in the area of electrical treeing [61, 62, 64] as a mechanism by which stress on a system can be relieved. These can provide context for how fracturing may lead to channel formation and growth in electrical treeing.

It is notable the channel formed has typically grown beyond the degraded region. In Section 4.1.2.2 we identified chemical degradation around 6 µm from the needle tip void, but the channel grew to as far as 15 µm away from the needle tip void. This indicates that while chemical degradation may be necessary for the tree to form, the applied stresses are sufficient for the crack to propagate in non-degraded regions. This could suggest tree electromechanical fracturing may also be a potential mechanism in tree growth.

Electromechanical fracturing as a mechanism for tree initiation is able to explain many aspects of tree initiation which have been observed. Of course cracking of itself does not produce significant chemical degradation. It has been noted that trees when they form tend to immediately grow a distance away from the needle tip [120]. A tree of <5 µm in length does not tend to form, instead larger trees of even up to 20 µm are common immediately after initiation. The exact distances tend to be dependent on the voltage applied [120]. This can be explained by the use of a theory of electromechanical fracturing. Here a significant amount of stress is assumed to be necessary to form a channel and when a channel does form this will relieve some of this stress but not all of it, while producing a weak point vulnerable to further crack propagation. The immediate channel growth will then continue until the stress which has built up due to the electromechanical forces is reduced below that necessary for further channel formation. The distance would depend upon the applied stress and so the applied voltage as observed in practice.

161

Electromechanical fracturing would also be able to explain the tendency for tree growth to occur along weak points such as the amorphous regions between lamellae and spherulites in semi- crystalline polymers such as polyethylene. These will be mechanical weak points and the most likely areas for cracks to form. However fracturing does not require inherent inhomogeneity, meaning treeing can occur independent of polymer material. Similarly branching is explained; observed here (Sample AFM(1)-A1(2)), as branching is common in many fracturing effects. It typically occurs when larger stresses are applied, sufficient to induce multiple cracks to form. The stress relief is dependent upon the surface area released [64] during the fracture, explaining why more branching would be observed at higher voltages and stresses.

The epoxy used in this test was below the glass transition temperature, meaning it was in more of a glassy state than a rubbery state (the latter would be observed above the transition temperature). This makes the material less flexible and ultimately more brittle and subject to fracturing. Other materials including rubber state epoxy, LDPE and XLPE will need to be tested to see if they behave in the same way.

5.3. Discussion on Tree Growth Findings

As with the tree initiation discussion, this section will discuss a number of aspects of electrical tree growth which are raised from these results. These will each be discussed separately. These will be; smaller channels raised above the surface, when and why degradation is able to occur or disperse beyond the walls of the channel including the effect of distance from the needle tip on this, the removal of material from within exposed sections of channels and the chemical signature of branching points.

5.3.1. Raised Channels Two cases in which channels were raised above the surrounding surface were identified in this study (Samples AFM(1)-A1(2) and AFM(1)-G2). In both cases these were in smaller, sub-micron channels. The clearest of these is shown in Figure 5-4.

The channels themselves are not exposed in these cases. The top layer of the material of the channel remains intact despite the material surrounding it being polished away. This creates the appearance in which the channel appears to be raised above the surrounding material. This is believed to occur due to the channels being hollow or at least low density. When the surfaces were polished, the top layer was compressed downwards into the channel instead of being removed. This is illustrated in Figure 5-5.

162

a) b)

15 µm 15 µm

Figure 5-4: The channel from sample AFM(1)-G2. On the left a) shows the IR absorbance at 1742 cm-1, b) shows the AFM surface profile. Circled in each image is a small channel which rises above the rest of the surface

The compression effect was only identified in the smaller, sub-micron channels. It appears that in larger channels this does not occur, or at least not as easily. It is possible that this is due to the size of the layers meaning that rather than compressing it will just fall apart. An alternative however could be that the outer layers are fundamentally different in larger channels, and less mechanically strong.

Figure 5-5: Schematic illustrating the compression of channels at the surface during polishing

163

Finding channels can be compressed during polishing provides further evidence for channels being hollow or at least of significantly reduced density; although this was not in significant doubt after SEM imaging and because it was believed to be necessary to allow discharges to occur down their length. But it raises the question of where does the material, which once filled this space, go. The material which occupied the space in which these channels form must go somewhere. Logically the material can either be within the channel, outside of the channels or distributed further down the length of the channel.

5.3.2. Exposure of channels during polishing We have seen evidence that some of the degraded material produced during tree growth remains within the channel. When channels are exposed both the white optical reflection (Figure 4-31b)) and IR absorption from within those channels are lost. This is likely due to material from within the channels being lost. This could happen either during the polishing process or through them being water soluble and then dissolved and washed away. Material loss in exposed channels was observed in both samples AFM(1)-G1 and AFM(1)-G2 (Channel 1). It is not yet clear where this material is situated, whether it is distributed around the walls of the channels or is more evenly distributed throughout it. There is no significant finding either way regarding this in this project. However this material cannot account for all the material removed in order to form the channel.

In the case of sample AFM(1)-G2 (Channel 2) a lot of material is observed in the bulk material outside of the channel. This may explain where the material which is removed during tree growth has gone. The material is able or is forced to leave the channel and then becomes trapped in the surrounding material. This is supported by the similarity in chemistry which was found between the material within and beyond the channel walls.

It is however more difficult to explain the absence of the material in the cases of AFM(1)-G2 (Channel 1) and AFM(1)-G1. In these cases, while some chemically degraded material is observed within the channels it is not observed at all outside of the channels. The question remains, where did these material go?

This author proposes it is some combination of; electromechanical fracturing (which can create space within channels without removing any material), reactions causing gases and other more volatile products to be released (and so reducing the amount of material within the channel) and the material being compressed by the discharges as it is damaged meaning it takes up less space.

For future use of the AFM-IR technique in electrical treeing it is important to consider the effect of material loss during sample preparation. If possible, alternative techniques besides polishing could

164 be applied; microtoming or cyrofracturing are potential options. Alternatively, not exposing the channel (as done with AFM(1)-G2 (Channel 2)) allows the chemistry within the sub-surface channels to be observed. This uses the fact that the AFM-IR detects absorption from below the surface, provided it is sufficiently close to the surface.

5.3.3. Branching points Why and how branching occurs remains a significant question in understanding electrical treeing, and no model purporting to describe tree growth could be seen as complete without an explanation for this. Studying branching points was seen as a very important aspect of the study of trees. Specifically, with the high resolutions provided by the AFM-IR technique, the effects of the branching points could be differentiated to an extent from that of the connecting channels.

Three branching points were studied in this project. They each came from sample AFM(1)-G2; with one exposed within the exposed section of channel in G2 (Channel 1) and two unexposed branching points in G2 (Channel 2).

 The exposed branching point came from the channel closest to the needle tip (~100 µm away) and showed no evidence of chemical degradation. While this would be expected given exposed channels consistently showed no degradation, it is also notable there was no degradation observed outside of the channel, in the surrounding bulk material, either.  Looking at the channels captured in AFM(1)-G2 (Channel 2) it has been found the degradation is concentrated around the channels. Increases in absorption are observed at both branching points. This is particularly clear in Figure 4-43. This degradation is not just limited to within the channel but is also observed to extend into the bulk outside of the channel. It is unclear, however, the extent to which degradation observed in this sample comes from the branching points or comes from the connecting channels more generally.

We appear then to have two distinct chemical signatures from branching points. One in which chemical degradation has appeared well beyond the channel walls (in samples further from the needle tip) and a second in which no degradation products are identified beyond the channel walls (in samples closer to the needle tip). This does however require further work to find more evidence to confirm this is typical behaviour.

There are a number of different mechanisms by which the increases in degradation at the branching points in sample AFM(1)-G2 (Channel 2) can be explained.

165

 Through an accumulation of degradation particles which are carried down the channel during discharges (meaning the degradation does not necessarily form there, just collects there). This theory is unlikely to explain the increase in IR absorption as it is found to cause increases beyond the channel walls when this would be expected to limit to within the channel.  The new branches form and then the branching triggers an increase in damage occurring in that location. Degradation occurs through discharges, a flow of plasma through the channels, hitting the channel walls. At branching points the flow will be split between the two possible paths, and presumably some will be more likely to hit the walls between the two paths causing damage to the surrounding material.  There is an increase in degradation in certain regions, for one reason or another, and this then makes branch formation more likely in those locations. o This could be an effect created by the discharges forming within the channels. This would be quite a logical conclusion to reach given that bifurcation and general tree structure have been noted to be dependent upon the applied voltage. More branching is found to occur at higher fields [19] and tree structure is changed by applied frequencies [19]. It could be then the discharges naturally produce regions which experience greater levels of damage, depending upon the applied fields, and these regions then bifurcate more readily. o Alternatively it could be a property of the material, if it was mechanically stronger it would be expected to slow channel growth as the channel reached it. This would then experience more discharges striking the channel walls until further tree growth could occur (causing more chemical damage as observed using the AFM-IR). This weakened area could then have many areas weakened sufficiently to produce more branches emanating from them.

The chemical analysis of branching points presents an opportunity to understand the mechanisms of branching and the physical conditions in which they occur. The findings presented here do suggest however that there are significant differences between branching points and the connected channels, with chemical degradation appearing to be concentrated at the branching points in sample AFM(1)-G2 (Channel 2). Whether this is also true in channels closer to the needle tip will require further investigation. It is believed with sufficient study using the AFM-IR in combination with other techniques it may be possible to identify the physical mechanisms behind branching.

166

5.3.4. Degradation products beyond the channel walls The AFM-IR has allowed us to observe the difference in chemistry between areas within and areas beyond the channel walls. This produced two distinct cases in which the behaviour was very different.

In channels closer to the needle tip (~100 µm) there appears to be a clearly defined difference between the channels (containing chemically damaged material) and the bulk material (which shows no degraded material). Here there is a sharp boundary at the channel walls beyond which there is no change in absorption. This is observed both in the AFM-IR maps of unexposed section, which show a sharp difference in absorbance between treed and non-treed regions. And in the exposed sections of channels which show no absorbance whatsoever either beyond the channel walls or upon their surface. There is no evidence of a transition or interface region between them in which the material is only partially damaged. This is reminiscent of surface discharge ageing, in which the surface is severely damaged however degradation does not extend far beyond the surface layer [121].

In the channel further from the needle tip (~500 µm), the degradation products are observed well beyond the channel walls extending as far as 5 µm from the channel. This material was identified to be chemically very similar in nature to the material within the channels. In fact using the spectral maps the channel is indistinguishable from the surrounding material. The likely explanation for this is that the material is produced within the channel walls. The material is then able to (or is made to) diffuse from the channel and enter the surrounding bulk in which it becomes trapped. It was notable in this case the degradation appeared to be concentrated at the branching points. It is not clear yet what exactly this means when comparing degradation signatures at channels different distances from the needle tip.

It should be noted the channel walls in all cases are clearly defined in optical imaging suggesting the mechanism(s) which defines the shape and structure of the channel is unaffected by this difference in chemical signature.

If it is taken that the chemical degradation observed outside of the channel in AFM(1)-G2 (Channel 2) is formed within the channel then a question which should be considered is why does this occurs here and not in other cases? It is likely material formed inside of the channel requires heat, pressure and time in order to be able to escape the channel and diffuse through the surrounding material. This may indicate a fundamental difference in the growth of channels at different distances from the needle tip. With channels formed closest to the needle tip not being sufficiently hot for product diffusion to occur at all. This could be due to channel growth not being a ‘hot’ process or through it

167 simply not reaching a necessary threshold. Within the channel further from the needle tip it remains unclear whether the process by which chemical products were able to diffuse was strongest at the branching points or if it occurred entirely because of the branching points.

The fact that material does not escape at all in the closer channels suggests the degraded material does not want to leave the channel in every case. It seems to require certain conditions to be met in order for this to occur; these could be that when there is ‘too much’ material produced the pressure effectively forces some of it to leave. Alternatively it could be with heat, time and energy they are able to leave. Each of these ideas implies different active processes and explanations for what is occurring. A model which could explain these results and others identified in literature is described in the next section.

5.4. Proposed Model of Tree Growth

Tests performed by Vaughan et al [53] used Raman spectroscopy to study electrical trees formed in polyethylene. Raman spectroscopy looks at the photon absorption and emission of photons from a material as illustrated in Figure 5-6.

Figure 5-6: Raman Scattering Illustration.

168

Crucially the molecule, when it absorbs the photon, becomes more energetic and enters a ‘virtual’ energy level, which is an extremely short lived state after which it will emit a photon and the molecule drops in energy. Typically the photons absorbed and emitted will be the same energy, this is Rayleigh scattering. Occasionally however a photon will be emitted at a different energy, giving a frequency difference, or Raman shift, between the absorbed and emitted photons. This is shown in the Stokes and anti-Stokes Raman scattering. These shifts are material dependent and allow the identification of the composition of a sample. Another phenomenon which typically occurs alongside Raman scattering is fluorescence. This involves the absorption to and from excited vibrational energy levels. This can produce a far stronger response than the Raman shifts and appears as a large background signal when compared to the Raman shifts. Examples of both can be seen in Figure 5-7. The small peaks in each line are Raman shifts while the broad background (strongest for 125 µm) is the fluorescence signal.

Vaughan et al [53] identified both Raman shifts and a strong fluorescent signal associated with the trees. The fluorescence was understood to form due to the melting of the polymer surrounding the channels and in this respect is similar to the heating which was identified here as controlling the diffusion of degradation products. Taken from [53] is Figure 5-7, showing the Raman and fluorescence responses around channels at different distances from the needle tip in a non- conducting tree (this is distinctive from the responses identified in conducting trees).

Figure 5-7: Raman spectra taken from channel cores of a non-conducting tree at different distance from the needle electrode. Taken from [53]

169

Here the fluorescence is given by the large peaks which hit a maximum at around 500 cm-1. Here we see there is no melting around 50 µm but that at 125 µm from the needle tip the fluorescence increases dramatically. The fluorescence, and therefore level of melting, then decreases with distance from the needle tip after this. It is proposed this melting and the observation of chemical products outside of the channels using AFM-IR are inherently linked. As indeed is the lack of them observed near channels closer to the needle tip. Early stage tree growth and later stage tree growth appear to have fundamental differences.

Also believed to be related is Sample AFM(1)-A1(2), in which a channel of ~20 µm length was found to not contain any chemically damaged material. This was proposed as evidencing tha tree initiation was a process driven by mechanical fracturing instead of partial discharges. This model on tree growth suggests that early stages (and possibly later stages) of tree growth are also driven by mechanical fracturing. This would then be true for both epoxy trees and non-conducting polyethylene trees.

Early stage tree growth is believed to occur first through electromechanical fracturing which creates an open channel in which discharges are then able to occur. The discharges damage the surrounding material, widening the channels and causing the observed chemical degradation. However the growth driven by mechanical fracturing precedes the discharges. This means the discharges are not sufficiently concentrated to produce the temperatures necessary to melt the surrounding material and to cause the degradation products to diffuse into the bulk.

Channels further from the needle tip in non-conducting trees experience lower fields, these slow the mechanical fracturing growth. At some point a threshold is reached (between 50 µm and 125 µm in Vaughan’s results and after 100 µm in this project) at which the growth by mechanical fracturing is sufficiently slowed for the discharges (which cause high energies and temperatures) to become more active in tree growth. These discharges are now able to be concentrated against the ends of the channel, depositing a lot of energy in these locations in a short space of time. This produces high temperatures which then cause the sudden shift in melting and material diffusion noted above.

As we then move further from the needle tip the discharges have further to travel through the non- conducting channels. This will increase the magnitude of discharges which reach the ends of the channels but crucially decreases the rate at which they will reach due to the larger extent of the tree and number of branches. The rate of energy being transferred to the channels will decrease with distance giving the results in Figure 5-7 of a fluorescence which decreases with distance from the

170 needle tip. The concept of there being different stages of tree growth is not new, and there is strong evidence for multi-stage tree growth in epoxy [55].

The proposed model of tree growth is illustrated in Figure 5-8 discussing the different stages of growth and how these affect the surrounding material.

 The first stage captures the moment of tree formation, which in an epoxy sample at least, has been shown to not be a phenomenon which produces chemical damage instead being a mechanically driven process. No partial discharges are believed to occur at the moment of initiation however charge injection precedes the formation of a channel. This charge injection chemically damages the surrounding material (creating a degraded region), mechanically weakening it until fracturing is able to occur, causing tree formation.  The second stage then covers the early stages of tree growth. This occurs more rapidly and is driven by electromechanical fracturing. Partial discharges also occur within the channels once established widening them and causing the chemical degradation all of which is concentrated within the channels.  The third stage is the point at which electromechanical fracturing has sufficiently slowed for partial discharges to now reach the ends of the channels for extended periods of time. These discharges create significant heat in the region as well as a many chemical by-products. With the heat and the pressure of the products in this region for an extended time the material is able to spread into the surrounding regions. It may be that the discharges then drive tree growth, or that fracturing remains the dominant mechanism but discharges have a role in weakening the material enabling the electromechanical fracturing to continue.  The fourth stage is a continuation of the third stage, however further from the needle tip the discharges are dispersed within a greater tree volume which reduces the energies being deposited per unit volume in the region. Melting effects surrounding the channel are then reduced as observed in Figure 5-7 [53]. These are not believed to contribute to channel growth but occur as a consequence of it.

171

Figure 5-8: Illustrated model showing different stages of proposed electromechanical fracturing driven tree growth along with partial discharge and chemical degradation.

This model suggests that in the early stages of tree growth electrical treeing is fundamentally a mechanical fracturing driven event. This fits the chemical evidence found using Raman spectroscopy and AFM-IR as well as other findings. Including:

 Optical and PD evidence of fine trees [20, 55] as well as other filamentary channels [18, 38] for which no corresponding partial discharge has been observed.  The commonly observed tree initiation length of channels [22, 23, 118, 122] being between approximately 5 µm and 20 µm. As crack formation will occur across a length depending upon the stress which has built within that region.  The common findings of channels growing longer and then later widening or carbonising [18, 55]. This would be expected when the growth in length is driven by fracturing while widening follows due to partial discharges.  The slow-down in tree growth after early stages of tree growth [3, 123]. This occurs as the field drops with distance from the needle tip and suggests further investigation into tree growth without extremely sharp electrodes is necessary.  The effect of mechanical stresses [10, 63, 124] and mechanical properties [61, 125] on tree initiation and growth.

