Advancing NDE Technologies, Research, and Engineering in a Changing World This Proceedings Is Intended for the Sole Use of Registered Attendees

April 1 - 4, 2019 | Hyatt Regency Orange County | Garden Grove, CA

Advancing NDE Technologies, Research, and Engineering in a Changing World This proceedings is intended for the sole use of registered attendees. No license is granted to reproduce, copy, forward, or redis- tribute this proceedings in any form or fashion and is exclusively intended for personal use only. No part of this publication or its contents may be copied, uploaded to the internet, or stored in any shared retrieval system.

Exclusive of those papers that are a work of the federal government and not subject to copyright.

The American Society for Nondestructive Testing Inc. (ASNT) is not responsible for the authenticity or accuracy of information herein. Published opinions and statements do not necessarily reflect the opinion of ASNT. Products or services that are advertised or mentioned do not carry the endorsement or recommendation of ASNT.

No part of this publication may be reproduced or transmitted in any form, by means electronic or mechanical including photo- copying, recording or otherwise, without the expressed prior written permission of The American Society for Nondestructive Testing Inc.

IRRSP, NDT Handbook, The NDT Technician, and asnt.org are trademarks of The American Society for Nondestructive Testing Inc. ACCP, ASNT, Level III Study Guide, Materials Evaluation, Nondestructive Testing Handbook, Research in Nondestructive Evaluation, and RNDE are registered trademarks of The American Society for Nondestructive Testing Inc.

First printing 10/19

Errata, if available for this printing, may be obtained from ASNT’s website, asnt.org.

ISBN: 978-1-57117-466-6

Printed in the United States of America.

Published by: The American Society for Nondestructive Testing Inc. 1711 Arlingate Lane Columbus, OH 43228-0518 asnt.org

ASNT Mission Statement: ASNT exists to create a safer world by advancing scientific, engineering, and technical knowledge in the field of nondestructive testing.

ii TABLE OF CONTENTS

Papers...... 1

Influence of Abrasive Grinding on Through-Section Residual Stresses on Bearing Steel...... 2 Vikram Bedekar, Rohit Voothaluru, Jeff Bunn, and R. Scott Hyde

High Resolution SCC Depth Map in Pipeline Samples Using New X-Ray Imaging Techniques ...... 9 Yohan Belanger, Luc Perron, and Xavier P. V. Maldague

Combined NDT Correlation to Estimate the Compressive Strength of Concrete ...... 20 Abhishek Chitti, Sandeep G. Burra, Tsuchin P. Chu, Prabir Kolay, and Sanjeev Kumar

Evaluation of Nondestructive Evaluation Methods for Applicability to Precast Concrete Deck Panels...... 30 Saman Farhangdoust and Armin Mehrabi

Platform for Test Scripting and Automated Testing of Smart Meters...... 42 Vaibhav Garg

Ultrasonic Testing Beyond Flaw Detection...... 52 Hormoz Ghaziary and Alfred Haszler

Guided Wave Simulations in Pipes with Non-Axisymmetric and Inclined Angle Defects using Finite Element and Hybrid Modeling...... 59 Masoud Masoumi and Ryan Kent Giles

Ultrasonic Thickness Estimation using Multimodal Guided Lamb Waves Generated by EMAT...... 69 Joaquín García-Gómez, Roberto Gil-Pita, Antonio Romero-Camacho, Jesús Antonio Jiménez-Garrido, Víctor García-Benavides, César Clares-Crespo, and Miguel Aguilar-Ortega

Magnetostrictive Cold Spray Sensor for Long-Term or Harsh Environment Ultrasound...... 78 S. W. Glass, J. P. Lareau, K. S. Ross, S. Ali, F. Hernandez, and B. Lopez

MWM-Array and MR-MWM-Array Eddy Current Testing for Piping and Vessels...... 86 N. Goldfine, T. Dunford, A. Washabaugh, S. Chaplan, and K. Diaz

The Use of the Pull-Off Test Method to Characterize the Performance of a Concrete Repair System...... 94 Evan Karunaratne, Dr. Julie Ann Hartell, and Dr. Norbert Delatte

TFM Acoustic Region of Influence ...... 104 Chi-Hang Kwan, Guillaume Painchaud-April, and Benoit Lepage

Automated Non-Destructive Evaluation of Spot Welds using the Imaging Analyses of the Residual Magnetic Flux Density...... 113 Christian Mathiszik, Jörg Zschetzsche, and Uwe Füssel