As discussed before (Section 1.4.5), a number of models have been developed of tree growth through electromechanical fracturing [61, 62, 64, 70] discussing how channels would be able to propagate in a polymer. No attempt is made to distinguish between these models, only to contextualise the experimental evidence for such a mechanism to be active.

172

6. Conclusion

This section considers the key developments and findings from this project.

 A localised breakdown was produced within a sample with uniform electrodes (Section 2.3). In this sample controlled variation in water saturation levels had produced changes in permittivity (similar in nature to water treeing) which changed the field experienced in different locations. These change in permittivity, possibly along with some surface roughness effects are believed to have created sufficient field enhancement to trigger a localised breakdown. This suggests alternative techniques to produce electrical trees may benefit from ways of altering the fields to develop localised field enhancements. Similar attempts to trigger electrical tree formation or growth were unsuccessful in the use of fine trees (Section 2.1) and surface tracking (Section 2.2). These were not believed to create sufficient field enhancements at any point to trigger the discharges necessary for tree growth.  The Atomic Force Microscopy with Infrared spectroscopy (AFM-IR) technique was applied to electrical trees (Chapter 3). This allowed chemical analysis of tree channels on scales below the channel width for the first time. A polishing and grinding technique, in combination with optical microscopy was utilised for sample preparation with a demonstrated effectiveness (Section 3.2.2). The AFM-IR has a spatial resolution of approximately 50 nm and allows a full chemical characterisation to be performed, greatly outperforming similar techniques such as Raman spectroscopy and FT-IR.  Pre-initiation chemical degradation was studied over a number of samples, showing the degradation is less homogeneous than has been observed in previous tests which had implied a relatively uniform sphere of degradation (discussed in Section 5.2.2). This has implications when considering the mechanisms which cause the pre-initiation degradation as well as triggering tree initiation. It must also be considered in discussions on the stochasticity of times taken for tree initiation as well as the location of tree initiations on the needle tip. This suggests the variation of time and location is likely to be a product of the variation in chemical degradation around the needle tip.  Pre-initiation degradation was shown to result in multiple chemical products besides those found in previous literature (discussed in Section 5.2.1), indicating different chemical pathways are active. This could be the result of the early stages of degradation only being captured due to the high sensitivity of the AFM-IR technique. The most striking result of this was that in the region of tree initiation for sample AFM(1)-A1(2) (Section 4.1.2.2) carbonyl (C=O) groups were not found to have formed. Their formation has often taken to be the strongest indication of pre-

173

initiation degradation. Other groups were found to form before and alongside (but independently of these). This should be further investigated to ensure this is consistently found and to test whether this is consistent across materials.  A 20 µm long channel was chemically analysed, having had its voltage removed immediately following initiation. This was found not to contain any evidence of chemical degradation. The lack of any evidence of chemical degradation strongly suggests partial discharges did not cause the channel formation given that these are believed to trigger channel growth through material vaporisation, which would leave a chemical signature. The initiation process in this case is believed to have been due to electromechanical fracturing (discussed in Section 5.2.4). This is a process for tree initiation and growth which would not induce significant chemical degradation and has often been proposed and theorised, yet lacked clear direct evidence.  Mature channels were studied showing changes in the distribution of chemical products with distance. At 100 µm from the needle tip degradation products were entirely contained within the channel walls, as observed in samples AFM(1)-G1 (Section 4.2.2) and AFM(1)-G2 (Section 4.2.3) whereas at 500 µm (Section 4.2.4) degradation products were found within the surrounding bulk extending as far as 5 µm from the tree channel. The similar chemical nature of the materials within and outside of the channel leads to the conclusion this material formed within the channel and dissipated outwards into the bulk.  A model of tree growth was developed using these results and results from the literature applicable to trees grown in glassy state epoxy and non-conducting channels in polyethylene. This model is discussed in Section 5.4. This model proposes channel growth is initially driven by electromechanical fracturing, this occurs closest to the needle tip where fields are strongest. This coincides with the periods of fastest tree growth often identified in treeing tests. Partial discharges are then able to occur within the channels left by the mechanical fracturing and chemically change the material. After a certain distance (around 100 µm) the growth driven by fracturing is slowed as the field drops with distance from the needle electrode. As the growth due to electromechanical fracturing is slowed, the discharges are able to ‘catch up’ with the channel growth and become involved in further tree growth causing chemical changes in the channels and surrounding material..  Degradation products were found to be most concentrated at two different branching points in sample AFM(1)-G2 (Section 4.2.4) in the channel 500 µm from the needle tip. The cause of this is not entirely clear, and nor is whether it is a consequence of distance from the needle tip or due to variations in material properties. It needs to be determined whether this effect is consistent across different samples and different distances from the needle tip.

174

7. Future Work

This project has largely looked to develop new methods for the study and analysis of electrical treeing. This began looking at alternative experimental configurations which did not require well- defined sharp points, and then looked at using a state-of-the-art chemical analysis technique (AFM- IR) to study localised chemistry of electrical tree initiation and growth for the first time on the nanometre scale. The successful application of this new technique to studying electrical treeing, along with the results and models developed using these methods, demonstrates there is a great deal of further work which could be done in these areas. Discussed in this section are:

 The continued adaption of planar samples to build upon the water absorbent sample finding as well as further configurations and tests for which there was not enough time to test. The AFM-IR could then be used to study the samples tested in these ways to identify if ageing is occurring.  Further AFM-IR tests looking at gaining a fuller picture of tree chemistry including using different materials such as LDPE and XLPE as well as trees grown in cables in service to see how these compare chemically.  An improved sample preparation technique for the AFM-IR which could allow channels to be more precisely approached, giving additional control in the choice of where to test.  The inclusion of electroluminescence tracking equipment in the study of pre-initiation chemical degradation. Allowing a detailed comparison between the best techniques for real- time studying of pre-initiation degradation and the most detailed analysis technique we have to study the ongoing chemical degradation.

7.1. Planar Sample Experimental Configurations

Understanding planar samples remains an area of electrical treeing in need of further investigation. Needle electrodes, while effective at growing electrical trees, are wholly unrepresentative of electrode configurations in cables. Developing alternative mechanisms to test these is an ongoing study.

7.1.1. Reverse Tree Testing Reverse tree testing did not lead to any evidence of continued tree growth or reverse tree initiation. However the nature of tree initiation in non-planar samples is that it would be expected to be a slow process. Even using extremely sharp and precise needle tips does not always lead to immediate tree initiation. Further tests looking to use such a technique should then account for this and be prepared

175 to utilise techniques for identifying pre-initiation ageing. The most obvious of these is a chemical analysis; this is something which has been developed in this project.

There are issues which would be associated with the application of AFM-IR to planar samples however. Even if small amounts of chemical degradation were occurring, identifying where and when these occurred would be a difficult task. The AFM-IR is a destructive technique and one which requires being located at the precise location of the degradation. Planar samples by their nature cover a significant area and volume and so probable locations for pre-initiation degradation would have to be located in advance. A technique which may help with this is FT-IR, which utilises IR absorbance in the same way as AFM-IR however does so with a lower resolution but wider field of view. The micro-FTIR version of this technique may be then able to identify locations with increases in chemical degradation.

A second issue associated with utilising such a prognostic technique is that it would be difficult to correlate chemical degradation identified using these techniques with future treeing events. While using the AFM-IR may allow more detailed aspects of the chemical degradation to be observed these tests are by their nature very distinctive from tests performed at needle tips. It would be difficult to know whether a type of chemical degradation observed was actually indicative of the early stages of tree initiation.

The most effective method which could be applied to understand reverse treeing using the AFM-IR, is believed to first chemically study reverse trees formed in typical samples. This does have the negative issues of there still being a needle tip in the system, though its impact is unclear, and a tree forming, which would be expected to remove and complicate much of the chemical information in the area. However as a first look at this chemically, understanding what is happening in the region is crucial to further study of the effect.

7.1.2. Water Absorbent Samples As a planar sample which appeared to produce a localised breakdown channel (Section 2.3.4) this sample configuration requires further testing to see under what circumstances this can be repeated and to see whether it can be controlled to produce electrical tree growth from such channels. Again a chemical analysis should be applied, in this configuration. There is however an obvious location in which to perform the analysis. The area within, around and ahead of the breakdown channels should be studied using the AFM-IR to look for evidence of further chemical degradation which can be correlated with pre-tree degradation. If this technique is viable for the growth of electrical trees

176 then long-term testing should be considered to understand how and why tree growth occurs in polymers with planar electrodes.

7.1.3. Alternative Sample Designs Although a number of different sample designs were tested in this project there remain others which could be tested to see whether tree growth can occur under different configurations, and if so what affect these have on tree structure and growth.

Semicon layers are an aspect of cables which have been accepted as improving the overall reliability of cable designs. Being extruded with the insulation ensures smooth and defect free interfaces between these layers, reducing the prospect of electrical trees forming in these regions. However the semicon layer itself is not a perfectly uniform material, instead being composed of a polymer embedded with carbon particles to increase the bulk conductivity of the material. There has been minimal work however to consider the effect of individual or small groups of carbon particles upon the field experienced by the insulation. We know from testing that a 3 µm radius steel needle tip at

10 kVRMS will quickly initiate electrical trees. Whether a group of 10 µm diameter carbon particle in the semicon layers can also initiate trees at the higher voltages experienced in practice is something which should be investigated further.

Another sample design which bears consideration is the PNP (plane-needle-plane) configuration shown in Figure 7-1.

Figure 7-1: Plane-needle-plane electrode configuration. Adapted from [126]

177

In the plane-needle-plane design, which has been used in liquid dielectric testing [126], two planar electrodes are used however one of these has a small needle electrode protruding ahead of it. This has the effect of altering the overall field, making it far more uniform than that of a single needle electrode. This is more representative of the types of fields which would be expected in cables. It was shown to have significant impact upon the shape of streamers in liquid dielectrics, making them branch far less often than a single needle electrode [126]. In electrical treeing it is likely such a design change, making treeing tests more representative of in service cables, will have significant impact upon the structure and potentially the fundamental growth mechanisms of electrical trees.

7.2. AFM-IR

The AFM-IR has been shown to work in practice in studying electrical treeing. The use of this powerful analysis technique has allowed the creation of detailed maps of the chemical degradation associated with tree initiation and with mature channels. Given the power of this technique to provide unique insights into chemical ageing it is a technique with many applications in the field of electrical treeing. First and foremost testing should be continued upon standard trees as done in this project, which is based upon a small number of tested channels. This will build a more comprehensive and thoroughly tested body of evidence on which to discuss the chemical nature of different stages of electrical treeing. Beyond this however there are many more particular areas in which the AFM-IR can be developed and applied as discussed here.

7.2.1. Sample Preparation The sample preparation technique used in this project, grinding and polishing, has been effective yet also shown to have significant drawbacks in terms of material loss on exposing channels. It is also a time consuming process which can also lead to over grinding rendering samples unusable. Finding an alternative method would therefore be an initial step worth considering in further use of the AFM- IR.

Alternative methods were previously discussed in the methodology section for the AFM-IR and their reasons for not being taken were discussed as well. The method which appears to have the most applicability in this case is the use of serial block-face scanning electron microscopy (SBFSEM) which involves microtoming away individual layers of the sample and then SEM scanning the top layer of the sample. Doing this repeatedly allows you to build a high-resolution 3D image of the sample, as has been done previously in The University of Manchester [107]. The issues with this are identifying when to stop microtoming before going too far and destroying the channel and in selecting the correct angle from which to approach if looking to analyse a particular section of a channel.

178

For pre-initiation samples microtoming to a precise location would be relatively trivial as the void from the needle tip creates a very clear target to aim for and is rotationally symmetric. However for tree channels this is more complicated. The proposed solution for this would be to use grinding and polishing in combination with optical microscopy as before to approach the desired channel. Optical microscopy would then be used to work out approximately how far to microtome and markers to look for on the SEM which indicate when to stop. At this point SBFSEM would be used to precisely prepare the final section of the sample for the AFM-IR.

7.2.2. Alternative Materials The tests performed in this project were upon trees grown in epoxy. The use of this material made sense in this project given the experience in the use of epoxy in Manchester as well as the optical properties making it significantly easier to prepare the samples. However moving forward it would be ideal to perform tests upon other materials, particularly polyethylene (LDPE and XLPE). These are chemically more similar to the materials used in real systems making the work more directly applicable. As well as this they are less complex chemically, being just chains of carbon and hydrogen would simplify the analysis of the treeing process. As identified in pre-initiation samples here, there can be many types of reactions ongoing during treeing in epoxy. It should be tested whether these are also active in treeing in polyethylene.

Different materials may bring different issues in sample preparation. Polyethylene is generally a softer material than the glassy state epoxy used in this project and may therefore be more difficult to polish down. This would only be known once trialled and if difficulties are found with this then alternative methods including microtoming and freezing of the sample could be attempted.

Given the chemical analysis from AFM-IR allows us insights into the physical conditions and the mechanisms active during tree growth it would be greatly beneficial to be able to test trees grown in service. These would allow us to confirm whether the chemical signatures identified in trees grown in the lab environment are similar to those in trees which occur in cables. This would be important for identifying the extent to which the current accelerated ageing techniques we use (sharp needle tips for field enhancement and smaller insulations to aid with analysis) cause treeing to occur in fundamentally different ways. Ideally this would be compared to a tree grown in the exact same type of material using needle electrodes to see how they compare chemically and structurally.

7.2.3. Initiation testing performed with electroluminescence analysis Pre-initiation degradation is very difficult to study, the only known testing techniques for it are to study it chemically, such as through Raman spectroscopy or AFM-IR, and to analyse the associated

179 electroluminescence. The electroluminescence, a small light emission, has been noted as having different stages [31] which are understood to correlate to different stages of ageing prior to initiation.

Combining these mechanisms, by tracking the electroluminescence and then at different stages of ageing studying the samples to see the chemical degradation which has occurred, would give a new insight into the pre-initiation ageing of samples. It would allow a new physical and chemical relevance to be applied to the electroluminescence and to understand the processes which occur in the lead up to tree initiation. Chemical testing here would look to see if there is anything distinctive chemically in each stage of ageing (for example different types or different rates of ageing occurring).

Chemical testing would be best performed by a combination of FT-IR and AFM-IR. While AFM-IR can produce very high resolution and detailed maps it is limited to a small 2-dimensional plane. As observed here the chemical ageing does not necessarily occur in a perfect sphere, different locations around the needle tip will experience different levels and different types of ageing. It would be possible to miss something in the AFM-IR or not gain the full picture. Ideally the tests would use FT- IR for the bigger picture and AFM-IR for the high sensitivity provided by this technique.

180

8. References

[1] A. D. Bartlett, J. J. Daly, B. Gregory, and S. Phillip, "Cost Reductions in Power Cable Systems Using Lean Cable Technology," International Conference and Exhibition on Electricity Distribution, Nice, 1999. [2] H. Orton, "History of underground power cables," IEEE Electrical Insulation Magazine, vol. 29, no. 4, pp. 52-57, 2013. [3] E. J. McMahon and J. R. Perkins, "Evaluation of Polyolefin High-Voltage Insulating Compounds; Dendrite (Tree) Formation Under Highly Divergent Fields," IEEE Transactions on Power Apparatus and Systems, vol. 83, no. 12, pp. 1253-1260, 1964. [4] R. Schurch Brandt, "Three-Dimensional Imaging and Analysis of Electrical Trees," Thesis, The University of Manchester, 2014. [5] E. Moreau, C. Mayoux, C. Laurent, and A. Boudet, "The structural characteristics of water trees in power cables and laboratory specimens," IEEE Transactions on Electrical Insulation, vol. 28, no. 1, pp. 54-64, 1993. [6] I. Radu, M. Acedo, J. C. Filippini, P. Notingher, and F. Ftutos, "The effect of water treeing on the electric field distribution of XLPE. Consequences for the dielectric strength," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 7, no. 6, pp. 860-868, 2000. [7] A. Bulinski, S. Bamji, and R. Densley, "Factors affecting the transition from a water tree to an electrical tree," International Symposium on Electrical Insulation, 1988: IEEE, pp. 327-330. [8] M. Mashikian and A. Szarkowski, "Medium voltage cable defects revealed by off-line partial discharge testing at power frequency," IEEE Electrical Insulation Magazine, vol. 22, no. 4, pp. 24-32, 2006. [9] P. D. Bastidas and S. M. Rowland, "Interfacial aging in composite insulators as a result of partial discharge activity," Electrical Insulation Conference (EIC), 2017: IEEE, pp. 13-16. [10] E. Kobayashi, F. Komori, and Y. Suzuoki, "Partial discharge degradation at insulating polymer- polymer interface," International Symposium on Electrical Insulating Materials., 1998: IEEE, pp. 569-572. [11] N. Shimizu, K. Uchida, and S. Rasikawan, "Electrical tree and deteriorated region in polyethylene," IEEE Transactions on Electrical Insulation, vol. 27, no. 3, pp. 513-518, 1992. [12] K. Kaminaga, T. Suzuki, T. Uozumi, T. Haga, N. Yasuda, and T. Fukui, "The mechanism of degradation of polyethylene in a high electrical field," Conference on Electrical Insulation and Dielectric Phenomena, 1993: IEEE, pp. 666-671. [13] N. Shimizu and K. Horii, "The effect of absorbed oxygen on electrical treeing in polymers," IEEE Transactions on Electrical Insulation, no. 3, pp. 561-566, 1985. [14] M. Ieda and M. Nawata, "A consideration of treeing in polymers," Conference on Electrical Insulation & Dielectric Phenomena, 1972: IEEE, pp. 143-150. [15] J. H. Mason, "Assessing the resistance of polymers to electrical treeing," Physical Science, Measurement and Instrumentation, Management and Education - Reviews, IEE Proceedings A, vol. 128, no. 3, pp. 193-201, 1981. [16] F. Noto and N. Yoshimura, "Voltage and frequency dependence of tree growth in polyethylene," Conference on Electrical Insulation & Dielectric Phenomena, 1974: IEEE, pp. 207-217. [17] D. Auckland, J. Cooper, and B. Varlow, "Factors affecting electrical tree testing," in Science, Measurement and Technology, IEE Proceedings A, 1992, vol. 139, no. 1, pp. 9-13. [18] X. Chen, Y. Xu, X. Cao, S. Dodd, and L. Dissado, "Effect of tree channel conductivity on electrical tree shape and breakdown in XLPE cable insulation samples," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 18, no. 3, pp. 847-860, 2011. [19] R. Densley, "An investigation into the growth of electrical trees in XLPE cable insulation," IEEE Transactions on Electrical Insulation, no. 3, pp. 148-158, 1979.

181

[20] H. Zheng, S. M. Rowland, I. Iddrissu, and Z. Lv, "Electrical treeing and reverse tree growth in an epoxy resin," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 24, no. 6, pp. 3966-3973, 2017. [21] J. Champion and S. Dodd, "The effect of material composition and temperature on electrical tree growth in epoxy resins," International Conference on Dielectric Materials, Measurements and Applications, pp. 30 – 34, 2000. [22] T. Tanaka, "Tree initiation mechanisms," International Conference on Properties and Applications of Dielectric Materials, 1991: IEEE, pp. 18-24. [23] N. Hozumi, T. Okamoto, and H. Fukagawa, "Simultaneous measurement of microscopic image and discharge pulses at the moment of electrical tree initiation," Japanese Journal of Applied Physics, vol. 27, no. 4R, p. 572, 1988. [24] K. Uchida and N. Shimizu, "The effect of temperature and voltage on polymer chain scission in high-field region," IEEE Transactions on Electrical Insulation, vol. 26, no. 2, pp. 271-277, 1991. [25] L. Dissado and R. Hill, "The statistics of electrical tree inception," IEEE Transactions on Electrical Insulation, vol. 25, no. 4, pp. 660-666, 1990. [26] T. Tanaka and A. Greenwood, "Effects of charge injection and extraction on tree initiation in polyethylene," IEEE Transactions on Power Apparatus and Systems, no. 5, pp. 1749-1759, 1978. [27] H. Zheng, S. Rowland, F. Liu, and Z. Lv, "The Influence of an Air Gap at the Needle/Polymer Interface on Electrical Tree Growth," International Conference on Electrical Materials and Power Equipment (ICEMPE), 2019: IEEE, pp. 174-177. [28] S. Bamji, A. Bulinski, and R. Densley, "Threshold voltage of luminescence and electrical tree inception in low-density polyethylene," Journal of Applied Physics, vol. 63, no. 12, pp. 5841- 5845, 1988. [29] J. Champion, S. Dodd, and G. Stevens, "Quantitative measurement of light emission during the early stages of electrical breakdown in epoxy and unsaturated polyester resins," Journal of Physics D: Applied Physics, vol. 26, no. 5, p. 819, 1993. [30] S. Zhang and J. He, "Light emission characteristics during the initiation of the electric tree in insulating polymers," The Journal of Engineering, vol. 2017, no. 14, pp. 2650-2654, 2017. [31] J. Champion, S. Dodd, and G. Stevens, "Long-term light emission measurement and imaging during the early stages of electrical breakdown in epoxy resin," Journal of Physics D: Applied Physics, vol. 27, no. 3, p. 604, 1994. [32] S. S. Bamji, A. T. Bulinski, Y. Chen, and R. J. Densley, "Photodegradation versus hot-electron impact for electrical tree inception at low electric fields," International Conference on Properties and Applications of Dielectric Materials, 1991, pp. 51-54 vol.1. [33] N. Shimizu, T. Takahashi, and S. Iemura, "Initiation mechanism of electrical tree under alternating stress-electron impact or UV photo-degradation?," International Conference on Solid Dielectrics, ICSD'01, 2001: IEEE, pp. 423-426. [34] N. Shimizu and C. Laurent, "Electrical tree initiation," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 5, no. 5, pp. 651-659, 1998. [35] N. Shimizu, H. Katsukawa, M. Miyauchi, M. Kosaki, and K. Horii, "The space charge behavior and luminescence phenomena in polymers at 77 k," IEEE Transactions on Electrical Insulation, no. 5, pp. 256-263, 1979. [36] S. Bamji, A. Bulinski, Y. Chen, and R. Densley, "Threshold voltage for electrical tree inception in underground HV transmission cables," IEEE Transactions on Electrical Insulation, vol. 27, no. 2, pp. 402-404, 1992. [37] S. Zhang, Q. Li, J. Hu, B. Zhang, and J. He, "Electroluminescence and electrical degradation of insulating polymers at electrode interfaces under divergent fields," Journal of Applied Physics, vol. 123, no. 13, p. 135106, 2018.

182

[38] L. A. Dissado, "Understanding electrical trees in solids: from experiment to theory," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 9, no. 4, pp. 483-497, 2002. [39] T. Tanaka, "Space charge injected via interfaces and tree initiation in polymers," Conference on Electrical Insulation and Dielectric Phenomena, 2001: IEEE, pp. 1-15. [40] A. Hadid, C. Laurent, M. Mammeri, R. Clavreul, F. Duchateau, and J. Berdala, "An investigation into electrical ageing of insulating polymers by detection of charge injection under alternative divergent field," Journal de Physique III, vol. 4, no. 1, pp. 35-53, 1994. [41] C. Laurent, C. Mayoux, and S. Noel, "Mechanisms of electroluminescence during aging of polyethylene," Journal of Applied Physics, vol. 58, no. 11, pp. 4346-4353, 1985. [42] C. Laurent, F. Massines, and C. Mayoux, "Optical emission due to space charge effects in electrically stressed polymers," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 4, no. 5, pp. 585-603, 1997. [43] L. Niemeyer, "A generalized approach to partial discharge modeling," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 2, no. 4, pp. 510-528, 1995. [44] T. Okamoto and T. Tanaka, "Detection of non-destructive pulse currents in polyethylene prior to its dielectric breakdown," IEEE Transactions on Electrical Insulation, vol. 3, no. EI-20, pp. 643-645, 1985. [45] F. Noto, N. Yoshimura, and T. Ohta, "Tree initiation in polyethylene by application of DC and impulse voltage," IEEE Transactions on Electrical Insulation, no. 1, pp. 26-30, 1977. [46] Y. Sekii, "Initiation and growth of electrical trees in LDPE generated by impulse voltage," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 5, no. 5, pp. 748-753, 1998. [47] L. Ying and C. Xiaolong, "Electrical tree initiation in XLPE cable insulation by application of DC and impulse voltage," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 20, no. 5, pp. 1691-1698, 2013. [48] N. Yoshimura and F. Noto, "Effects of DC prestressing on tree initiation in polyethylene with and without needle-shaped void," IEEE Transactions on Electrical Insulation, no. 2, pp. 135- 140, 1984. [49] Y. Saito, M. Fukuzawa, and H. Nakamura, "On the mechanism of tree initiation," IEEE Transactions on Electrical Insulation, no. 1, pp. 31-34, 1977. [50] J. H. Mason, "Breakdown of solid dielectrics in divergent fields," Proceedings of the IEE-Part C: Monographs, vol. 102, no. 2, pp. 254-263, 1955. [51] M. Mammeri, C. Laurent, and J. Salon, "Influence of space charge buildup on the transition to electrical treeing in PE under ac voltage," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 2, no. 1, pp. 27-35, 1995. [52] M. G. Danikas and T. Tanaka, "Nanocomposites-a review of electrical treeing and breakdown," Electrical Insulation Magazine, IEEE, vol. 25, no. 4, pp. 19-25, 2009. [53] A. Vaughan, I. Hosier, S. J. Dodd, and S. Sutton, "On the structure and chemistry of electrical trees in polyethylene," Journal of Physics D: Applied Physics, vol. 39, no. 5, p. 962, 2006. [54] Y. Suzuoki, F. Komori, and T. Mizutani, "Partial discharges due to electrical treeing in polymers: phase-resolved and time-sequence observation and analysis," Journal of Physics D: Applied Physics, vol. 29, no. 11, p. 2922, 1996. [55] I. Iddrissu, Z. Lv, and S. M. Rowland, "The dynamic character of partial discharge in epoxy resin at different stages of treeing," International Conference on Dielectrics (ICD), 2016, vol. 2: IEEE, pp. 728-731. [56] R. Schurch, S. M. Rowland, R. S. Bradley, and P. J. Withers, "Imaging and analysis techniques for electrical trees using X-ray computed tomography," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 21, no. 1, pp. 53-63, 2014. [57] G. Chen and C. Tham, "Electrical treeing characteristics in XLPE power cable insulation in frequency range between 20 and 500 Hz," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 16, no. 1, pp. 179-188, 2009.

183

[58] X. Zheng and G. Chen, "Propagation mechanism of electrical tree in XLPE cable insulation by investigating a double electrical tree structure," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 15, no. 3, pp. 800-807, 2008. [59] S. J. Dodd, "A deterministic model for the growth of non-conducting electrical tree structures," Journal of Physics D: Applied Physics, vol. 36, no. 2, p. 129, 2002. [60] L. Hu, Y. Xu, X. Huo, and Y. Liao, "Chemical component analysis of electrical treeing in polyethylene by micro-infrared spectroscopy," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 23, no. 2, pp. 738-747, 2016. [61] J. Fothergill, "Filamentary electromechanical breakdown," IEEE Transactions on Electrical Insulation, vol. 26, no. 6, pp. 1124-1129, 1991. [62] J.-P. Crine, "Influence of electro-mechanical stress on electrical properties of dielectric polymers," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 12, no. 4, pp. 791- 800, 2005. [63] M. Arbab and D. Auckland, "Growth of electrical trees in solid insulation," IEE Proceedings A (Physical Science, Measurement and Instrumentation, Management and Education), vol. 136, no. 2, pp. 73-78, 1989. [64] H. Zeller and W. Schneider, "Electrofracture mechanics of dielectric aging," Journal of Applied Physics, vol. 56, no. 2, pp. 455-459, 1984. [65] A. El-Zein, M. Talaat, and M. El Bahy, "A numerical model of electrical tree growth in solid insulation," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 16, no. 6, pp. 1724- 1734, 2009. [66] B. Varlow and D. Auckland, "Mechanical aspects of electrical treeing in solid insulation," IEEE Electrical Insulation Magazine, vol. 12, no. 2, pp. 21-26, 1996. [67] D. Auckland, A. Taha, and B. Varlow, "Correlation of mechanical properties with electrical treeing behaviour at elevated temperatures," Conference on Electrical Insulation and Dielectric Phenomena, 1993: IEEE, pp. 636-641. [68] K. Stark and C. Garton, "Electric strength of irradiated polythene," Nature, vol. 176, no. 4495, pp. 1225-1226, 1955. [69] I. Iddrissu, S. M. Rowland, H. Zheng, Z. Lv, and R. Schurch, "Electrical tree growth and partial discharge in epoxy resin under combined AC and DC voltage waveforms," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 25, no. 6, pp. 2183-2190, 2018. [70] T. Lewis, J. Llewellyn, M. Van der Sluijs, J. Freestone, and R. Hampton, "A new model for electrical ageing and breakdown in dielectrics," International Conference on Dielectric Materials, Measurements and Applications, 1996: IET, pp. 220-224. [71] B. Bolasodun, "Microwave effects on the curing, structure properties and decomposition of epoxy resins," Thesis, The University of Manchester, 2011. [72] A. Hajlane, H. Kaddami, R. Joffe, and L. Wallström, "Design and characterization of cellulose fibers with hierarchical structure for polymer reinforcement," Cellulose, vol. 20, no. 6, pp. 2765-2778, 2013. [73] D. Hepburn, I. Kemp, A. Shields, and J. Cooper, "Degradation of epoxy resin by partial discharges," IEE Proceedings-Science, Measurement and Technology, vol. 147, no. 3, pp. 97- 104, 2000. [74] S. Bahadoorsingh and S. M. Rowland, "Investigating the influence of the lubricant coating on hypodermic needles on electrical tree characteristics in epoxy resin," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 17, no. 3, pp. 701-708, 2010. [75] P. D. Bastidas and S. M. Rowland, "Developing test methods for interfacial ageing in composite insulators," Conference on Electrical Insulation and Dielectric Phenomena (CEIDP), 2015: IEEE, pp. 594-597. [76] T. Liu, B. Hui, M. Fu, S. Hou, B. Luo, and G. Wang, "Experimental and simulation analysis of electrical breakdown for 220kV silicone rubber pre-moulded cable joints," International Electrical Insulation Conference (INSUCON), 2017: IEEE, pp. 1-5.

184

[77] S. Morsch, P. D. Bastidas, and S. M. Rowland, "AFM-IR Insights into the Chemistry of Interfacial Tracking," Journal of Materials Chemistry A, 2017. [78] T. Miyashita, "Deterioration of water-immersed polyethylene-coated wire by treeing," IEEE Transactions on Electrical Insulation, no. 3, pp. 129-135, 1971. [79] T. Tabata, T. Fukuda, and Z. Iwata, "Investigations of water effects on degradation of crosslinked polyethylene insulated conductors," IEEE Transactions on Power Apparatus and Systems, no. 4, pp. 1361-1370, 1972. [80] M. Shaw and S. Shaw, "Water treeing in solid dielectrics," IEEE Transactions on Electrical Insulation, no. 5, pp. 419-452, 1984. [81] R. Ross and J. Smit, "Composition and growth of water trees in XLPE," IEEE Transactions on Electrical Insulation, vol. 27, no. 3, pp. 519-531, 1992. [82] J. Chen and J. Filippini, "The morphology and behavior of the water tree," IEEE Transactions on Electrical Insulation, vol. 28, no. 2, pp. 271-286, 1993. [83] T. Uematsu, Z. Iwata, S. Irie, and O. Fujii, "Bow-tie-tree in EPR cables are accelerated water treeing test," IEEE Transactions on Power Delivery, vol. 7, no. 4, pp. 1667-1676, 1992. [84] S. Boggs, J. Densley, and J. Kuang, "Mechanism for impulse conversion of water trees to electrical trees in XLPE," IEEE Transactions on Power Delivery, vol. 13, no. 2, pp. 310-315, 1998. [85] I. Radu, "Electric field calculation and the influence of water trees on insulation breakdown in needle–plane geometry," Journal of Electrostatics, vol. 60, no. 1, pp. 49-67, 2004. [86] M. Acedo, I. Radu, F. Frutos, J. Filippini, and P. Notingher, "Water treeing in underground power cables: modelling of the trees and calculation of the electric field perturbation," Journal of Electrostatics, vol. 53, no. 4, pp. 267-294, 2001. [87] C. Stancu et al., "Electric field computation in water treed polyethylene with space charge accumulation," IEEE International Symposium on Electrical Insulation, 2006: IEEE, pp. 186- 189. [88] C. Stancu, P. V. Notingher, and P. Notingher, "Influence of space charge related to water trees on the breakdown voltage of power cable insulation," Journal of Electrostatics, vol. 71, no. 2, pp. 145-154, 2013. [89] C. Grave, I. McEwan, and R. A. Pethrick, "Influence of stoichiometric ratio on water absorption in epoxy resins," Journal of Applied Polymer Science, vol. 69, no. 12, pp. 2369- 2376, 1998. [90] A. A. C. Pereira and J. R. M. d’Almeida, "Effect of the hardener to epoxy monomer ratio on the water absorption behavior of the DGEBA/TETA epoxy system," Polímeros, vol. 26, no. 1, pp. 30-37, 2016. [91] S. Manjang, M. Arief, G. Duma, and T. Waris, "Impact of Water Absorptionon Dielectric Properties and Breakdown Voltage of Polymer Epoxy and Xlpe," International Journal of Engineering and Science Applications, vol. 3, no. 1, pp. 23-28, 2016. [92] M. Kosaki, N. Shimizu, and K. Horii, "Treeing of polyethylene at 77K," IEEE Transactions on Electrical Insulation, no. 1, pp. 40-45, 1977. [93] N. Shimizu, K. Uchida, and K. Horii, "Initiation mechanism of electrical tree—chain scission by injected charge and role of oxygen," Conference on Electrical Insulation & Dielectric Phenomena, 1987: IEEE, pp. 419-424. [94] K. Uchida, H. Asai, N. Shimizu, and T. Takahashi, "Initiation mechanism of electrical tree and deteriorated region in polymers," International Conference on Properties and Applications of Dielectric Materials, 1991: IEEE, pp. 240-243. [95] A. Vaughan, S. J. Dodd, and S. Sutton, "A Raman microprobe study of electrical treeing in polyethylene," Journal of Materials Science, vol. 39, no. 1, pp. 181-191, 2004. [96] C. LibreTexts. (2019, 02/03/2020). Infrared Spectroscopy Absorption Table [Online]. Available:

185

https://chem.libretexts.org/Bookshelves/Ancillary_Materials/Reference/Reference_Tables/S pectroscopic_Parameters/Infrared_Spectroscopy_Absorption_Table. [97] A. Hammiche, H. Pollock, M. Reading, M. Claybourn, P. Turner, and K. Jewkes, "Photothermal FT-IR spectroscopy: A step towards FT-IR microscopy at a resolution better than the diffraction limit," Applied Spectroscopy, vol. 53, no. 7, pp. 810-815, 1999. [98] S. N. Magonov and D. H. Reneker, "Characterization of polymer surfaces with atomic force microscopy," Annual Review of Materials Science, vol. 27, no. 1, pp. 175-222, 1997. [99] O. Custance, R. Perez, and S. Morita, "Atomic force microscopy as a tool for atom manipulation," Nature Nanotechnology, vol. 4, no. 12, p. 803, 2009. [100] K. Kjoller, J. Felts, D. Cook, C. Prater, and W. P. King, "High-sensitivity nanometer-scale infrared spectroscopy using a contact mode microcantilever with an internal resonator paddle," Nanotechnology, vol. 21, no. 18, p. 185705, 2010. [101] C. Mayet, A. Dazzi, R. Prazeres, J.-M. Ortega, and D. Jaillard, "In situ identification and imaging of bacterial polymer nanogranules by infrared nanospectroscopy," Analyst, vol. 135, no. 10, pp. 2540-2545, 2010. [102] A. Dazzi, C. B. Prater, Q. Hu, D. B. Chase, J. F. Rabolt, and C. Marcott, "AFM–IR: combining atomic force microscopy and infrared spectroscopy for nanoscale chemical characterization," Applied Spectroscopy, vol. 66, no. 12, pp. 1365-1384, 2012. [103] S. Morsch, Y. Liu, S. B. Lyon, and S. R. Gibbon, "Insights into epoxy network nanostructural heterogeneity using AFM-IR," ACS applied materials and interfaces, vol. 8, no. 1, pp. 959- 966, 2015. [104] H. McDonald, P. D. Bastidas, S. Rowland, and S. Morsch, "Chemical Analysis of Solid Insulation Degradation using the AFM-IR Technique," International Conference on Dielectrics (ICD), 2018: IEEE, pp. 1-4. [105] H. McDonald, S. Morsch, and S. M. Rowland, "High Resolution Chemical Analysis of Electrical Trees through AFM-IR Spectroscopy," Conference on Electrical Insulation (EIC), Calgary, 2019 (In Press): IEEE. [106] H. McDonald, S. Morsch, and S. Rowland, "Chemical Analysis of Tree Growth in Epoxy Resin Using AFM-IR Spectroscopy," IEEE Transactions on Dielectrics and Electrical Insulation, 2020 (In Press). [107] R. Schurch, S. M. Rowland, R. S. Bradley, and P. J. Withers, "Comparison and combination of imaging techniques for three dimensional analysis of electrical trees," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 22, no. 2, pp. 709-719, 2015. [108] A. Garton, R. Densley, and A. Bulinski, "Initial results of a study of water trees in XLPE by Infrared spectroscopy," IEEE Transactions on Electrical Insulation, no. 6, pp. 500-501, 1980. [109] R. Schurch, S. M. Rowland, and T. Starborg, "Serial block-face scanning electron microscopy for three-dimensional imaging of electrical trees," International Conference on Solid Dielectrics (ICSD), 2013: IEEE, pp. 271-274. [110] P.-H. Vallotton, M. Denn, B. Wood, and M. Salmeron, "Comparison of medical-grade ultrahigh molecular weight polyethylene microstructure by atomic force microscopy and transmission electron microscopy," Journal of Biomaterials Science, Polymer Edition, vol. 6, no. 7, pp. 609-620, 1995. [111] J.-i. Aihara, "Why aromatic compounds are stable," Scientific American, vol. 266, no. 3, pp. 62-69, 1992. [112] Savitzky, A. and M. J. Golay, "Smoothing and differentiation of data by simplified least squares procedures." Analytical chemistry, 1964, 36(8): 1627-1639. [113] N. Baden, M. Yasuda, A. Yoshida, and N. Muraki, "New method for chemical characterization of polymer materials in industrial devices: AFM-IR with FIB sample preparation," International Symposium on the Physical and Failure Analysis of Integrated Circuits, 2015: IEEE, pp. 496-499.