Process for Nondestructive Testing of Pressure Vessels Using Electronic Distance Measurements to Measure 3-D Coordinates of Cardinal Points While Pressure and/or Force Testing...... 125 David H. Parker

Resistivity Behavior of Concrete Mixtures with Included Supplementary Cementitious Materials...... 134 Cody Shults and Julie Hartell

iii Laser Shearography and Ultrasound Inspection of Composite Laminates with Overlapping Fiber Plies ...... 144 Sarah L. Stair, David G. Moore, Corinne Hagan, and Ciji L. Nelson

Process Monitoring at Capacitor Discharge Welding...... 154 Jörg Zschetzsche, Max-Martin Ketzel, Uwe Füssel, Hans-Jürgen Rusch, and Nicolas Stocks

Author Index...... 161

iv PAPERS

1 Influence of Abrasive Grinding on Through-Section Influence of abrasive grinding on through-section residual stresses in bearing Residual Stresses on Bearing Steel steel Vikram Bedekar1, Rohit Voothaluru1, Jeff Bunn2, and R. Scott Hyde1 Vikram Bedekar1, Rohit Voothaluru1, Jeff Bunn2, and R. Scott Hyde1

1 1The Timken Company The Timken Company Material Science Research and Development Material Science Research and Development North Canton OH 44720, USA North Canton OH 44720, USA (234) 262-2396; Fax (234) 262-2282; email [email protected] (234) 262-2396; Fax (234) 262-2282; [email protected]

2 2Oak Ridge National Laboratory Oak Ridge National Laboratory Neutron Scattering Division Neutron Scattering Division Oak Ridge, TN 37830, USA Oak Ridge, TN 37830, USA (865) 241-6133; email [email protected] (865) 241-6133; [email protected]

ABSTRACT In today’s industrial world, bearings are used in traditional applications such as automotive, steel mills, paper mills, railways as well as high precision applications such as aerospace, medical, renewable energy etc. Regardless of application, residual stresses play an important role in the service life of a bearing. In general, bearings experience Hertzian stresses in the order of 2-3 GPa with maximum stresses generated in the subsurface. Depending upon the magnitude and nature (tensile or compressive), the residual stresses enhance or reduce the application stresses. The residual stresses are generated during each stage of manufacturing process. For bearing components, the manufacturing stages include – tube rolling, green machining followed by heat treatment and final finishing. Majority of bearings are finished using abrasive grinding to achieve the final size and shape. The work holding and material removal during the grinding process can cause significant distortion. Distortion of bearing components is a significant issue, especially for thin sectioned components, leading to high scrap rates. In this study, neutron diffraction was utilized to understand the evolution of through thickness residual stresses after abrasive grinding on a thin walled bearing component. The results from neutron diffraction were coupled with distortion measurements especially the out of roundness to understand the effect of the final processing on the residual stress state present in the bearings.

Keywords: Neutron diffraction, Residual stress, Grinding

INTRODUCTION Over the past few decades residual stresses have been studied in great detail by academia and industry alike. As the name suggests, “residual stresses” are the stresses that reside in a component after processing or experiencing in-service operating conditions. Manufacturing of bearing or gear components involves steel making, forging/rolling/tube making, soft machining, and heat treatment followed by final finishing. Residual stress fields are generated and stored during each manufacturing stage. Depending upon the part geometry and temperature of the manufacturing stage, the stored residual stresses are released in the form of distortion. In the case of bearing components, the tolerances are narrow and distortion can lead to significant scrap and higher manufacturing costs. Residual stresses can cause beneficial or detrimental effects on the product life cycle. Tensile residual stresses drastically increase the propensity to crack, while compressive residual stress (CRS) counteract the application stresses thereby enhancing the service life. Considerable efforts have been performed to enhance the structural integrity of components by inducing CRS. Recent literature has shown that end processes such as shot peening [1], laser peening [2], hard turning [3] induce CRS resulting in enhanced fatigue life.

2 Zaretsky et al. (1992) [4] indicated that presence of compressive residual stress reduces maximum shear stress. They presented this phenomenon by following equation-

1 )(     S )( (Eq. 1) max r max 2 r

Where (max)r is the final maximum shear stress and Sr is the residual stress ( + if tensile and – if compressive).