186

[114] S. Rasikawan, H. Ishihara, and N. Shimizu, "Comparison between water-treed and deteriorated regions. Electrical tree precursor," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 1, no. 4, pp. 597-603, 1994. [115] G. Stevens, "Microstructure in crosslinked epoxy polymers," Structural adhesives: Developments in resins and primers(A 87-47826 21-37). London and New York, Elsevier Applied Science Publishers, 1986, pp. 201-267, 1986. [116] R. Schurch, S. M. Rowland, and R. S. Bradley, "Partial discharge energy and electrical tree volume degraded in epoxy resin," Conference on Electrical Insulation and Dielectric Phenomena (CEIDP), 2015: IEEE, pp. 820-823. [117] M. Brown, J. Champion, S. Dodd, and P. Mudge, "An investigation of partial discharge energy dissipation and electrical tree growth in an epoxy resin," International Conference on Solid Dielectrics, (ICSD), 2004, vol. 1: IEEE, pp. 288-291. [118] N. Hozumi, T. Okamoto, and H. Fukagawa, "TEM observation of electrical tree paths and microstructures in polyethylene," Japanese Journal of Applied Physics, vol. 27, no. 7R, p. 1230, 1988. [119] K. Wu and L. Dissado, "Model for electrical tree initiation in epoxy resin," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 12, no. 4, pp. 655-668, 2005. [120] T. Tanaka, "Charge transfer and tree initiation in polyethylene subjected to AC voltage stress," IEEE Transactions on Electrical Insulation, vol. 27, no. 3, pp. 424-431, 1992. [121] K. S. Johansson, "Surface modification of plastics," in Applied Plastics Engineering Handbook: Elsevier, 2017, pp. 443-487. [122] C. Mayoux, "Degradation of insulating materials under electrical stress," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 7, no. 5, pp. 590-601, 2000. [123] E. J. McMahon, "A Tutorial on Treeing," IEEE Transactions on Electrical Insulation, vol. EI-13, no. 4, pp. 277-288, 1978. [124] B. X. Du, J. G. Su, J. Li, and T. Han, "Effects of mechanical stress on treeing growth characteristics in HTV silicone rubber," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 24, no. 3, pp. 1547-1556, 2017. [125] M. Hikita, S. Tajima, I. Kanno, I. Ishino, G. Sawa, and M. Ieda, "High-field conduction and electrical breakdown of polyethylene at high temperatures," Japanese Journal of Applied Physics, vol. 24, no. 8R, p. 988, 1985. [126] Z. Liu, Q. Liu, and Z. Wang, "Effect of electric field configuration on streamer and partial discharge phenomena in a hydrocarbon insulating liquid under AC stress," Journal of Physics D: Applied Physics, vol. 49, no. 18, p. 185501, 2016.

187

9. Appendices

Appendix A – Energy conversion calculation from partial discharges for sample AFM(1)-A1(2)

To see whether the channel could be explained by the observed discharges the approach of Schurch [115] and Dodd [116] is applied. Standard equations are used to calculate the energy from the discharges and the energy required to vaporise a channel this size.

The Partial Discharge Energy is calculated using:

PD Energy=∑ (qi · vi)

Schurch found around 7% of the PD energy was converted to vaporisation energy. Although this channel is much smaller and closer to the needle tip it is assumed this value remains applicable.

In this equation the energy from each discharge is dependent upon the instantaneous voltage. With this being a preliminary calculation to determine the possibility of these discharges causing the channel formation, an approximation is made that the discharges can be taken as having occurred at the RMS voltage (10.6 kV). The average discharge size is taken to be 750 fC, with 112 pulses occurring. (These assumptions can be made as there appears to be no relationship between instantaneous voltage and discharge size meaning rough averages can be taken.)

This gives a total PD Energy of 8.9 x 10-7 J, assuming a 7% conversion rate this gives a vaporisation energy of 6.2 x 10-8 J.

We can now calculate the energy required to produce a channel of this volume, again using [115] which was also performed in epoxy we can assume and energy of vaporisation per unit volume of 7.56 x 1010 J/m3.

The tree volume is calculated assuming the channel to be approximately cylindrical with a diameter of 500 nm and a length of 15 µm.

The equation for the volume of a cylinder is V=π x r2 x L, which gives a volume of 3.927 x 10-18 m3. The total energy required can now be calculated as:

Vaporisation Energy Required = 7.56x1010 x 3.927x10-18 = 29.6 x 10-8 = 2.96 x 10-7 J

Vaporisation Energy Required = 2.96 x 10-7 J

188

Energy/Energy Required (assuming 7% conversion) = (6.2 x 10-8)/(2.96 x 10-7) = 0.209 = 20.9% of the energy required.

Using the standard figure of 7% energy conversion the PDs are shown to not have sufficient energy to be the cause of the tree initiation and subsequent channel formation. Instead alternative processes should be considered.

Important provisos should be applied to this value of 20.9% however as a number of assumptions were made in reaching this figure which were determined for a larger, more mature tree. These may not hold for such a young tree as this, however much is not known about trees this small making truly accurate estimations impossible. There could also be discharges below the detection limit of 500 fC which also contribute to the vaporisation of the material.

If there is a significantly different conversion rate of energy for new channels (35% as opposed to 7%) then these energies are well matched though this is a significant change in the energy conversion rate. This calculation cannot be entirely conclusive either way, it neither clearly indicates the initiation channel is formed by partial discharges, nor suggests this is impossible. It is believed however it is indicative of non-PD processes involved in tree initiation.

189

Appendix B – List of Publications

[1] McDonald, H., et al. (2020 (In Press)). "Chemical Analysis of Tree Growth in Epoxy Resin Using AFM-IR Spectroscopy." IEEE Transactions on Dielectrics & Electrical Insulation.

[2] McDonald, H., et al. “Chemical Analysis of Solid Insulation Degradation using the AFM-IR Technique” IEEE 2nd International Conference on Dielectrics (ICD) 2018

[3] McDonald, H., et al. (In Press). “High Resolution Chemical Analysis of Electrical Trees through AFM-IR Spectroscopy”. IEEE Conference on Electrical Insulation, Calgary, 2019

[4] McDonald, H., et al. (In Submission). “A High-Resolution Study of Chemical Aging Prior to Electrical Tree Growth” IEEE 3rd International Conference on Dielectrics (ICD) 2020

190

Chemical Analysis of Tree Growth in Epoxy Resin using AFM-IR Spectroscopy

H. McDonald1, S. Morsch2 and S. M. Rowland1 1 Department of Electrical and Electronic Engineering 2 Department of Materials The University of Manchester Manchester, M13 9PL, UK

ABSTRACT Recent advances have allowed submicron features of electrical trees to be imaged. Until recently it has not been possible to analyze chemical aging on the same scale. Here, the use of atomic force microscopy – infrared spectroscopy (AFM-IR) to characterize chemical degradation around a needle tip, and in the vicinity of electrical tree branches grown in epoxy resin is explored. The spectral signatures associated with the degradation of epoxies during electrical treeing are successfully identified with 50 nm resolution. This allows identification of chemical degradation in unprecedented detail. The method of polishing and grinding to prepare the samples was found to remove degraded material from exposed channels. This issue was overcome by not fully exposing the tree, but bringing it close enough to the surface that the AFM-IR was able to see within the channel. A short tree examined immediately after initiation showed no chemical degradation, indicating that its growth was not a chemically active process, suggesting electromechanical fracturing is more likely. Mature channels showed no singular chemical signature. Carbonyl groups were identified in all mature channels and were particularly concentrated at the branching points studied, suggesting higher levels of degradation during branching or accumulation at points of bifurcation. It is believed that these reactions are precipitated by hot electron impacts during discharges. AFM-IR has provided unprecedented detail concerning chemical damage incurred during the treeing process and can provide vital information to inform models of aging. Index Terms — AFM-IR, electrical tree, epoxy, chemical, characterization, imaging

1 INTRODUCTION growth, but as a consequence of it. An alternative is to UNDERSTANDING electrical tree formation is crucial consider chemical degradation preceding channel growth, to the development of more reliable and cost-effective weakening the surrounding region until localized power cable systems. Electrical trees consist of tubules mechanical or electrical breakdown occurs causing channel which form within polymers under high electric fields. growth. Many variations are seen in the chemical and When trees bridge a dielectric between two conductors they structural make-up of trees, depending on material [2], trigger catastrophic failure. The growth of an electrical tree temperature [3], and electric field magnitude and frequency requires the formation of extended voids within a solid [4]. Variations can be observed at different stages of growth region of insulation. It is widely assumed that this is driven of the same tree [1, 5, 6]. by the occurrence of partial discharges which accompany Three forms of channel can be distinguished with channel growth [1]. fundamental differences in their formation and structure: For a tree channel to grow, sufficient energy must be Firstly conducting channels in which discharges occur only transferred at the tip of the channel to the polymer to induce at the tip of trees due to graphitic carbon formation on the further damage. The source of this energy is taken to be the channel walls reducing the field within the tree [5]; secondly electrical field and a key conversion process is through there are non-conducting channels, in which discharges partial discharges (PDs) within the channel or at the channel occur along the length of the channels; finally filamentary or tip [2]. Another theory considers electromechanical forces fine trees which have been observed in epoxy resins [1]. created at the channel tip as a result of Maxwell stresses. These channels are smaller in diameter (< 1 µm) and, The consequence of such mechanical strain and stress is contrary to the cases of wider channels, there is no fracturing at the channel tip, extending the tree further into observable discharge during tree growth. Each of these the bulk. Here discharges occur, not as the driver of channel treeing types has distinctive physical structures and PD behavior. Further understanding requires improved Manuscript received on 30 September 2019, in final form 2 December 2019, accepted xx Month 20yy. Corresponding author: H. McDonald. 191

knowledge of the progression of chemical degradation within and around the channels. Infrared (IR) spectroscopy identifies chemical bonds by mapping the absorption spectrum of the material. This technique can distinguish functional groups as well as the wider structure of the molecules. FT-IR has been used in the study of tree initiation [7] and the presence of carbonyl groups was identified. Micro-IR spectroscopy has been used in the study of mature channels [6] identifying carbonyls, hydroxyls and ether groups. However, the resolution of these techniques is ~15 µm and since tree channels have widths ~1 µm, IR spectroscopy cannot resolve the locations Figure 1. AFM-IR with top-down illumination. An IR source is pulsed, of chemical products within channels [6]. Raman inducing rapid thermal expansion of the sample. This is detected by spectroscopy, which measures the inelastic scattering of deflection of the AFM probe cantilever [15]. light, is complementary to FT-IR and has been applied to tree initiation [7] and mature channels [5, 8]. The resolution kick can be calculated, along with the amplitude of the of Raman is ~2 µm, again limiting the precision with which absorption. Using this method the absorbance spectrum of a groups can be located with respect to tree channels. particular location can be determined, or the absorbance at a Other chemical techniques used to study electrical treeing particular wavelength can be mapped across an area of the include methylene blue dye to identify the presence of sample. Both methods are utilized in this paper. oxygen or oxidation around high field points prior to tree 2.2 SAMPLE PREPARATION initiation [9]. Electron Spin Resonance has also been used to confirm the production of free radicals during the pre- Araldite (LY 5052) and Aradur (HY 5052) were used to initiation period of degradation [10] and during tree growth fabricate samples. The Araldite is Novalak-based with an [11]. ether diluent; the Aradur is composed of a number of Atomic force microscopy combined with infrared polyamines producing a cured epoxy with high thermal and spectroscopy (AFM-IR) provides a technique for high chemical stability. The samples are mixed in a 100:38 ratio resolution chemical analysis on the scale of tree channels, according to manufacturer specifications. This solution is with a resolution of 50 nm. The technique was developed in hand-mixed for 2 mins and then by magnetic stirrer for 7 1999 by Hammiche et al [12]. It is analogous to FT-IR in mins. The solution is then degassed at room temperature for that it depends upon the infrared absorption of a region to 30 mins after which the solution is transferred to 25 x 25 determine the chemical composition. The combination of mm acrylic cubes, and a steel needle (Ogura, 3 µm tip infrared excitation with AFM overcomes the diffraction radius) is inserted leaving a 2 mm gap to the bottom surface. limit. The nature of the AFM-IR means that sub-surface This assembly is degassed for a further 40 mins and then left materials will contribute to the measured absorption. The to cure at room temperature and pressure for 24 hrs. The ° depth of material which can be detected using the AFM-IR samples are then post-cured at 100 C for 4 hrs and stored in is material and AFM-IR dependent [12]. However a desiccator for at least a week before testing. experience indicates that in these samples a molecule must 2.3 HIGH VOLTAGE TESTING be within a few micrometres of the surface to contribute to The experimental setup is shown in Figure . A 50 Hz the AFM-IR measurements. voltage was applied to the needle giving a typical point- The use of AFM-IR in the study of electrical treeing has plane configuration for the treeing tests [16]. For the previously been reported by the authors examining both the initiation tests only, an Omicron MPD600, in combination degraded region formed prior to tree initiation [13] and the with a dummy sample and balanced circuit, measured partial distribution of products within and around mature channels discharge activity. A CCD camera tracked tree initiation and [14]. This paper presents a continuation of this work, in growth. The sample in which a short tree was examined was which the degradation products in early tree formation and subject to 10.6 kV with the voltage being removed on at branch points are examined for the first time. RMS first visible signs of tree initiation. The mature channels 2 METHODOLOGY studied in this sample were first subject to a voltage of 17.7 2.1 THE AFM-IR TECHNIQUE kVRMS to encourage initial tree formation and growth before An AFM probe, as shown in Figure 1, is attached to a dropping the voltage to 8.85 kVRMS for further slow growth. cantilever and traced across the sample surface. The vertical The voltage was removed when channel growth approached deflection of the probe is tracked as it traverses the surface the plane electrode. to build a surface height profile. A broadband, pulsed 2.4 AFM-IR MEASUREMENTS tunable laser illuminates the area under the probe. When tuned to a wavelength corresponding to the excitation Before a region of interest can be studied by AFM-IR, the energy of a chemical bond within the sample, light is location must be exposed in a flat surface. To do so, the surrounding material was first cut away to reduce the sample absorbed and the bond excited, leading to thermal expansion 3 of the region below the AFM probe. This expansion ‘kicks’ to the size required for the AFM-IR (around 1 cm ). The the probe and causes it and the cantilever to oscillate. This needle was then carefully removed. The epoxy surface was mechanical movement is tracked by a photodiode. By ground down to the area of interest, first using silicon Fourier transforming the oscillation, the amplitude of the 192

a) b)

50 µm 25 µm

c) 15 µm

Figure 2. The circuit diagram illustrating the system used during sample growth including HV supply, CCD camera and PD monitoring. carbide paper and then with a diamond compound polishing Figure 3. A polished surface with an exposed tree channel: a) Optical paste. Image (bottom-lit); sub-surface channel visible. b) Optical Image (top-lit); An Anasys Instruments NanoIR2 was used in contact reflective material visible within channels below the surface. c) Selected mode with a gold-coated silicon nitride probe. The light AFM-IR spectra of channel and surrounding areas (normalized to 1504 cm-1 source was an Optical Parametric Oscillator (OPO) laser peak). Inset: AFM-IR height profile in spectra location, 30 µm x 15 µm. used in top-down configuration. This is a pulsed, tunable They have been divided into three distinct areas: Red points infrared light source with a beam size of ~30 µm and a and associated spectral lines correspond to regions closest to spectral resolution of 4 cm-1. Spectra are acquired at a the exposed channel, blue points to areas further from the number of different locations for each sample. These use channel and green points to regions above a buried channel. 1024 co-averages for each data point, and are then smoothed The spectra obtained suggest the presence of different by Savitzky-Golay filter and normalized. The compositional chemical products in each of these regions. In both the maps are produced at a number of chosen wavenumbers for 1800 – 1620 cm-1 and the 1150 – 1000 cm-1 spectral regions each sample with 600 points per 300 scan lines, using 32 co- the green spectra (representing buried channels) show averages per data point. increased absorbance when compared with both the red spectra (areas closest to exposed channel) and the blue 3 RESULTS spectra (further from the channel). The 1800 – 1620 cm-1 The AFM-IR technique has been applied to four channels region is indicative of C=O bonds. In particular a broad peak to identify a range of chemical signatures associated with is noted at 1740 cm-1 in the area of the unexposed channel. treeing degradation. Each is reported below: These appear to be on or in the channel walls, and there is no evidence of degradation extending beyond the channel Sample 1 – 100 µm from a needle tip in a 1600 µm tree walls. Previously [14] discussed the testing and analysis of AFM-IR maps of the area in Figure 3c are given in Figure sample 1 and this paper will expand upon this. The channel 4. In these maps, red indicates a greater level of IR is within a tree of 1600 µm length, has a width of 3 µm, and absorbance. Figure 4d is a map of the distribution of is considered to be non-conducting due to the white absorption at 1726 cm-1, with absorbance in this region reflective material within the channel [16]. The sample is taken to be indicative of carbonyl groups (potentially esters). 100 µm from the needle tip. These show an increase in the number of carbonyl bonds in Optical images of the channel are given in Figure . Bottom the region above the buried channels, however no lighting shows the structure of the tree below the surface degradation is detected in the exposed section of the channel and top lighting demonstrates that this tree reaches the and less IR absorption is observed within the channel. This surface. AFM-IR was used to study the region around the reduced signal is believed to be due to degradation products exposed channel. The points selected for spectral analysis being removed during the polishing process. However, it are shown in Figure c with selected corresponding spectra. may also be indicative of the AFM probe

a) b) c) d)

e) f) g) 15 µm

Figure 4. AFM-IR images of the surface around the channel in sample 1: a) surface profile. IR Absorption amplitude at wavenumbers: b) 1664 cm-1, c) 1702 cm-1, d) 1726 cm-1, e) 1248 cm-1, f) 1604 cm-1, g) 3300 cm-1. Each map is ratioed against 1504 cm-1 (corresponding to an aromatic peak). The images represent a 30 µm x 15 µm surface. Scale bar applies to all spectral maps. 193

a) b) 20 µm 25 µm

15 µm c) Figure 5. Sample 1, optically imaged in the background with the 1702 cm- 50 µm chemical map overlaid. Red areas indicate greater absorption at this wavelength. being unable to make clean contact within the channel void. Enhanced absorbance over the buried channel is noted at all wavenumbers across the carbonyl region (5c). In Figure 5 the 1702 cm-1 map of Figure 4c is overlaid on an optical image of the channel, and this confirms that the degradation products are contained within the unexposed channels and that their formation or dispersion beyond the channel walls is limited. The wavenumber 1664 cm-1 can be Figure 6. Epoxy sample with a sub-surface channel. a) Optical Image (top- assigned to C=C bond formation (Figure 4b) and follows the lit); channel exposed at the surface. Exposed areas circled. b) Optical Image same pattern as the carbonyl maps, degradation products (bottom-lit); channel structure visible below surface. c) Selected AFM-IR being concentrated within the buried section of the channel, spectra of channel and surrounding areas (normalized to 1504 cm-1 peak). not extending far beyond the channel and not being evident Inset: 50 x 50 µm AFM-IR height profile displaying location of spectra. within the exposed sections of the channel. Others mapped are 1604 cm-1 (Figure 4f) and 1248 cm-1 (Figure 4e) which (blue) there is a significant peak at 1740 cm-1 indicating the give weaker responses yet follow the same trends. In these formation of a C=O group, potentially an aldehyde. A region cases, due to the complexity of the epoxy chemistry, no of increased absorption from 1200-1000 cm-1 is again attempt has been made in this paper to assign specific identified. The chemistry is further examined through the functional groups. chemical maps in Figure 7. As with sample 1, Figure 7b -1 Sample 2 – 500 µm from a needle tip within a 1600 µm shows little increase in absorption at 1248 cm . This again tree indicates that although there are differences in the spectra The second channel studied is from a different sample, but (Figure 6c), these do not appear to represent large scale a similarly sized channel, and grown at the same voltage as changes to the chemical structure in this region. Local -1 in sample 1. This location was ~500 µm from the needle tip concentrations of carbonyl functional groups (1708 cm , -1 -1 in a 1600 µm long tree. Small parts of the channel are 1726 cm , 1742 cm ) and unsaturated carbon to carbon -1 exposed to the surface as indicated in Figure 6a (circled). bonds (C=C) (1656 cm ) are again identified. The rest of the tree structure remains unexposed In sample 1 the degradation products were localized immediately below the surface. The spectra (Figure 6c); all within the channel and did not extend beyond the channel taken from areas above or around the unexposed channel walls. In this sample however, analysis indicates that the invite comparison with those from sample 1. Here the blue degradation products extend into the bulk of the epoxy. and red markers indicate areas above the channels and in This is shown in Figure 8b, an overlay of the AFM-IR surrounding areas respectively. The green spectra are from absorbance map at 1742 cm-1 (corresponding to aldehydes) an area over 15 µm from any channel. and an optical image. The degradation products extend 5 µm Again, an increase across the 1800-1600 cm-1 range (red beyond the outer sections of the channel. Figure 8c, another -1 compared to green) suggests the formation of C=O bonds overlay at 1742 cm , is contrast-adjusted to show the within the channels below the surface. Notably however, in locations of highest absorbance. These can be seen to be the one of the spectra from within the centre of the channel branching points, indicating a higher number of C=O

a) b) c) d)

e) f) g) h) 30 µm

Figure 7. AFM-IR images of the surface of sample 2. a) Surface profile. IR absorption amplitude at wavenumbers: b) 1248 cm-1, c) 1604 cm-1, d) 1640 cm-1, e) 1656 cm-1, f) 1708 cm-1, g) 1724 cm-1 h) 1742 cm-1. Each is ratioed against 1504 cm-1 (aromatic peak). Maps are 30 µm x 30 µm. Scale bar applies to all maps.

194

a) b) c) d)

30 µm

Figure 8. Sample 2, background images show the channel structure below the epoxy surface. Overlaid images use the AFM-IR chemical mapping at different wavelengths and contrast settings. a) Optical Image b) 1742 cm-1 and set to show the extent of the carbonyls c) 1742 cm-1 and set to show the highest concentrations of carbonyls c) 1656 cm-1 and set to show the highest concentrations of alkene bonds. functional groups present. Figure 8d, an overlay of 1656 cm- the exposed section of the tree, as well as above the buried 1, indicative of C=C groups, shows an increase within the channel (each given in black), show no clear absorbance channels in general but any localization of the products is changes. The dominant factor controlling the chemical less defined than for the carbonyls. In Figure 8b a smaller composition in each spectrum is the location around the channel (0.5 – 1 µm in width) can also be identified as needle tip. The channel itself does not appear to have an containing carbonyl groups (1742 cm-1). associated chemical signature. This map also allows Sample 3 – 15 µm tree at the needle tip consideration of the degradation products generated in the The third sample is a small tree from which the voltage degraded region prior to initiation. The spectra taken from was removed immediately following initiation, after 7 areas closest to the needle tip are coloured in green, areas minutes at 10.6 kVRMS. Optically imaged in Figure 9 it further from this in blue and those furthest from the needle measures approximately 15 µm in length and had undergone tip are in red. The spectra appear to indicate an increase in a first bifurcation near its tip. Figure 9b is a top-lit image of carbonyls as they move further from the needle tip. This is the same location after the sample had been cross-sectioned. difficult to explain and contradicts previous findings [7, 13]. The points where the channel is exposed at the surface are This may be a by-product of the AFM-IR data processing: -1 circled. The remainder of the channel is immediately below In this case the spectra are normalized against the 1500 cm the surface. aromatic peak, chosen as a typically unreactive group. If this was being degraded during the aging process then it may Figure 9c shows the IR absorption spectra at different appear as an increase in the spectra of other groups. This locations (shown inset). Spectra taken from within or around does not however explain the lack of carbonyls being formed as is typically identified. This suggests there a) b) remains much to understand regarding the tree initiation process. 10 µm PDs were recorded during and after tree initiation. These 10 µm were small in number with a maximum of 1.4 pC after tree initiation. This paucity suggests that the process of tree formation was not directly driven by the PDs. c) 15 µm Chemical maps at a number of wavelengths are given in Figure 10. These, as with the spectra, show that there is no chemical degradation product associated with the electrical tree within or surrounding the channel. It can also be seen that the channel extends far beyond the degraded region (particularly clear in Figure 10b) indicating that while this region may have been relevant in the tree formation, it is not directly the limiting factor in how far it is able to extend from the needle tip. Circled in each map is the Figure 9. a) Optical Image (bottom-lit); projection of channel observed to exposed section of the channel, showing no increase in initiate from needle tip. b) Optical Image (top-lit); after polishing of surface degradation product, though each appears to show a slight channel can be seen at the exposed surface. c) Selected AFM-IR spectra of decrease in the channel. This, as with sample 1, may be -1 area (normalized to 1504 cm peak). Inset: AFM-IR height profile showing because the AFM probe tip did not make effective contact the location of the spectra, 30 µm x 30 µm. within the channel void.

a) b) c) d)

30 µm

Figure 10. Sample 3 a) Surface Profile. Absorption amplitude at wavelengths b) 1650 cm-1 c) 1726 cm-1 d) 1752 cm-1. Scale bar applies to all maps. 195

channel is no lower than in the surrounding areas (which b) a) 25 µm would indicate a chemical change or poor contact); rather, it matches very closely (the channel cannot be distinguished in 100 µm Figure 13b-12f), suggesting that they are chemically similar. This channel is distinct in having a larger width-depth ratio, allowing easier access for the AFM probe. This suggests that the material observed within unexposed channels is c) 50 µm entirely removed from the exposed channels by the preparation process. The chemistry above the unexposed section is similar to the previous cases. The locally enhanced absorbance in Figure 13b (1604 cm-1) is indicative of the presence of C=C bonds formed within the channels but could also be due to primary amine (N-H) bonds. Figure 13c (1656 cm-1) may also be due to C=C bonds, however C=O and C=N (imine) Figure 11: Epoxy sample with an exposed channel. a) Optical Image bonds can also be absorbed at this wavelength making it (bottom-lit); channel structure visible below surface. b) Optical Image difficult to assign with certainty. Aromatics are also (top-lit); channel exposed at the surface. c) Selected AFM-IR spectra of channel and surrounding areas (normalized to 1504 cm-1 peak). Inset: absorbent at this wavenumber, however the maps are -1 AFM-IR height profile displaying location of spectra. normalized against a different aromatic band (1504 cm ) meaning that variations in absorbance here are unlikely to be Sample 4 – 100 µm from the tree tip in a 1600 µm tree due to aromatics. Absorption changes in Figure 13d, e and f Sample 4 shown in Figure 11, considers a channel from are considered to be due to carbonyls, which appear to have the same tree as sample 2. A 20 µm length has been exposed consistently formed in large amounts across all samples. The at a point ~100 µm from the needle tip, including a section 1742 cm-1 absorption, which we have assigned to aldehydes, at which branching has occurred. A further channel is also is the most prominent and well-defined absorption in this partially exposed. region, and this matches well with the spectra which The spectra of the channel are shown in Figure 11c. There displayed a peak in this region. -1 is a prominent 1740 cm peak, corresponding to carbonyl These results can be compared to those of sample 2 from (probably aldehyde) groups formed within the unexposed the same tree. The distribution of the chemical products is sections of the channel (black spectra). As in sample 1, there very different. In sample 2, the chemical products extend far are no degradation products within the exposed sections of beyond the channel walls and into the bulk of the material the channel (blue spectra). This is clearly different from the whereas here (Figure 12) the products are contained entirely unexposed areas (black spectra). Another similarity with within the channel walls. sample 1 is that the degradation does not appear to extend beyond the channel walls with spectra close to the channel 4 DISCUSSION and further away appearing largely the same. 4.1 SAMPLE PREPARATION A clear region of absorption is seen in the right side of the maps in Figure 13. Figure 12 (the 1742 cm-1 absorption map As discussed in Section 2 a grinding and polishing technique overlaid on an optical image) shows that this corresponds was used in conjunction with optical imaging to track the closely to the unexposed sections of the channel, with position of the channels at or near the surface and to produce exposed sections again showing no increased signal. In a smooth surface. This paper has demonstrated the efficacy contrast to samples 1 and 3, absorption detected inside this of this method in a variety of cases; however there are a number of findings which must be considered when using this technique. In samples 1 and 4 a difference can clearly 25 µm be seen between exposed and unexposed sections of the tree channels. The exposed sections appear to have lost their degraded products during the polishing process. This prevents us from observing how degraded products are distributed within and around the channel in such sections but also provides insight into the nature of the degraded material, which appears to be only weakly bound to the surrounding epoxy. The AFM-IR technique struggles with obtaining readings Figure 12. Background: Optical Image of a tree channel with an overlaid on rough surfaces. The polishing technique helps to produce -1 AFM-IR absorption map. (1742 cm ). smooth surfaces, however exposed channels will necessarily

a) b c) d) e f) 50 µm

Figure 13. Sample 4 a) Surface Profile. Absorption amplitude at wavelengths: b) 1604 cm-1 c) 1656 cm-1 d) 1702 cm-1 e) 1726 cm-1 f) 1742 cm-1. Scale bar applies to all spectral maps. 196

produce areas which are uneven and while steps are taken to account for this, it can introduce error into the readings at those points. In sample 2 these issues were managed by leaving the channel largely below the surface but sufficiently close to the surface to be studied. This also Figure 15. The Norrish Type 2 Reaction, the mechanism for carbonyl groups helped to solve the issue of material being removed due to to form alkene and enol groups. This occurs first through the hydrogen abstraction on the aldehyde or ketone group. the polishing process. molecular groups can be found in [15]. 4.2 MATURE CHANNELS The chemical groups identified in this paper are largely Three mature channels were studied which are visually consistent with those identified in previous studies of similar, both in terms of color and channel width, however electrical treeing, including those on trees formed in they present with very different chemical signatures. polyethylene [7]. The appearance of carbonyl groups (C=O), Samples 1 and 4 show no evidence of degradation products previously noted in both tree growth [6] and initiation [7], is beyond their channel walls; these were both around 100 µm consistently observed in this study. In particular the -1 from the needle tip. However sample 2, at ~500 µm from the aldehyde bond (peak centred at 1742 cm ) is prominent, this needle tip, shows significant amounts of degradation is particularly clear in Figure 11c, although the band is products within the bulk. This contrast could be due to the broad, and corresponds to other carbonyl peaks such as difference in distance of the channels from the needle tip. ketones and esters. These are identified within all channels This would be expected to change the field strength and except for sample 4, the youngest of the channels which had discharge magnitudes experienced within the channels. not been exposed to many discharges. There is a consistent -1 Though the channels appear superficially similar, they peak at around 1660 cm , associated with C=C bonds, as represent different stages of treeing. identified in previous studies in polyethylene trees. Sample 2 shows degradation products concentrated around However, in epoxy it is possible that this corresponds to an both of the branching points considered. This suggests that imine peak (C=N bond) which is also active at this greater volumes of degradation products are formed during wavelength. the branching process, indicating a difference in process An important note is that in sample 2, there is no clear compared to typical channel formation. distinction in the chemical maps (Figure 7) or in the spectra Despite the evidence that degradation products are (Figure 6) for areas within or outside the channel. This removed from exposed channels during the polishing suggests a common origin for these materials and may process, Figure 13 suggests that a significant contrast exists indicate that they are not formed outside of the channel, but between treed areas and the bulk epoxy. The difference formed inside and then permeated beyond the channel walls. between these appears binary; there is no evidence of an This evidence supports findings of previous papers which intermediate, partially degraded region existing between the have proposed oxidation as an early degradation mechanism undamaged bulk epoxy and the totally degraded material as well as β-chain scission as a chemical pathway [6]. This within the channels, which appears to be entirely a channel would be expected to occur via alkoxy radicals as shown in surface phenomenon. Figure 14. Also possibly active is the Norrish Type II reaction of Figure 15, via which aldehydes form alkenes and 4.3 NEWLY INITIATED TREE ketones. As discussed by Hu et al [6] these reactions could In sample 3 a channel has been studied which has only just be both active due to hot electron impacts from partial undergone initiation, and has been exposed to very limited discharge and due to UV light emissions from charge PD activity. The 20 µm long feature did not have any recombination. The chemical mapping properties of AFM- observed degradation products within or near the channel. IR however allow degradation due to UV light emission to This was true for both the exposed and unexposed sections be confidently ruled out as a significant factor in tree of the channel. Electrical tree initiation has previously been growth, at least closer (~100 µm) to the needle tip. UV shown to be associated with chemical changes where a emissions would be expected to penetrate into the bulk of ‘degraded region’ formed around the needle point prior to the epoxy significantly and cause degradation beyond the initiation [17]. In this sample such an area was identified channel walls, which was not observed in samples 1 and 4. (Figure 10b), however the channel extends far beyond this It is considered most likely from this study then that the region. There were minimal PDs detected during initiation chemical degradation occurs via oxidative and β-chain and there is a lack of evidence for chemical degradation scission pathways as well as via the Norrish Type II associated with the channel. This is indicative of the channel reaction, with energy being directly transferred to the epoxy initiation not being discharge driven [19] but of involving an molecules via hot electron impact due to partial discharge. electro-mechanical fracture mechanism [18]. Other reactions are possible, in particular if the 1660 cm-1 peak has been wrongly attributed to the C=C bond and is 4.4 TREEING CHEMISTRY instead imine based. Alternative degradation mechanisms A table of peak position wavenumbers alongside assigned which could be active include imine hydrolysis [19] and ozonolysis [20]. The primary reason for these being considered less probable than the chain scission and Norrish Type II pathways is due to the apparent similarities between polyethylene and epoxy based treeing.