9   max  LF.      max )( r  (Eq. 2)

In ball bearing, the maximum shear stresses are a function of Hertzian stresses and the fatigue life (Life Factor-LF) is inversely proportional to maximum Hertz stress to the power of nine (Eq.2). For the first time, Zaretsky et al. (1965) [5] indicated that a compressive residual stress that exists at the depth of maximum shear stress decreases the maximum shear stress. It is well known that the final material removal operation plays in influential role in the surface and surface residual stresses in a final product. In today’s industry, most of the final material removal is performed using abrasive grinding or hard turning. For high volume production abrasive grinding is economical given the short process times. The residual stress development after finish grind operation has been studied in literature. Balart et al. [6] stated that mechanical deformation, thermally induced plastic deformation, and phase transformation during grinding induce residual stresses. The magnitude of residual stresses are not only dependent upon material properties but also grinding parameters such as wheel speed, depth of cut, abrasive, lubricant etc. Residual stresses can be measured using direct and indirect methods [7]. Indirect methods involve monitoring changes in the dimension of a component after cutting or removing material. These techniques involve hole drilling, curvature measurements and crack compliance methods. On the other hand, direct measurements involve measuring change in the atomic d-spacing using methods like X-ray diffraction, X-ray synchrotron, and more recently neutron diffraction. Due to cost and accessibility, most of the studies utilize X-ray diffraction to measure the surface and subsurface residual stresses. The subsurface measurements are performed by removing material layers by electropolishing methods. This method is time consuming and does not provide residual stresses through the thickness of the material. Therefore, until now most of the investigations are restricted to a few hundred micron depths. With the advent of new state of the art techniques such as neutron diffraction and X-ray synchrotron, it has become possible to understand the development of residual stresses through the section of component. Both techniques perform measurements in a non-destructive manner, thus preventing the relaxation of stresses due to sectioning. Recent work at the Oak Ridge National Laboratory (ORNL) has shown that the engineering diffractometer at 2nd Generation Neutron Residual Stress Facility (NRSF2) at ORNL’s High Flux Isotope Reactor (HFIR) can be used for wide range of materials to map the residual strains [8-11]. Neutrons can provide unique insights into material’s atomic structure because of their ability to penetrate deeply in to the material. The penetration depth of neutrons can be upto a few centimeters [12] with spatial resolution as small as 1mm3. This work investigates through section residual stress development on a thin walled and asymmetrical bearing ring after final grinding operation. Until now most of the research has been focused on near surface residual stresses development in circumferential direction. In this paper, the through section residual stresses in all three directions were computed using neutron diffraction. Consequently, area maps on residual strains in three directions are presented. The influence of stresses on distortion is also discussed.

3 EXPERIMENTAL DETAILS Cylindrical tubes of AISI 52100 steel (Table 1) were soft machined to produce asymmetrical bearing rings as shown in Fig. 1. The part dimensions after soft machining were 206mm (OD), 143mm (smallest ID) and 190mm (largest ID). The heat treatment was conducted by heating the ring to austenitizing followed by rapid quenching. The ring was tempered for 2 hours to improve toughness and avoid cracking. The microstructure of the ring consisted of tempered martensite, retained austenite (2-3%) and carbides. The hardness was 58-60 HRc. The ring was finished using abrasive grinding. First rough grinding was performed to remove the oxide layer formed after heat treatment. Final passes of grinding were designed to meet the desired size and shape.

Table 1: Chemical Composition of AISI 52100 (balance Fe). %C Mn Si Cr S P Fe 1.03 0.34 0.29 1.46 0.009 0.006 Balance

(a) (b)

Figure 1: (a) Bearing ring (b) Neutron diffraction set-up at HFIR

The neutron diffraction experiments were conducted using 2nd Generation Neutron Residual Stress Mapping Facility (NRSF2). The neutrons generated in beamline HB-2B in High Flux Isotope Reactor (HFIR) at Oak Ridge National Laboratory (ORNL) were used as a source. The bearing ring (Fig. 1a) was set-up into the instrument as shown in Fig. 1b. The schematic of the ray diagram is shown in Fig. 2a.

(a) (b)

Figure 2: Neutron diffraction set-up (a) Ray diagram (b) Sketch detailing use of slits to investigate grains inside the component [8]

4 The details of this instrument have been discussed in previous literature [8]. The measurement technique is similar to powder diffraction. However, this technique involves spatially mapped signal which is achieved by defining a volume of grains in space. This is achieved by restricting the irradiated region and the field of view of the detector by slits and radial collimators as shown in Fig. 2b. For each gauge volume, a diffraction signal is generated with spatial information of the sample and orientation of scattering vector (Q in Fig. 2b). The peak positions were determined by pseudo voight function [8].

Strains are determined from the ratio if the d-spacing (d0) to the stress free d-spacing as shown in Eq. 3.

(Eq. 3)

Residual stresses (ij) are determined from 3-D strain as follows:

(Eq. 4)

Where  = poisson’s ratio (0.3 for steel), E = 210 GPa and , , are the measured strain in principal directions. For the experiments a gauge volume of 2 x 2 x 2 mm was used. A neutron wavelength of 1.54A° was utilized. The strains were measured on martensite {211} family of planes. It is known that partial burial of the beam causes erroneous results [9]. Therefore, care was taken to ensure that all the measurements were conducted with beam completely buried within the sample. This was achieved by continuous monitoring of peak intensities. A total of 64 measurements were conducted. A beam size of 2mm was used. During the set-up, it was difficult to locate the beam precisely near the surface. Since bearing race surface experiences the contact fatigue, the beam was aligned to be as close to the raceway surface as possible. In that effort, the near surface data on the rest of the surfaces could not be captured in a reliable manner and therefore ignored.

RESULTS AND DISCUSSION The through section strain and stress area maps are shown in Fig. 3 and Fig. 4 respectively. From the stress and strain fields, it is evident that the near surface effects of grinding process are predominantly seen in circumferential direction compared with radial or axial directions. It is well founded that the stresses that are perpendicular to the direction of crack propagation offer best resistance to fatigue damage [14]. In the classical bearing fatigue, the damage occurs in axial direction. Therefore, circumferential (or hoop) direction offers most valuable information. In circumferential direction, the near surface strain or stress is highly compressive compared with the core. Grinding is a high energy material removal process. In the previous literature, near surface high compressive residuals stresses after grinding operation have been reported [15]. It is also known that the magnitude and the direction of the stresses strongly depend upon the grinding parameters such as wheel speed, coolant condition, depth of cut etc. Due to abrasive action between the grinding wheel and the surface, high temperatures and stresses are generated, causing altered material zone (AMZ) [16]. It is also observed that most of the compressive stresses (or strains) are observed in circumferential direction while most of the tensile stresses are seen in the radial direction. It is perhaps due to the rotational speed of the grinding wheel is in the circumferential direction.

5 (a)

(b)

Figure 3: Strain area maps after grinding (a) axial, (b) radial and (c) circumferential direction

(a)

(b)

(c)

Figure 4: Residual stress area maps after grinding (a) axial, (b) radial and (c) circumferential direction

6 The thin walled bearing ring was gently finished to ensure that material removal process is plastically dominated. The high strain generated during plastic deformation is accomodated by localized dislocations. Therefore, the mechanical interactions between the abrasive grains and the workpiece surface gives rise to compressive residual stresses. It should be noted that the high compressive stress reported in the literature [15, 17] is generally observed within top 12-25m before it returns to zero or slightly compressive. However, in Fig. 4, the high magnitudes of residual stresses are observed upto a few millimeters. It is likely that this is a combined artifact of image processing and the larger beam size. The magnitude of the compressive residual stress decreases once the surface effects due to grinding taper off. However, the stresses are still in the compressive regime. As mentioned previously, before final grinding, the component was was austenitized to 850°C and quenched rapidly to form body centered tetragonal structure (BCT) from face centered cubic (FCC) structure of austenite. The formation of martensite is a shear dominated diffusionless transformation which causes expansion of atomic cells. Due to this expansion, compressive stresses are generated. However, the transformation does not occur uniformly due to asymmetric shape of the component. Due to direct contact with the quench media, the surfaces transform first. Additionally, the thinner sections transform ahead of the thicker sections. As the transformation initially occurs at the surfaces, the surfaces undergo expansion. This results in compressive stresses in the surface/subsurface region. Once the surface transformation is complete, the core starts transforming (and expanding) causing reversal of stresses. Thus, the final stress state results in tensile stresses in the surface region and compressive stresses in the core. While the tensile stresses in the surface/subsurface region alter to compressive stresses due to the plastic flow during grinding, the core stresses are the result of the heat treatment. It is worthwhile noting that pockets of increased stresses are observed in core section of the component. These are perhaps the last regions to transform during heat treatment. This effect is predominantly observed in the radial stress/strain maps (Fig. 3b and Fig. 4b). The tensile stresses near the flat surface along the Y- axis is mostly likely due to stacking of the multiple test samples in a laboratory furnace which led to limited exposure of that surface to the coolant. The observation was confirmed using X-ray diffraction. The gradients in the stress state in the component results in to distortion. Due to variation in cooling rates during heat treat causes bending deflection [18]. The distortion tendancy of the component is further increased due to the asymmetric and thin walled geometry. It was observed the out of roundness (or ovality) at the thinnest section was much higher (45m) compared with the thickest section (38m). This is most likely because the thinnest section has to accommodate higher stress gradient compared with the thickest section. This is in addition to the structural effects due to clamping and the cutting forces during grinding operation.