Figure 14. Diagram showing potential β-Chain Scission degradation 5 CONCLUSIONS

197

The spectral signature associated with the chemical [7] N. Shimizu, K. Uchida, and S. Rasikawan, "Electrical tree and degradation of epoxies during electrical treeing has been deteriorated region in polyethylene," IEEE Trans. Electr. Insul., vol. 27, no. 3, pp. 513-518, 1992. successfully identified using AFM-IR. This has allowed [8] X. Chen et al, "Effect of tree channel conductivity on electrical tree identification of the precise location and extent of chemical shape and breakdown in XLPE cable insulation samples," IEEE Trans. degradation in unprecedented detail with 50 nm resolution. Dielectr. Electr. Insul, vol. 18, no. 3, pp. 847-860, 2011. A grinding method was used to prepare the samples, [9] N. Shimizu et al, "The space charge behavior and luminescence phenomena in polymers at 77 k," IEEE Trans. Electr. Insul., no. 5, pp. bringing the targeted channels to the surface ensuring a 256-263, 1979. smooth surface for testing; however, it has been found [10] K. Kaminaga et al, "The mechanism of degradation of polyethylene however that this method removes degraded material from in a high electrical field," Annu. Rep. Conf. Electr. Insul. Dielectr. exposed channels. This issue was overcome by not fully Phenom. (CEIDP), 1993, pp. 666-671. [11] G. Bacquet et al., "ESR study of free radicals in electrical trees in exposing the tree, but bringing it close enough to the surface polyethylene," IEEE Trans. Electr. Insul., no. 3, pp. 157-163, 1978. that the AFM-IR was still able to see within the channel. [12] A. Hammiche et al, "Photothermal FT-IR spectroscopy: A step A short tree examined immediately after initiation showed towards FT-IR microscopy at a resolution better than the diffraction no chemical degradation within or around the channel, limit," Applied spectroscopy, vol. 53, no. 7, pp. 810-815, 1999. [13] H. McDonald, P. D. Bastidas, S. Rowland, and S. Morsch, "Chemical indicating that its growth was not a chemically active Analysis of Solid Insulation Degradation using the AFM-IR process. This makes it unlikely that the earliest stages of tree Technique," in IEEE Int. Conf. Dielectr. (ICD), 2018, pp. 1-4. growth are driven by electrical discharges. Rather, [14] H. McDonald, S. Morsch, and S. M. Rowland, "High Resolution alternative mechanisms such as electromechanical fracturing Chemical Analysis of Electrical Trees through AFM-IR Spectroscopy," IEEE Conf. Electric. Insul. (CEI), 2019. must be considered more likely. [15] S. Morsch, P. D. Bastidas, and S. M. Rowland, "AFM-IR Insights into Mature channels, further from the needle tip and having the Chemistry of Interfacial Tracking," J. Mat. Chem. A, 2017. seemingly reached a steady state in terms of degradation, [16] J. Champion and S. Dodd, "Simulation of partial discharges in have also been studied. These showed that no singular conducting and non-conducting electrical tree structures," J. Phys. D: App. Phys., vol. 34, no. 8, p. 1235, 2001. chemical signature could describe electrical tree growth. [17] K. Uchida, H. Asai, N. Shimizu, and T. Takahashi, "Initiation Two of the samples studied showed no degradation products mechanism of electrical tree and deteriorated region in polymers," reaching beyond the channel walls, whilst the other showed IEEE Int. Conf. Prop. Appl. Dielectr. Mat. (ICPADM), 1991, pp. 240- extensive products in the epoxy bulk reaching as far as 5 µm 243. [18] H. Zeller and W. Schneider, "Electrofracture mechanics of dielectric from the channels. The reason for this variation is not yet aging," Journal Appl. Phys., vol. 56, no. 2, pp. 455-459, 1984. clear although some potential explanations are proposed in [19] T. Vidil et al, "Control of reactions and network structures of epoxy the discussion. Furthermore it is not determined whether thermosets," Progress in Polymer Science, vol. 62, pp. 126-179, 2016. degradation occurs beyond the walls or whether the material [20] N. M. Donahue, J. H. Kroll, J. G. Anderson, and K. L. Demerjian, "Direct observation of OH production from the ozonolysis of olefins," is first generated within the channels and then permeates Geophysical Research Letters, vol. 25, no. 1, pp. 59-62, 1998. outwards. Carbonyls are concentrated at both of the branching points Harry McDonald (S’18) is a Research Associate in studied in sample 2, suggesting higher levels of degradation the Dept. of Electrical and Electronic Engineering at during branching or accumulating at branch points. The University of Manchester. He received an MPhys in physics at University of St Andrews in 2014 and is Carbonyl groups, in particular aldehydes, have are currently a PhD student with the EPSRC CDT in identified in all mature channels. In addition alkene groups Power Networks. His research interests include have formed, consistent with the literature. These are electrical treeing and high voltage solid insulation believed to form through the chain scission pathway, whilst degradation. the Norrish Type II reaction is also possible. It is believed Suzanne Morsch is a Research Associate working in that these reactions are precipitated by hot electron impacts the AkzoNobel laboratory for Corrosion Protection in during discharges. the Dept. of Materials at The University of Manchester. She received an MChem from Newcastle University in ACKNOWLEDGMENT 2006, and completed her PhD in Surface Science at Durham University in 2012. Her research interests McDonald is thankful for the support of the ESPRC include the development of nanostructural through the Power Networks Centre for Doctoral Training. characterisation techniques such as AFM-IR and the degradation of network polymers. Morsch is grateful to AkzoNobel for financial support. Simon M. Rowland (F ‘14) completed a B.Sc. in REFERENCES physics at The University of East Anglia, and his PhD [1] I. Iddrissu, Z. Lv, and S. M. Rowland, "The dynamic character of at Chelsea College, London University. He has worked partial discharge in epoxy resin at different stages of treeing," IEEE for many years on dielectrics and their applications. He Int. Conf. Dielctr. (ICD), 2016 vol. 2, pp. 728-731 worked in multinational companies prior to joining [2] M. Olyphant, "Breakdown by treeing in epoxy resins," IEEE Trans. The Dept. of Electrical and Electronic Engineering in Power App. and Syst. vol. 82, no. 69, pp. 1106-1112, 1963. The University of Manchester in 2003. He was [3] M. Kosaki, N. Shimizu, and K. Horii, "Treeing of polyethylene at appointed Professor of Electrical Materials in 2009. He 77K," IEEE Trans. Electr. Insul, no. 1, pp. 40-45, 1977. was President of the IEEE Dielectric and Electrical Insulation Society in [4] F. Noto and N. Yoshimura, "Voltage and frequency dependence of 2011 and 2012. tree growth in polyethylene," Annu. Rep. Conf. Electr. Insul. Dielectr. Phenom. (CEIDP), 1974, pp. 207-217. [5] A. Vaughan, I. Hosier, S. J. Dodd, and S. Sutton, "On the structure and chemistry of electrical trees in polyethylene," J. Phys. D: App. Phys., vol. 39, no. 5, p. 962, 2006. [6] L. Hu, Y. Xu, X. Huo, and Y. Liao, "Chemical component analysis of electrical treeing in polyethylene by micro-infrared spectroscopy," IEEE Trans. Dielectr. Electr. Insul., vol. 23, no. 2, pp. 738-747, 2016. 198

Chemical Analysis of Solid Insulation Degradation using the AFM-IR Technique

Harry McDonald, Pablo D. Bastidas, Simon Rowland Suzanne Morsch School of Electrical and Electronic Engineering Corrosion and Protection Centre, School of Materials The University of Manchester The University of Manchester Manchester, United Kingdom Manchester, United Kingdom E-mail: [email protected]

Abstract—To enable the continued development of power In order to study such localised phenomena more powerful transmission cabling, an understanding of the processes which investigative techniques are needed. While previous techniques result in their failures is essential. In order to do so powerful have allowed the limited studying of the degradation they lack analysis techniques are required, however those which consider the sensitivity and resolution to sufficiently measure and chemical degradation are lagging far behind those of visible localize chemical damage to the same extent as visible degradation. This paper presents the AFM-IR chemical analysis degradation. Methylene Blue dyeing has been utilised by technique, able to conduct surface level chemical analysis with Shimizu et al in the identification and study of the degraded resolutions on the scale of 50 nm across the infrared spectrum. region formed during low exposure to oxygen [6] however as a Two cases are considered; that of interfacial tracking between dye it could not provide a full chemical characterisation of the epoxy and silicone rubber, and the degraded region formed region. Fourier transform infrared spectroscopy (FTIR) and around a defect in the tree initiation process. The results obtained using AFM-IR are compared to the outcomes from Raman spectroscopy were used by Uchida et al with the same other techniques. It is found that this offers a unique and aim [8], identifying an increase in C=C and C=O bonds powerful insight into visible and non-visible degradation of solid however these have resolutions approaching 10 µm due to the dielectrics. diffraction limit of IR light, greatly limiting the available resolution. Similarly Hu et al [7] used micro-infrared Keywords—Electrical treeing, interfactial tracking, epoxy, spectroscopy in the chemical study of tree channels. Further AFM-IR, degradation, chemical analysis methods have been used in the study of treeing and tracking, such as electron spin resonance (ESR) and X-ray Photon I. INTRODUCTION Spectroscopy (XPS) [9], however none have been able to Polymeric materials such as XLPE are increasingly used in combine the resolving capabilities used in visible light with the cabling owing to their excellent breakdown strengths and ability to clearly identify and distinguish across the range of dielectric properties. They are however vulnerable to a number molecular damage. of degradation mechanisms, among the most common of which Analysis methods with a resolution in excess of a channel are electrical treeing and interfacial tracking. There has been width are not sufficient to accurately study the chemical extensive recent work in the study of the degradation of degradation process, which is highly inhomogeneous. Instead dielectrics, particularly with the visible aspects [2] with an aim they capture the bulk processes whilst failing to distinguish of reducing cabling costs while increasing their lifetimes. localised processes. AFM-IR with an X-Y resolution of 50 nm However the investigative tools for non-visible damage, such is able to capture such variation in great detail and here the as chemical degradation, lack the same high resolution. This method is presented with the NanoIR2 system (Anasys has limited the available evidence to guide the development of Instruments). Analysed were an interfacial tracking channel theories and models for the degradation processes. This paper [1] and the degraded region around a high field defect. will present a new method for the investigation of chemical degradation called Atomic Force Microscopy- Infrared II. AFM-IR Spectroscopy (AFM-IR) which can resolve down to 50 nm. The AFM technique was developed in 1982 by IBM. This A degraded region has been found to develop around allowed for topographical analysis at resolutions beyond the defects at high fields and this is generally taken to be the nanometer, and the later integration of infrared spectroscopy by precursory step before tree and track initiation [3-5]. The study Alexander Dazzi [10] supported surface level chemical of this has however been largely limited to low oxygen analysis. Utilising an AFM probe cantilever, this is traced environments [6] where it is more pronounced and much about across the surface and subtle deflections due to height its formation and how it leads to channel inception is not yet variations can be tracked to produce a 3D profile of the surface. known. Similarly degradation occurs within and around A pulsed tuneable top-down IR laser is used for the channels during their growth [7], however again the chemical spectroscopy, when the wavelength of the beam corresponds to pathways involved are not yet known, limiting the the excitation energy of chemical bond vibrations, this is understanding of the physical processes acting and the accompanied by transient thermal expansion of the specimen. development of methods to increase insulation resistance. This expansion is then registered by the AFM probe and

199

Fig. 1: AFM-IR experiment with top-down illumination. An IR source is pulsed, inducing rapid thermal expansion of the sample, this is detected by deflection of the AFM probe cantilever. The recorded AFM-IR signal corresponds to the amplitude following fast Fourier transform of the deflection signal. [1] indicates the presence of such chemical bonds in that location. silicone layer. As the tracks are surface degradation this allows This is done a number of times to allow for an average to be their investigation by the AFM-IR technique. taken in that location and can be used to view the absorbance spectrum across the infrared range in that position, or to build B. Electrical Treeing Samples high resolution chemical density mapping at a target The area around the excitation point (a needle tip) in a wavelength. This process is illustrated in Fig. 1. sample immediately following initiation was of interest in this experiment. Using the same Araldite and Aradur composition, Whilst typical spectrographical methods are resolution degassed for 70 minutes, cured and post-cured similarly using a limited with the diffraction limit, AFM-IR is able to bypass this 25 mm x25 mm acrylic cube. Partway through the degassing by the use of an AFM probe. This allows the level of process a needle (Ogura, 3 µm tip radius) was inserted into the absorption across the surface area contacting the probe to be epoxy, with a 2mm gap to the bottom. This creates a point- measured, approximately 50 nm in this experiment. Through plane configuration with an AC voltage applied to the needle this method the diffraction limit can be overcome and highly tip. This voltage was steadily increased until initiation occurred localised spectra can be taken. In samples in which large (from 8.52 kV to 10.35 kV RMS over 19 minutes), on initiation variations in chemical composition can be expected over small the voltage was stopped. This was monitored visually and areas this is a valuable tool and can provide crucial context to electrically as with the tracking samples. other chemical analysis methods. The AFM-IR technique has had previously demonstrated capability in the chemical The area of interest, the region around the needle tip, here is characterisation of epoxies [1, 11]. within the bulk of the material, in contrast to tracking along the surface, this must first be exposed. In order to do this a cutting machine was used to remove most of the material and the needle itself was then withdrawn. The samples were then III. METHODOLOGY carefully ground down until the desired point was exposed. A. Interfacial Tracking Samples This was done using polishing paper and a diamond compound Interfacial tracks were generated between epoxy and paste (3 µm, 1 µm and 1/4 µm) with regular visual inspections vulcanised silicone rubber layers. The epoxy was prepared with through a microscope to ensure the correct point was reached Araldite-LY5052 and Aradur-HY5052 (Huntsman) in a 100:38 before chemical investigation with the AFM-IR. by weight mix. This was degassed for 50 minutes and cured in C. AFM-IR a 150 mm x 50 mm x 5 mm (L x W x D) plaque for 24 hours at These samples were then analysed using the AFM-IR room temperature, then post-cured at 100oC for 4 hours. The system, NanoIR2 (Anasys Instruments) in contact mode, the silicone rubber layer (Polymax) was used as received. probe being traced along the surface. Scan rates of 0.04 Hz for A 45 µm thick stainless-steel wire was used as an electrode the tracking samples and 0.08 Hz for the treeing samples were defect, curved into a circle of diameter 1.2 mm, this was placed used, with a gold-coated silicon nitride probe to produce the between the epoxy and silicone rubber plaques with the air at chemical map images. With each excitation allowing a the interface removed [12]. An AC voltage of 42 kV RMS was measurement, for the tracking samples 32 co-averages were applied to the steel wire for 1020 minutes, interfacial tracks used for each point, of which there were 600 for each of 300 were generated here and monitored visually by camera and scan lines. For treeing samples, there were 16 co-averages for electrically by OMICRON MPD600 for partial discharge each point with 1024 points per line over 300 scan lines. Wider readings. The epoxy surface was then exposed by removing the IR band spectra were produced at various points on the

200

Fig. 3: 80 µm x 80 µmAFM-IR images of the epoxy resin surface of sample A after 1020 Fig. 2: AFM-IR spectra corresponding to tip locations minutes of electrical aging: (a) contact mode height image, and maps of the amplitude indicated by markers on the contact mode height image signal induced by pulsed infrared illumination at (b) 1080 cm-1; (c) 1248 cm-1; (d) 1336 cm-1; (inset) of epoxy resin after 1020 minutes of electrical (e) 1456 cm-1; (f) 1505 cm-1; (g) 1660 cm-1, and (h) 1752 cm-1. [1] stress. Normalised to 1505 cm-1 peak. Dotted line shows ATR-FTIR Spectrum [1] samples, just outside of the channels (green dots) whilst the signal for each of these 1024 co-averages were used for each value. from within the channels is more similar to those further from the channels (red dots). This could indicate either a IV. RESULTS AND DISCUSSION mechanism by which channels are able to widen, through In each case wideband spectral sweeps were performed at degrading peripheral areas, or the transportation of areas considered interesting, such that they could be compared degradation products following their generation. to identify wavelengths, and potentially molecular densities,  The red dots further from the channels still showed signs of which were worth further investigation. This was followed by degradation, in particular it was noted that C=O and N-O the production of chemical maps at specific wavelengths to see bond formation was indicated. That this occurs away from the variation in the absorption levels across the area. the channels indicates damage does occur at a range in A. Interfacial Tracking Samples seemingly intact areas. Full spectral sweeps were performed within and around the The AFM-IR spectra are compared to an ATR-FTIR tracking channels, shown in Fig. 2 is the spectrum after 1020 spectrum (dotted line) from the same sample and this minutes of aging at these different locations. The absorbance demonstrates the limitation of low resolution, wider volume for each spectrum was normalized to 1505 cm-1. It reveals that scanning techniques. Here not only is much of the the degradation products found inside of the channel walls, and inhomogeneity of the sample lost such as the differences in in the micrometres beyond it, have distinct differences. chemical compositions within and around the channels but it is also apparent that -the ATR-FTIR measurements were  The strongest, most apparent signal was that of the peak at impacted by degradation products from below the surface, 1752 cm-1, attributed to the C=O stretch bonds. These were beyond tracking channels. It is through using higher resolution strongest within the channels (blue dots). This and a peak at methods in conjunction (such as the AFM-IR) that such 1248 cm-1 (C-O stretch) were taken as an indication of a differences can be discerned. large of concentration of esters within the channel.

-1 Following this, to better observe the changes across an area,  A distinct peak at 1660 cm differed in that it was strongest chemical map images were produced at target wavelengths as

Fig. 4: AFM-IR spectra corresponding to tip locations indicated by markers on the contact mode height image (right) of epoxy resin following the initiation of an electrical tree from needle tip. Spectra are normalised to 1600 cm-1 peak.