SUMMARY Through thickness residual stresses assessment was conducted in a non-destructive manner using neutron diffraction. The measurements were performed on a thin walled bearing ring after final grinding operation. The results provided a whole picture of residual stress development in radial, axial and circumferential direction. In a great detail, the data indicated the grinding influences the residual stresses only in surface/subsurface regions while the previous operations such as heat treatment dominate the stresses in the core section. The stresses developed in the heat treatment are uneven due to variation in the thickness and cooling rates. This variation influences the distortion after grinding operation. The information from this study can be directly used in structural integrity calculation or validate FEM to develop predictive model to further control and manipulate processes to reduce distortion which will ultimately results in reduction in scrap rates.

7 ACKNOWLEDGEMENTS A neutron diffraction portion of this research used resources at the High Flux Isotope Reactor, a DOE Office of Science User Facility operated by the Oak Ridge National Laboratory. The authors would like thank Matt Boyle, Jeremy Kimble, Ike Mann and Prototype shop at The Timken Company for fabricating and analyzing the bearing rings.

REFERENCES (1) Torres, M. A., Voorwald, H. J. C., 2000, An Evaluation of Shot Peening, Residual Stress and Stress Relaxation on the Fatigue Life of AISI 4340 Steel, International Journal of Fatigue, 24:877-886. (2) Hammersley, G., Hackel, L., Harris, F., 2000, Surface Prestressing to Improve Fatigue Strength of Components by Laser Shot Peening, Optics and Lasers in Engineering, 34:327-337. (3) Guo, Y. B., Warren, A. W., Hashimoto, F., 2010, The Basic Relationships Between Residual Stress, White Layer, and Fatigue Life of Hard Turned and Ground Surfaces in Rolling Contact, CIRP J.of Manufacturing Science and Technology, 2: 129-134. (4) Zaretsky E. 1992, STLE Life Factors for Rolling Bearings. STLE Publication, SP-34. Zaretsky, E. V., Parker, R. J., Anderson, W. J., & Miller, S. T. (1965). Effect of component differential hardness on residual stress and rolling-contact fatigue (No. NASA-TN-D-2664- NASA Cleveland Lewis center). (6)(5) Balart M. J., Bouzina A., Edwards L., Fitzpatrick M., 2004, The Onset of Tensile Residual Stresses in Grinding of Hardened Steels, Materials Science and Engineering A, 367: 132-142. (7) Brinksmeier, E., Cammett, J. T., König, W., Leskovar, P., Peters, J., Tönshoff, H. K., 1982, Residual Stresses – Measurement and Causes in Machining Processes, CIRP Annals – Manufacturing Technology, 31/2:491-510. (8) Cornwell, P., Bunn, J., Fancher, C.M., Payzant, E.A., Hubbard, C.R., 2018, Current Capabilities of the Residual Stress Diffractor at the High Flux Isotope Reactor, Review of Scientific Instruments, 89/9:092804. (9) Hempel, N., Bunn, J., Nitschke, T., Payzant, E. A., Dilger K., 2017, Study on Residual Stress Relaxation in Girth-Welded Steel Pipes Under Bending Load Using Diffraction Methods, Material Science and Engineering A, 688: 289-300. (10)Ikeda T., Bunn J., Fancher C., Seid A., Motani R., Matsuda H., Okayama T., 2018, Non-destructive Measurement of Residual Strain in Connecting Rods Using Neutrons, SAE Technical Paper. (11)Eisazadeh, H., Bunn, J., Aidun, D., 2017, Numerical and Neutron Diffraction Measurement of Residual Stress Distribution in Dissimilar Weld, Welding Journal, 96(1):21-30. (12)Withers P. J. and Bhadeshia H. K. D. H. 2001, Residual Stress. Part 1 – Measurement Techniques, Material Science and Technology, 17/4: 355-365. (13)Wang, X. L., S. Spooner, et al., 1998, Theory of the peak shift anomaly due to partial burial of the sampling volume in neutron diffraction residual stress measurements, Journal of Applied Crystallography, 31: 52-59. (14)Wulpi, D.J., 1966. How Components Fail. American Society for Metals. (15)Sridharan, Uppiliappan, Vikram Bedekar, and Francis M. Kolarits. "A functional approach to integrating grinding temperature modeling and Barkhausen noise analysis for prediction of surface integrity in bearing steels." CIRP Annals 66.1 (2017): 333-336. (16)Oliveira J. F. G., Silva E. J., Guo C., Hashimoto F., 2009, Industrial Challenges in Grinding, CIRP Annals – Manufacturing Technology, 58/2:663-680 (17)Malkin S. and Guo C., 2007, Thermal analysis of grinding, CIRP Annals – Manufacturing Technology, 56/2:760-782. (18)Pan, J., 2002, Factors Affecting Final Part Shaping, Handbook of Residual Stress and Deformation of Steel, ASM International: 151 –157.

NOTICE OF COPYRIGHT This manuscript has been authored by UT-Battelle, LLC under Contract No. DE-AC05-00OR22725 with the U.S. Department of Energy. The United States Government retains and the publisher, by accepting the article for publication, acknowledges that the United States Government retains a non-exclusive, paid- up, irrevocable, worldwide license to publish or reproduce the published form of this manuscript, or allow others to do so, for United States Government purposes. The Department of Energy will provide public access to these results of federally sponsored research in accordance with the DOE Public Access Plan (http://energy.gov/downloads/doe-public-access -plan).

High Resolution SCC Depth Map in Pipeline Samples High Resolution UsingSCC D Newepth X-Ray Map i nImaging Pipelin eTechniques Samples Using New X-Ray Imaging Techniques Yohan Belanger1,2, Luc Perron2, and Xavier P. V. Maldague1 1,2 2 1 Yohan Belanger , Luc Perron , and Xavier P. V. Maldague 1Université Laval 1 Département de génieU nélectriqueiversité L etav deal génie informatique Départem1065,ent de av. gé ndeie laél eMédecine,ctrique et dbureaue géni e1300 informatique 106Québec,5, av. de Qc,la M Canada,édecine ,G1V bure a0A6u 1300 (418) 656-2984;Québec email, Qc, [email protected], G1V 0A6 (418) 656-2984; [email protected] 2LynX Inspection 2 2740L yn XRue In sEinsteinpection Québec,27 4Qc,0 R Canada,ue Einst eG1Pin 4S4 (418) 657-7706;Québ emailec, Q [email protected], Canada, G1P 4S4 (418) 657-7706; [email protected]

ABSTRACT Materials in pipelines are subjected to a series of stress and corrosive environments. These conditions eventually lead to stress corrosion cracking (SCC). Several NDT techniques already exist to detect and assess such cracks in the field, some being more efficient than others at providing accurate depth measurements. Depth measurements are essential to help decide between a low-cost repair and an expensive section replacement, but any new NDT method first needs to be evaluated in the lab before its gets deployed in the field. Validating that the method provides trustworthy results can hardly be done without a reliable ground truth. A new low cost X-Ray imaging method, involving scatter corrections and the physical aspects of SCC, was developed to generate high resolution 3D SCC depth maps of pipeline samples showing precise crack positions and depth measurements. This method could be used as a gold standard for evaluating traditional NDT techniques, such as ultrasound or eddy current, since digital X-Ray imagery in a controlled laboratory environment can yield better resolution. We will discuss the challenges of making such measurements and the techniques used to overcome them using state of the art hardware and software.

Keywords: Digital Radiography, DR, X-ray, Computer Vision, SCC, Pipelines

INTRODUCTION Pipeline transport is a method of long-distance transportation for fluids. They are often used to transport hydrocarbons and chemicals, but can also be used to transport water or any other liquid and gaseous products. Because pipelines are used for large volumes of fluids, any failure can have dramatic consequences. This is where non-destructive testing (NDT) can play a big role in finding preventable defaults. One such default is Stress Corrosion Cracking (SCC). It is defined as the growth of crack formation in a corrosive environment. Several NDT techniques can be used to spot such formations in the field. Newer techniques aim to find the depth of those formations in order to help decide between a low-cost repair and an expensive section replacement.

In this paper, we will talk about new radiography imaging techniques that can provide a high resolution map of cracks in a pipeline sample, both in position and in depth. These techniques use a combination of X-Ray Physics and Computer Vision in order to get the most information out of pipeline samples. In time, we want to verify that these non destructive techniques can be used to produce ground truth data to help assess in a laboratory environment other NDT solutions that are designed to be deployed in the field. We will detail the materials and methods, explain the theory behind them and discuss our results.

THEORY Radiation Physics Here are the primary effects that have an impact on image quality and thickness calibration.