201

interfacial tracking and electrical treeing is discussed, looking at the visible, macroscopic tracking channels and the non- visible degraded region formed around a needle defect following tree initiation. With a resolution at submicron levels, the extent to which non-visible degradation may reach into the bulk material from a defect or channel can be observed, while also allowing details within this region to be clearly resolved far beyond the capabilities realized with previous techniques. This can be seen both through the wide-band spectra taken at key points and through the generation of chemical map images over an area. As well as this the low sampling depth ensures that chemical signals are taken from the intended locations, minimising signals from elsewhere. Here providing necessary context to spectra produced using the ATR-FTIR technique, which penetrates further below the surface. Fig. 5: 30 µm x 30 µm AFM-IR images of the epoxy resin surface ACKNOWLEDGEMENTS can be seen in Fig.3. These further confirmed the previous observations of ester carbonyl (C=O) absorbance increasing H. McDonald is thankful for the financial support of the within the tracks and while actually reducing peripheral to ESPRC through the Power Networks Centre for Doctoral them. Alkene absorbance (C=C) was found to increase around Training. S. Morsch is grateful to AkzoNobel for financial the outside of tracks but not inside of them. Further the support. P. Bastidas is grateful to EPSRC and The University generally decreased levels of C-O, O-H and N-O bonds of Manchester for a DTA grant for financial support. indicated that these were the bonds in the epoxy being damaged during the degradation process. REFERENCES [1] S. Morsch, P. D. Bastidas, and S. M. Rowland, "AFM-IR Insights into B. Electrical Treeing Samples the Chemistry of Interfacial Tracking," Journal of Materials Chemistry A similar process was performed on the electrical treeing A,vol 5, pp.24508-24517, 2017. samples, looking at the degraded region beyond the needle tip [2] R. Schurch Brandt, "Three-Dimensional Imaging and Analysis of immediately following initiation. The section analysed in this Electrical Trees," 2014. sample did not include the tree channel. An initial spectral [3] N. Shimizu, H. Katsukawa, M. Miyauchi, M. Kosaki, and K. Horii, "The space charge behavior and luminescence phenomena in polymers at 77 sweep, shown in Fig. 4 revealed a distinct increase in carbonyl k," IEEE Trans on Electrical Insulation, no. 5, pp. 256-263, 1979. (C=0) bonds in areas close to the needle tip, which is indicative [4] S. Bamji, A. Bulinski, and R. Densley, "Threshold voltage of of oxidation during the degradation process. As well as this, luminescence and electrical tree inception in low‐density polyethylene," lower amounts of alkene (C=C) absorption were detected. Journal of applied physics, vol. 63, no. 12, pp. 5841-5845, 1988. [5] N. Shimizu, K. Uchida, and S. Rasikawan, "Electrical tree and Chemical maps were again produced to view wavelengths deteriorated region in polyethylene," Electrical Insulation, IEEE -1 of interest. It can be seen from Fig. 5, looking at the 1708 cm Transactions on, vol. 27, no. 3, pp. 513-518, 1992. wavenumber, which can be considered an indicator of the [6] N. Shimizu and K. Horii, "The effect of absorbed oxygen on electrical presence of C=O bonds, that the degraded region extends to treeing in polymers," IEEE transactions on electrical insulation, no. 3, slightly below 10 µm (roughly 8 µm) from the needle tip. pp. 561-566, 1985. [7] L. Hu, Y. Xu, X. Huo, and Y. Liao, "Chemical component analysis of The observations here correlate well with previous findings electrical treeing in polyethylene by micro-infrared spectroscopy," IEEE on the degraded region, providing further confidence in the Trans. on Dielectrics and Elec. Ins, vol. 23, no. 2, pp. 738-747, 2016. results and the method. Previous tests had indicated a region [8] K. Uchida, H. Asai, N. Shimizu, and T. Takahashi, "Initiation with greater densities of oxidised bonds would form following mechanism of electrical tree and deteriorated region in polymers," in tests performed in low levels of oxygen. The size of this region Properties and Applications of Dielectric Materials, 1991., Proceedings was typically found to be around 10 µm. The use of the AFM- of the 3rd International Conference on, 1991, pp. 240-243: IEEE. [9] K. Kaminaga, T. Suzuki, T. Uozumi, T. Haga, N. Yasuda, and T. Fukui, IR technique, with high sensitivity and resolution, is here "The mechanism of degradation of polyethylene in a high electrical demonstrated in performing a detailed chemical analysis of this field," in Electrical Insulation and Dielectric Phenomena, 1993. Annual region in a treeing sample immediately following initiation. Report., Conference on, 1993, pp. 666-671: IEEE. This demonstrates the existence of such a region under oxygen [10] A. Dazzi, C. B. Prater, Q. Hu, D. B. Chase, J. F. Rabolt, and C. Marcott, rich conditions following tree initiation, and allows a detailed "AFM–IR: combining atomic force microscopy and infrared chemical analysis to be performed across the infrared spectrum spectroscopy for nanoscale chemical characterization," Applied and with high resolution. Spectroscopy, vol. 66, no. 12, pp. 1365-1384, 2012. [11] S. Morsch, Y. Liu, S. B. Lyon, and S. R. Gibbon, "Insights into epoxy V. CONCLUSION network nanostructural heterogeneity using AFM-IR," ACS applied materials & interfaces, vol. 8, no. 1, pp. 959-966, 2015. The capability of the AFM-IR in the study of degradation in [12] P. D. Bastidas and S. M. Rowland, "Interfacial aging in composite solid dielectrics is well demonstrated, both as a stand-alone insulators as a result of partial discharge activity," in Electrical technique and in providing a worthwhile complement to other Insulation Conference (EIC), 2017 IEEE, 2017, pp. 13-16: IEEE. analysis tools. The application of this technique to both

202

High Resolution Chemical Analysis of Electrical Trees through AFM-IR Spectroscopy

Harry McDonald, Simon Rowland School of Electrical and Electronic Engineering Suzanne Morsch The University of Manchester Corrosion and Protection Centre, School of Materials Manchester, United Kingdom The University of Manchester E-mail: [email protected] Manchester, United Kingdom

Abstract—Accurate modelling of electrical tree growth is branching patterns [2]. In some cases these will change the dependent upon a physical and chemical understanding of the chemical nature of the channels. Some channels are considered insulation and degradation processes on the scale of the tree to be ‘conductive’ as carbon forms upon the walls [3], yet branches. While imaging techniques for the physical elements of others will be ‘non-conductive’ without such carbon deposits tree growth have improved in recent years, those for the [4]. Theories intending to describe tree propagation must be chemical regime are lagging behind. In this paper AFM-IR able to explain such differences. Electrical treeing is typically (Atomic Force Microscopy - Infrared Spectroscopy) is applied to assumed to be the result of localised degradations, whether this a non-conducting tree channel grown in epoxy resin. This is due to partial discharge [5], electro-mechanical cracking [6] provides chemical analysis with a spatial resolution of 50 nm. or chemical degradation [7]. In the development of such The distinct chemistries from within buried channels, exposed theories, understanding of the material condition at resolutions channels and the epoxy bulk are revealed for the first time. beyond that of an individual channel is of utmost concern. Keywords—Electrical treeing, epoxy, AFM-IR, degradation, B. Epoxy chemical analysis, imaging Epoxy exhibits good dielectric properties, combined with I. INTRODUCTION optical transparency and a relatively easy curing process. It is also a well-understood material, having been the subject of A. Electrical Treeing Background significant studies into degradation processes under different Electrical tree channels are gaseous tubules which develop stresses. This makes it an ideal material for studying electrical in polymeric dielectrics under high voltages. Microns in treeing. Typically, treeing studies are in epoxy or polyethylene diameter, they form in power cable insulation over the course and it is important to appreciate both the similarities and of years. They originate due to defects such as protrusions, differences between these materials. Both are subject to the contaminants or voids. Once formed, these channels grow in a treeing phenomenon, which may appear on first inspection to tree-like structure (see Figure 1) until they bridge the insulation manifest with similar properties. However there are significant and breakdown occurs. The properties of these structures are physical and chemical differences between trees which are highly dependent upon the material and conditions in which superficially similar. Differences include channel size, tree they develop. This makes the diagnosis of the underlying geometry, growth rate and level of carbonisation. causes of electrical treeing a complex task. In most cases, channel growth is accompanied by, and apparently driven by C. Imaging Work large partial discharges within the channels. However, they Techniques such as Scanning and Tunnelling Electron have also been observed to grow without observable Microscopy (SEM & TEM) provide detail of degradation on discharges. Channel structure can differ dependent on factors the orders of nanometres while Serial Block Face Scanning such as voltage, frequency and temperature, and small Electron Microscopy (SBFSEM) (Figure 1) and X-ray differences can produce large changes in channel widths and Coherence Tomography (XCT) allow 3-dimensional reconstructions of the visible elements of electrical tree degradation [8]. Our ability to study the chemical factors associated with degradation is less complete. This paper provides insight into the nature of the chemical degradation associated with tree channels. For the first time, chemical maps with a resolution of approximately 50 nm are obtained by application of Atomic Force Microscopy with Infrared Spectroscopy (AFM-IR) to electrical tree channels. Previous studies of chemical ageing within and around tree Figure 1: Three-dimensional reconstruction of an electrical tree using channels have insufficient resolution to allow detailed SBFSEM [7] localisation of chemical bond types, and have limited the

understanding which can be developed. Hu et al [9] utilised Micro-FTIR to study the chemical products formed during the treeing process in polyethylene, identifying C=O, C-OH and C- O-C bonds in particular. However the resolution of this technique, at around 15 µm, prevents the accurate localisation of these degradation products. It can be seen that they form, but the context of location is lost. Vaughan et al [3] utilised Raman Spectroscopy. They found that channel walls contain a layer of graphitic carbon along with evidence of ‘background’ degradation throughout the bulk of the material, believed to emanate from the channels. Chen et al [4] studied a number of channel formations using this technique, aiming to understand why different structures form. They identified the presence of graphitic carbon as an influence on tree structure. While Figure 2: AFM-IR with top-down illumination. An IR source is pulsed, Raman spectroscopy provides higher resolutions than Micro- inducing rapid thermal expansion of the sample. This is detected by FTIR (~2 µm), this is significantly less than those achieved deflection of the AFM probe cantilever. The AFM-IR trace is recovered from fast Fourier transform of the deflection amplitude signal. [1] with AFM-IR and the technique struggles to accurately locate degradation products within and around channels. Furthermore, tree readily grows. The sample production method including Raman spectroscopy and IR absorbance methods are often mixing, curing and post-curing has been detailed in [11]. considered complementary, typically bonds are active (therefore visible) using only one of these techniques, and An AC voltage was applied to the needle tip which is 2 mm cannot be seen in the other. This means that full chemical from the grounded surface. Tree growth was monitored characterisation will typically require both methods. optically by a CCD camera. A voltage of 17.7 kVRMS was initially applied to initiate the tree and then dropped to 8.85 D. AFM-IR kVRMS for controlled growth. The voltage was removed before First developed in 1999 by Hammiche et al [10], this the tree reached the bottom surface. This is a standard method technique allows chemical characterisation with spatial for tree growth, chosen such that the chemical degradation resolutions of tens of nanometres. Whilst most spectrographic products obtained are representative of typical behaviour. techniques are diffraction limited, used in combination with B. Sample Polishing atomic force microscopy, this method is able to far exceed it. Samples were initially inspected under a microscope to An AFM probe is attached to a cantilever. When traced identify key areas of interest, and then roughly cut to access across a surface, the probe is able to detect minor deflections these areas. The needle is removed at this stage. The target area and build a 3D surface profile. A broadband, pulsed tuneable is then approached by polishing, first using Silicon Carbide laser is then used to illuminate the surface under the probe, and grinding/polishing paper and then a water soluble diamond scans across the IR spectrum. When wavelengths compound paste (Spectrographic Ltd) containing crystals of corresponding to the excitation energy of a constituent bond defined sizes (3 µm, 1 µm and 0.25 µm). Throughout this are incident upon a molecule, energy is absorbed and the bond process the sample is inspected by optical microscope using a is excited. As the surface below the probe expands in response combination of top and bottom lighting to ensure the target to the laser pulse, this causes the probe to be ‘kicked’ and to area is reached. This is shown in Figure 3a. vibrate. These vibrations are recorded, as shown in Figure 2, and Fourier transformed to find the amplitude of the induced C.AFM-IR vibration. By this method the absorbance spectrum of a The sample surface was analysed using the NanoIR2 specific location can be measured, or the absorption at a target (Anasys Instruments) in contact mode with a gold-coated frequency across an area can be mapped. Both methods are silicon nitride probe. Absorbance spectra measurements were demonstrated in this paper. taken at different locations around the exposed channel as detailed in Figure 3c. The spectra were taken between 900 cm-1 AFM-IR has previously been used in the study of electrical -1 treeing to examine chemical damage occurring during initiation and 1800 cm with 1024 co-averages for each point. The [11]. This paper now demonstrates the applicability of AFM-IR compositional mapping was performed at different to the tree channels themselves and provides insight into the wavenumbers across an area of 15 x 30 µm with a scan rate of chemical degradation occurring during their growth. 0.1 Hz and 16 co-averages per point. II. METHODOLOGY III. RESULTS The tree appears to be non-conductive, as evidenced by the A. Sample Production white reflective materials formed within the channels. While Araldite LY 5052 and Aradur HY 5052, both produced by finer channels form within this resin [12], a thicker channel Huntsman, are used. The epoxy is Novalak based; containing was chosen as the object to study as this is more typical of trees an ether diluent while the hardener contains a mixture of grown in many different materials. This gives a wider polyamines. This provides high thermal and chemical stability. relevance to the work, while also making the measurements Samples are moulded with an embedded needle from which a less vulnerable to errors in preparation.

204

the presence of C=O bonds. a) b) C. Chemical Composition Maps High resolution chemical composition maps were taken at 25 µm 50 µm a number of wavenumbers as shown in Figure 4. These show the spatial variation in absorbance across the sample, with red c) or orange colouration indicating higher absorbance. Relating absorbance at a particular wavelength to a specific functional group then forms a chemical map of this group. Figure 4 examines the location of a number of groups. Figure 4b) at 1664 cm-1 indicates unsaturated C=C bonds which are often associated with polymer ageing, though could suggest carboxylic acid formation. 4c) at 1702 cm-1, is indicative of ketone groups and 4d) at 1726 cm-1 is indicative of ester groups. Similar maps here suggest that there is no great difference in which locations the groups form. As observed in the spectra there is a region of high concentration Figure 3: Polished surface exposing a tree channel: a) Optical Image (bottom lit); sub-surface channel visible. b) Optical Image (top lit); reflective above the buried channel, yet minimal variation is seen around material visible within channels below the surface. c) Selected AFM-IR the exposed channel. Notably the absorption readings within spectra of channel and surrounding areas (normalised to 1504 cm-1 peak). the exposed section of the channel indicate a reduction in C=O Inset: AFM-IR height profile displaying location of spectra, 30 x 15 µm. bonds compared to the buried channels, and also to the surrounding areas. This is considered further in the discussion. A. Optical Imaging -1 The sample was optically imaged prior to chemical Looking outside of the C=O band, Figure 4e) at 1248 cm , analysis. In Figure 3a), bottom lighting shows a channel buried is indicative of the presence of C-O bonds in ethers. It should below the surface. In Figure 3b), top lighting makes it clear the be noted however that this region of the IR spectrum is channel emerges into the surface as the dark feature. complex, with many overlapping absorbences. This complicates identifying the contributions to a signal. Figure 4f) B. Spectra at 1604 cm-1 is associated with aromatic rings, however due to The absorption spectra are displayed in Figure 3c) the spectral proximity of the broad C=O and C=C bands, this alongside the AFM height profile; illustrating where these should be treated with caution. Figure 4g) at 3300 cm-1 is spectra were taken. Spectra denoted by red lines were taken in indicative of OH groups. Again there is an increase in density locations closest to the exposed channel, blue lines indicate above the buried channel and no clear variation with distance locations more distant, and spectra in green were taken at from the edges of the channel. Within the exposed section of locations above the buried channel. An increase in C=O the channel the detected concentration decreases. absorption is observed in the locations closest to the channel; however this difference is slight and requires further IV. DISCUSSION investigation. A significant increase is observed in areas above It is not intended in this paper to definitively assign each the unexposed channel (green spectra). The AFM-IR is largely map to a functional group, or to identify the exact mechanisms a surface detection system however as it detects thermal by which they form. To do so will require further expansion, material below the surface can influence results. measurements and confirmation via other analysis techniques. The IR absorbance from 1620-1800 cm-1 can be seen to have Nonetheless an attempt can be made to identify chemical greatly increased across the entire band. This is indicative of degradation products in some cases. Carbonyl groups (C=O) and hydroxyl (OH) are clearly

a) b) c) d)

e) f) g)

Figure 4: AFM-IR images of the surface around the channel. a) surface profile. IR Absorption amplitude at wavenumbers: b) 1664 cm-1, c) 1702 cm-1, d) 1726 cm-1, e) 1248 cm-1, f) 1604 cm-1, g) 3300 cm-1. Each map is ratioed against 1504 cm-1 (corresponding to aromatic peak). The images represent a 30 µm x 15 µm surface.

205

observed within a channel below the exposed surface. Due to demonstrated that graphitic carbon may form during treeing [3, the broad spectral excitation in the carbonyl region, we 9], although only within conductive channels. An important conclude that a range of carbonyl groups is being formed rather caveat to this is that the AFM-IR can have difficulties making than one particular functional group. Other chemical products measurements in locations with uneven surfaces, such as the are apparent, C-O and C=C are likely, but difficult to assign inside of a tree channel. This can distort absorption readings. with certainty. The results observed here are comparable to others obtained in polyethylene. Hu et al [9] using Micro-FTIR V. CONCLUSION to study treeing in LDPE identified C=O bonds as well as C- The suitability of AFM-IR to the study of chemical ageing OH and C-O-C bonds in aged areas. However they were of epoxy resin in and around electrical tree channels has been unable to localise them due to the resolution limitations of the demonstrated, providing insights impossible to find using other equipment. We can now see that these products were likely to means. It is identified that degradation products are retained on have formed primarily within the channels. These products are the surfaces of the tree channels, but do not appear to extend also notably similar to those formed with partial discharge beyond the channel walls. Work will continue and, as testing of epoxy in air by Hudon et al [13]. The comparability experience with the technique and interpretation improves, of chemical products in epoxy and polyethylene suggests that more detailed analysis of degradation in and around treeing similar processes may occur in each. channels can be expected. High numbers of C=O groups suggest oxidation. These ACKNOWLDEDGEMENTS typically form in the first stages of this process. This may occur McDonald is thankful for the financial support of the ESPRC via the chain scission mechanism. Here C-C and C-H bonds through the Power Networks Centre for Doctoral Training. are split by the high energy particles produced in partial Morsch is grateful to AkzoNobel for financial support. discharge. These generate the free radicals necessary for further reaction, initially interacting with oxygen within the REFERENCES channels or free volume. This is similar to observations in [1] S. Morsch, P. D. Bastidas, and S. M. Rowland, "AFM-IR Insights into surface discharge testing. This theory is supported by the lack the Chemistry of Interfacial Tracking," Jour. of Mat. Chem. A, 2017. of clear evidence of chemical degradation extending beyond [2] D. Auckland, J. Cooper, and B. Varlow, "Factors affecting electrical tree the channel walls. Spectrally there may have been a slight testing," in Science, Measurement and Technology, IEE Proceedings A, increase in C=O bond density close to the channel walls, 1992, vol. 139, no. 1: IET, pp. 9-13. however this was not conclusive. No ageing beyond the [3] A. Vaughan, I. Hosier, S. J. Dodd, and S. Sutton, "On the structure and channel walls was identified using the compositional mapping chemistry of electrical trees in polyethylene," Journal of Physics D: Applied Physics, vol. 39, no. 5, p. 962, 2006. of the AFM-IR. The formation of the further degradation [4] X. Chen, Y. Xu, X. Cao, S. Dodd, and L. Dissado, "Effect of tree products can happen via a number of mechanisms. Norrish channel conductivity on electrical tree shape and breakdown in XLPE Type II has been proposed [9] in treeing and the presence of cable insulation samples," IEEE Transactions on Dielectrics and ketones as well as C=C and OH bonds supports this. Electrical Insulation, vol. 18, no. 3, pp. 847-860, 2011. Alternatively hydrogen abstraction via alkoxy radicals has been [5] J. Champion, S. Dodd, and J. Alison, "The correlation between the proposed during interfacial tracking in epoxy [1]. partial discharge behaviour and the spatial and temporal development of electrical trees grown in an epoxy resin," Journal of Physics D: Applied Lack of damage beyond the channel walls is not surprising Physics, vol. 29, no. 10, p. 2689, 1996. considering that the process is similar to discharge on a [6] H. Zeller and W. Schneider, "Electrofracture mechanics of dielectric polymer surface in air. The result is however in contrast to aging," Journal of applied physics, vol. 56, no. 2, pp. 455-459, 1984. conclusions presented by Vaughan et al [3] who used Raman [7] S. J. Dodd, "A deterministic model for the growth of non-conducting Spectroscopy in the study of LDPE treeing. It was suggested electrical tree structures," Jour. of Phys. D: App. Phys., vol. 36, no. 2, p. there that fluorescence measured may be partially due to 129, 2002. chemical degradation extending beyond the channel walls. The [8] R. Schurch, S. M. Rowland, R. S. Bradley, and P. J. Withers, "Comparison and combination of imaging techniques for three results presented in this paper would suggest that either: the dimensional analysis of electrical trees," IEEE Transactions on range of this degradation is large with no great spectral change Dielectrics and Electrical Insulation, vol. 22, no. 2, pp. 709-719, 2015. as a function of distance from the channels, that it does not [9] L. Hu, Y. Xu, X. Huo, and Y. Liao, "Chemical component analysis of occur in all cases of electrical treeing, or that their findings electrical treeing in polyethylene by micro-infrared spectroscopy," IEEE were due to incidental sampling of channels below the surface. Trans. on Dielectrics and Elect. Insul., vol. 23, no. 2, pp. 738-747, 2016. [10] A. Hammiche, H. Pollock, M. Reading, M. Claybourn, P. Turner, and K. At each tested wavenumber a reduced absorption was Jewkes, "Photothermal FT-IR spectroscopy: A step towards FT-IR measured within the exposed section of the channel compared microscopy at a resolution better than the diffraction limit," Applied to the channel below the surface. This is also witnessed spectroscopy, vol. 53, no. 7, pp. 810-815, 1999. optically by the loss of reflective materials observed in [11] H. McDonald, P. D. Bastidas, S. Rowland, and S. Morsch, "Chemical channels below the surface when they are exposed. This Analysis of Solid Insulation Degradation using the AFM-IR Technique," IEEE 2nd International Conference on Dielectrics (ICD), 2018: pp. 1-4. indicates that these are materials weakly held on the surface of [12] I. Iddrissu, Z. Lv, and S. M. Rowland, "The dynamic character of partial channels, but do not necessarily represent degradation discharge in epoxy resin at different stages of treeing," IEEE extending into the solid polymer. The reduced signal on the International Conf on Dielectrics (ICD), 2016, vol. 2, pp. 728-731. exposed channel walls compared to the bulk polymer may be [13] C. Hudon, R. Bartnikas, and M. Wertheimer, "Chemical and physical indicative of the occurrence of carbonisation because C-C degradation effects on epoxy surfaces exposed to partial discharges," bands are at best weak in IR measurements. Raman testing has Proc of International Conf on Properties and Applications of Dielectric Materials, 1994, vol. 2, pp. 811-814.