X-Ray attenuation The major interactions causing X-Ray attenuation are the Photoelectric Effect, Compton Scattering and Pair Production. However, in the energy ranges that were used in this experiment, only the Photoelectric Effect and Compton Scattering should occur (Figure 1a). In reality, instead of computing the total cross-section of those two interactions, we use the Linear attenuation coefficient. This coefficient depends on the energy of the beam and the material it is going through(Figure 1b). Note here that the Linear attenuation coefficient is the Mass attenuation coefficient times the material’s density.

−μx (Eq.1) I = I0e

Where: = Output beam intensity I = Input beam intensity I0 μ = Linear attenuation coefficient x = Distance travelled in the material

(a) (b) Figure 1: X-Ray Physics figures. (a)Relative importance of the three major types of X-Ray interactions(1). (b)Example of X-Ray Mass attenuation coefficient(2).

In practice, the end result depends on the integral of the attenuation coefficient over the spectrum of the source and the materials the beam is going through. It is however possible to see from Eq.1 that material thickness predictions based on beam intensity ratios are possible if these intensities are obtained in the exact same conditions as the calibration piece.

Compton Scattering At the energy range used in this experiment, Compton Scattering is the first interaction most photons will encounter when going through the sample. It is a lossy interaction that changes the path and the energy of the incident photon

and produces a recoil electron that escapes with part of the incident energy. The energy of the outgoing photon is given by(1):

Eγ (Eq. 2) E = 2 γ′ 1 + (Eγ/mec )(1 − cosθ)

Where:

Eγ′ = Outgoing photon energy Eγ = Incident photon energy me = Electron rest mass c = Speed of light θ = Scattering angle

As it is possible to see, the energy of the outgoing photon depends on the scattering angle. This means that in a given material, the scattered photons that have a narrow angle will keep most of their incident energy and be able to go through the sample with a wrong angle. This wrong angle is in turn one of the reasons, along with the focal spot size of the source, for visual blurring. This means that if you have a cradle that has a too much absorbent material, the contrast lines created in the sample will be blurred before entering the detector. It is possible to see the extent of this effect by looking at the angular differential cross-section of a single photon scattered from a single free electron, also named the Klein-Nishina formula(Figure 2).

(a) Figure 2: Klein-Nishina distribution of scattering-angle cross sections over a range of commonly encountered energies.

Corrosion effects on X-Ray attenuation Most of the photons going through the sample will eventually be stopped by the Photoelectric Effect. Indeed, when looking at the proportionality of its cross section for photons with energies above the highest atomic binding energy, it is possible to see the relation between the Photoelectric Effect, the atomic number and the incident energy of the photon.

Zn (Eq. 3) ∝ m σ E

Where: σ = Approximate cross section n = Atomic number (with n≈4 at 0.1 MeV and n≈4 .6 at 3 MeV) Z m = Energy of the incident photon (with m≈3 at 0.1 MeV and m≈1 at 5 MeV) E

As it is possible to see from this equation, the photoelectric effect decreases steeply with the energy of the incident photon leaving room for compton scattering at higher energies. It also increases steeply with the atomic number of the material. (1).

Without going into the chemical details of corrosion, this process results in the bounding of lighter elements (like Oxygen or Chlorine) to the steel of the pipeline. From a radiation physics viewpoint, corrosion has the effect of lowering the effective atomic number of the samples, meaning that the photoelectric effect will be less effective at stopping radiation. It will have the effect of making parts of the samples appear thinner in the captured images.

MATERIAL AND METHODS In this section we will discuss the experimental setup and the methods used to clean the images.

Experimental Setup X-Ray Source We used a 150 kV, 500 μA X-Ray source with a focal spot size of around 50 μm. It is a cone-beam source with a solid angle of 45° and using a constant potential (as opposed to to a pulsed beam). In this experiment, the applied potential ranged from approximately 100 kV to 150 kV and the current from approximately 300 μA to 500 μA.

X-Ray Imaging Detector We used a detector equipped with an array of CMOS sensors coupled with a GadOx scintillator that produced 14 bpx images. The size of the array was 3096 x 3100 pixels with a resolution of 0.099 mm. The active area was 306 x 307 mm and could record images at 30 fps using two Camera Link cables. The source was placed at about 550 mm above the detector. We made sure that the detector array was circumscribed by the circle of the cone beam, meaning that the beam completely covered its active area.

Linear translation stage We used a high precision, high accuracy 2D linear translation stage that moved on a set of rails mounted on an optics table. The height of the sample was set with a sturdy assembly of aluminum pegs bolted to the head of the moving cradle assembly.

Radiography Cabinet We used a lead-lined NDT cabinet that was rated for the dose-rate of our source and met the radiation safety requirements of the Canadian Nuclear Safety Commission (CNSC).