206

A High-Resolution Study of Chemical Aging Prior to Electrical Tree Growth

H. McDonald, S. M. Rowland Suzanne Morsch Dept. of Electrical and Electronic Engineering Corrosion and Protection Centre, Dept. of Materials The University of Manchester The University of Manchester Manchester, United Kingdom Manchester, United Kingdom E-mail: [email protected]

Abstract- A study of chemical aging prior to electrical tree initiation has been largely formed by tests using methylene initiation has been undertaken using a high-resolution chemical blue [6], FTIR [7, 9] and Raman spectroscopy [7, 10]. These analysis technique, AFM-IR (atomic force microscopy-infrared techniques have shown a link between carbonyl formations spectroscopy). This technique has provided chemical and tree initiation [7], with carbonyls identified as forming a characterisation with a spatial resolution of around 50 nm. It is sphere of degradation around the needle tip. Similar tests have seen that the use of tests on this scale provide information and insight which was not previously observable using lower been performed upon tree channels after initiation, looking to resolution techniques. Previously unobserved degradation understand their chemical nature and the processes which products are observed in epoxy resin prior to tree initiation with underlie tree growth. ethers (C-O) and methyl groups (CH3) appearing to form Our understanding of the chemical processes involved in alongside carbonyls (C=O), each occurring in different locations tree formation and propagation is still limited however. around the needle tip. Such spatial resolution of aging chemistry Although these tests have advanced knowledge; the low will allow improved models of tree initiation and growth to be spatial resolutions of the techniques makes detailed developed. interpretation difficult. If we consider the scale of the most I. INTRODUCTION common techniques; FTIR has been used to study electrical Solid dielectrics under high voltage gradients degrade with trees with resolutions of around 10 µm [9], Raman has been time and this degradation, once sufficient in volume or in applied with a resolution that appears somewhat lower than quantity, can lead to localized or total breakdowns. This paper the anticipated 2 µm [10, 11]. These scales, when compared to considers electrical treeing, which is the formation and the common 1- 5 µm diameter channels, can be seen to make propagation of gaseous channels within solid polymer it impossible to gain detailed information about the locations dielectrics. These channels are typically 1-5 µm in diameter, in which chemical aging occurs around a tree channel. and are commonly accompanied by partial discharge signals. To analyze processes that occur on the scale of tree branch Asset failure can occur when these channels bridge the diameters, an ability to characterize at or below the insulation from one conductor to another. micrometre scale is required. This is necessary for the clear As a process, electrical tree growth develops in stages; interpretation of what has occurred and where it has happened. these include initial degradation, through growth, and Recent examples of this include Zhang’s [12] study of ultimately leading to the total breakdown and failure of the electroluminescence which utilized SEM energy dispersive insulation [1]. Understanding how this happens requires the spectrometry for analysis of chemical degradation in the analysis of these processes on physical size and time-scales region surrounding the needle electrode. This attempted to which are suitable for the element being studied. distinguish between distinct electroluminescence signals and Electrical treeing can be studied in a number of ways. to identify the cause and location of this light emission. Commonly the formation and propagation of a tree is tracked Previous work by the authors [13] used atomic force optically and using partial discharge measurements. These microscopy with infrared spectroscopy (AFM-IR) analysis to measurements allow the tree as a whole to be considered; they study properties of tree growth. This work identified a lack of can provide indications of the health of the insulation [2] and chemical degradation within the earliest stages of channel of the rate of channel propagation [3]. Whilst optical imaging growth as well as differing chemical signatures as channels is very useful for real-time tracking of tree growth, it lacks the propagated further from the needle tip. high resolution of techniques such as SEM [4] for observing New imaging techniques have then, begun to show aspects the detailed structure of trees. Similarly techniques in 3D of the electrical tree growth processes which can only be modelling have recently been developed [5], allowing the true observed when using tools and techniques with sufficiently structure to be viewed as opposed to a 2D projection. high resolving powers. This paper will describe the use of Chemical analysis has previously been utilized in the study AFM-IR to study chemical degradation around a needle tip of electrical treeing. Tests have been performed upon both the prior to tree initiation. The high spatial resolution and tuneable formation [6-8] of electrical trees and their propagation [9-11]. IR excitation laser allows detailed chemical characterisation of Our chemical understanding of the period of aging prior to this region to be performed. Different regions experiencing

207

different levels and types of chemical change are identified, and 0.25 µm) enabling a very smooth finish to be achieved. suggesting distinct processes occurring within them. This process is monitored using optical microscopy to ensure the intended area is reached, and a smooth finish is achieved. II. METHODOLOGY A. Sample Production and Preparation B. AFM-IR An epoxy material is chosen for this test, using the The polished sample was analyzed using the NanoIR2 Araldite LY 5052 and the Aradur HY5052 produced by (Bruker, formerly Anasys Instruments) in contact mode with a Huntsman. The samples are mixed and cured according to gold-coated silicon nitride probe. The AFM-IR allows high manufacturer instructions in a 100:38 ratio. The full sample resolution surface profiles to be produced and then IR production method is described in [13] covering mixing, absorbance spectra be taken from different locations around vacuuming, curing and post-curing procedures. the sample. In this test, locations around the needle tip are A point-plane electrode configuration is used for sample selected as shown in Fig. 1. Spectra are taken between 900 cm- aging with the sample immersed under silicone oil during 1 and 1800 cm-1, with a spectral resolution of 4 cm-1 using exposure to high voltage. The point is a steel needle with a 3 1024 co-averages for each point. IR absorbance maps are also µm tip radius manufactured by Ogura. A voltage of 8 kV was taken of 30 µm x 30 µm squares around the needle point, with applied to the sample for 40 minutes. Aging was monitored by 300 points per 300 scan lines. These are taken at selected optical imaging and partial discharge measurements. The wavenumbers using a scan rate of 0.1 Hz and 16 co-averages circuit diagram and further information on experimental setup per point. are provided in [13]. A large group of samples was aged, III. RESULTS many of which initiated trees. The sample investigated was one which had yet to show any visible indication of tree Spectra formation. The spectra from this sample are given in Fig. 1 along with It is necessary for AFM-IR testing that the area of the sample a surface profile showing the location from which the spectra chosen for testing is at the sample surface. The AFM-IR are taken. The markers closest to the needle tip (1-2 µm away) chemical analysis probes only a few micrometres below the have been coloured red or black and appear to show surface, meaning that the location of target areas is crucial. It chemically distinct areas. Blue markers signify spectra is also necessary that the surface be extremely smooth in order locations further from the needle tip (~5 µm) and green to attain the best results using this technique. This test markers even further away (> 10 µm). The spectra coloured in considers the degradation in the area surrounding the needle black show a number of absorption peaks, particularly at 1450 tip so this must be exposed. A polishing and grinding cm-1, 1250 cm-1, and a region of increased absorption from technique is used for this. First the sample is cut down around 1140 cm-1 to 1010 cm-1. The IR absorption maps are mechanically with a saw, to access the volume surrounding given in Fig. 2. Each map is 30 µm x 30 µm, with red areas the needle tip. The needle is removed at this stage, with great indicating higher levels of IR absorption and green indicating care so that the epoxy is not damaged. The sample is then lower levels. These maps appear to show a range of ground down using silicon carbide grinding/polishing paper absorptions, at different wavenumbers and in different followed by polishing using water-soluble diamond compound locations. To aid interpretation these can be to be split into pastes manufactured by Spectrographic Ltd. The pastes three distinct groups, or cases, which are illustrated in Fig. 3. contain diamond crystals of well-defined sizes (3 µm, 1 µm In Fig. 3, case 1 indicates areas of increased carbonyl and

10 µm

Fig. 1. Spectra for Sample AFM(2)-10. Inset: AFM-IR height profile showing the location of the spectra. The dark shadow on the right of the image is the original location of the needle tip of 3 µm radius.

208

a) b) c) d)

e) f) g) h)

i) j) k) l)

Fig. 2. AFM-IR Absorption Maps a) Surface Profile, b) 1056 cm-1, c) 1132 cm-1, d) 1248 cm-1, e) 1288 cm-1, f) 1360 cm-1, g) 1448 cm-1, h) 1604 cm-1, i) 1656 cm-1, j) 1702 cm-1, k) 1726 cm-1, l) 1742 cm-1. Each map is 30 x 30 µm. alkene absorption, immediately in front of the needle tip. identified in said sample. These are found at 1656 cm-1, 1702 cm-1, 1726 cm-1 and 1742 IV. DISCUSSION cm-1. Case 2 corresponds to increases in the wavenumbers 1056 cm-1, 1132 cm-1, 1248 cm-1 and 1288 cm-1. These are The striking result from these measurements is that a more difficult to assign, the most probable functional group number of functional groups appear to have formed and these responsible would be –CO and potentially indicative of ether are different in different locations. It is not yet clear whether formation. How ether would form in this case is unclear the occurrences of different groups are directly related to each however. Case 3 corresponds specifically to the 1448 cm-1 other (for example sequentially, such as if case 3 degradation absorption map and is believed to be indicative of CH3 groups. developed after case 1) or whether they are completely These groups were also identified in another sample (aged at 8 independent. These results differ from previous cases reported kV for only 20 minutes), and it was the only form of aging in studies which considered the chemical degradation occurring prior to tree initiation [6-8]. These previous cases in

Fig. 3. Schematic illustrating the location of three distinct groups of degradation identified in Sample AFM(2)-10

209

the literature tended to identify uniform, spherical regions It may be that the formation of the respective functional around the needle tip composed of homogeneous degradation groups is related, with CH3 groups potentially acting as a [14]. Typically this has focused upon the formation of precursor to carbonyl groups. Others, such as those suspected carbonyl groups. These papers report lower resolution and as being associated with ether groups, do not overlap in lower sensitivity techniques, which will have made it difficult location suggesting that they may occur independently of the to identify variations in the locations in which chemical other groups. products have formed. The low sensitivity will have also ACKNOWLEDGMENT required that the samples in these cases had to be further aged in order to produce clear evidence of chemical aging. AFM-IR McDonald is grateful for the support of the ESPRC through allows the identification of lower quantities of chemical the Power Networks Centre for Doctoral Training. Morsch is product, allowing insights into the earliest stages of chemical grateful to AkzoNobel for financial support. This publication aging. While the results in this paper do show the formation of was also supported by EPSRC grant EP/T001232/1 ‘DC carbonyl groups, other functional groups are also found to networks, power quality and plant reliability’. form, and in different locations. These are also not uniform or REFERENCES symmetric around the needle tip. This suggests that there may [1] I. Iddrissu, S. M. Rowland, H. Zheng, Z. Lv, and R. Schurch, "Electrical be more chemical complexity involved in the early stages of tree growth and partial discharge in epoxy resin under combined AC and electrical treeing, prior to tree formation, than has previously DC voltage waveforms," IEEE Transactions on Dielectrics and been considered. In particular the CH3 group has been Electrical Insulation, vol. 25, no. 6, pp. 2183-2190, 2018. identified in multiple samples and as a precursor to any [2] Y. Suzuoki, F. Komori, and T. Mizutani, "Partial discharges due to electrical treeing in polymers: phase-resolved and time-sequence carbonyl groups. The role of this in electrical treeing, in epoxy observation and analysis," Journal of Physics D: Applied Physics, vol. and in other materials, is something which should be a focus 29, no. 11, p. 2922, 1996. for further consideration. The overlap of Case 3 and Case 1 in [3] K. Wu, Y. Suzuoki, T. Mizutani, and H. Xie, "Model for partial discharges associated with treeing breakdown: I. PDs in tree channels," Fig. 3 could suggest that carbonyls form directly after and Journal of Physics D: Applied Physics, vol. 33, no. 10, p. 1197, 2000. from the CH3 groups. [4] T. Iizuka, Y. Ohki, and T. Tanaka, "Effects of coupling agent and filler Comparisons may be drawn with the results of Zhang et al dispersion on Vt characteristics of epoxy/silica nanocomposites," [12]. They suggested that different processes (hot electron and International Symposium on Electrical Insulating Materials, 2008, pp. UV recombination) occur during pre-initiation aging, and in 60-63. [5] R. Schurch, S. M. Rowland, R. S. Bradley, and P. J. Withers, different locations. Combining these results would suggest "Comparison and combination of imaging techniques for three that carbonyl groups are most likely to have formed by hot dimensional analysis of electrical trees," IEEE Transactions on electron collisions while ethers may have instead formed from Dielectrics and Electrical Insulation, vol. 22, no. 2, pp. 709-719, 2015. UV recombination. The possibility that these different [6] N. Shimizu and K. Horii, "The effect of absorbed oxygen on electrical treeing in polymers," IEEE Transactions on Electrical Insulation, no. 3, functional groups are formed from entirely different processes pp. 561-566, 1985. cannot be discounted, however further work is required to [7] N. Shimizu, K. Uchida, and S. Rasikawan, "Electrical tree and resolve this question. deteriorated region in polyethylene," IEEE Transactions on Electrical Clear evidence has been found for different functional Insulation, vol. 27, no. 3, pp. 513-518, 1992. [8] K. Kaminaga, T. Suzuki, T. Uozumi, T. Haga, N. Yasuda, and T. Fukui, groups forming prior to tree initiation, the impact of each of "The mechanism of degradation of polyethylene in a high electrical these upon tree formation is not yet clear. A channel studied field," Conference on Electrical Insulation and Dielectric Phenomena, during the earliest stages of tree initiation [13], found that it 1993, pp. 666-671. formed in a location in which there were no carbonyl groups [9] L. Hu, Y. Xu, X. Huo, and Y. Liao, "Chemical component analysis of electrical treeing in polyethylene by micro-infrared spectroscopy," IEEE detected. Moving forward, pending further testing we should Transactions on Dielectrics and Electrical Insulation, vol. 23, no. 2, pp. not yet assume that any of these groups are definitely involved 738-747, 2016. or not involved in tree formation. [10] A. Vaughan, I. Hosier, S. J. Dodd, and S. Sutton, "On the structure and chemistry of electrical trees in polyethylene," Journal of Physics D: V. CONCLUSION Applied Physics, vol. 39, no. 5, p. 962, 2006. [11] X. Chen, Y. Xu, X. Cao, S. Dodd, and L. Dissado, "Effect of tree The results shown here further demonstrate the need for channel conductivity on electrical tree shape and breakdown in XLPE suitably scaled tests in the study of electrical treeing. Tree cable insulation samples," IEEE Transactions on Dielectrics and growth is a microscopic or sub-microscopic phenomenon and Electrical Insulation, vol. 18, no. 3, pp. 847-860, 2011. so analysis is therefore also required on these scales to be able [12] S. Zhang, Q. Li, J. Hu, B. Zhang, and J. He, "Electroluminescence and electrical degradation of insulating polymers at electrode interfaces to observe the processes which are involved. Work in the field under divergent fields," Journal of Applied Physics, vol. 123, no. 13, p. of sub-micron chemical analysis is able to develop a more 135106, 2018. sophisticated understanding of the treeing process. [13] H. McDonald, S. Morsch, and S. Rowland, "Chemical Analysis of Tree The chemical degradation occurring around the needle tip Growth in Epoxy Resin Using AFM-IR Spectroscopy," IEEE Transactions on Dielectrics and Electrical Insulation, 2020 (In Press). prior to tree initiation has been studied using the AFM-IR [14] N. Shimizu, T. Takahashi, and S. Iemura, "Initiation mechanism of technique. This technique provides a significantly higher electrical tree under alternating stress-electron impact or UV photo- spatial resolution (with 50 nm resolution) than previous degradation?," International Conference on Solid Dielectrics, 2001, pp. techniques. As a result, the formation of different functional 423-426. groups in different locations around the tip has been identified.

210