Pipeline samples The pipeline samples were provided by a local NDT company and had already been inspected using magnetic particle inspection, meaning the cracks were visible on the sample. Their size was around 200 x 100 mm and were made from very corroded X52 steel with varying thicknesses from ?? to ?? mm. A separate sample made of the same steel contained a series of thin EDM notches with distinct pre-defined depths was used as a reference. All samples were scanned at a height of approximately 320 mm above the detector array lying on an extruded polystyrene sheet.

Software used Most of the software we used was based on open source libraries. Image processing and analysis was done using a combination of custom Python and C++ software utilities. Additional software tools were developed in Python and C++ to streamline the data analysis process. Some of the more detailed analysis required manual interventions.

Methods In this section we will discuss the methods we used for image production, processing/cleaning and analysis.

Image acquisition We used the intrinsic geometrical magnification of our X-Ray setup to increase the resolution of the images and make the cracks more visible. As a result, the sample size exceeded considerably the available field of view of the detector which lead to the development of a custom image stitching algorithm. This algorithm automatically stitched all of the images together based on the given height of the sample and its relative XY position on the linear stage. To reduce the effects of noise and the geometric magnification of the penumbras in the images, we captured many images at every position and used the averaged result.

Image pre-processing Detector background noise and intensity discrepancies caused by the source’s heel effect were cleaned by using a custom polynomial gain and offset correction algorithm. We then removed the bad sensors by using a bilinear filter and rescaled the images to 16 bpx to facilitate processing. This process chain, which was repeated with every averaged image received from the detector, could be compared in essence to a flat field correction.

Depth calibration We used a custom-made step wedge made out of a plate of X52 steel. The wedge was machined with steps ranging from 1 to 9 mm in height and 12 mm in length. The wedge was used as an intensity to thickness calibration piece. We were able to reduce the effects of X-Ray scattering by isolating each step of the wedge with other steel plates during capture. This effectively receded the edge effects and it was decided that the samples were to be scanned in this manner also. During this exercise we also found out that using polystyrene as the cradle base for the samples was prefered over a polyethylene base as it reduced diffusion blurring. We very carefully scanned each step and got the grey value to thickness curve from which we could start our work.

Depth analysis We first started by verifying that the thickness curve we got from the step wedge was self consistent and could predict thickness well enough to verify the depth of the EDM notches on our machined sample. The thickness curve was applied to the stitched images of the pipeline samples and we started the analysis. We used a rolling-ball background subtraction algorithm that could attenuate the error caused by the curvature of the sample. We later applied a correction that took into account the degree of corrosion around the cracks. This correction is based on X-Ray attenuation physics and the change in effective atomic number of steel after being corroded.

RESULTS In this section we will discuss the results obtained from the calibration plate and the samples.

Crack Depth EDM notched plate The first results came from the EDM machined plate with cavities of known depths. This plate was not corroded so we could test our image processing and depth finding algorithms on it. We had values for multiple notches but only a few will be presented here. The analysed depths were between 1 and 3 mm. The resulting images of each analysed EDM notches are presented in Figures 4 and 5.

(a) (b) Figure 4: (a) Notch #2, 1 mm depth (b) Notch #3, 2mm Depth

(a) (b) Figure 5: (a) Notch #4, 3 mm depth (b) Notch #11, 1mm Depth

Table 1: Predicted EDM notches depth and calibrated X-Ray measurements Notch # EDM depth Measured Depth Absolute Error Relative Error

2 (1.00 ± 0.05) mm 0.938 mm 0.063 mm 6.67 %

3 (2.00 ± 0.05) mm 2.166 mm 0.166 mm 7.68 %

4 (3.00 ± 0.05) mm 2.863 mm 0.137 mm 4.79 %

11 (1.00 ± 0.05) mm 0.982 mm 0.018 mm 1.83 %

Having these results in hand, we were ready to analyse an actual pipeline sample.

Pipeline Sample Several samples were thoroughly analysed during this research, but for simplicity, we will focus in this paper on the results from only one of these samples. The experiments and analysis of this sample were repeated enough to ensure the repeatability of the methods. For a better understanding of the regions of interest in the sample (ROI), we will start by showing in Figure 6 the full stitched depth map of the sample with highlighted ROIs and then move on to the analysis of each individual region.

Figure 6: Full depth map of the sample

In Figure 7 we can see a high resolution version of region 2. It is possible to see first hand the effect of corrosion on the images, as a kind of white cloud that envelops the cracks.

Figure 7: Zoomed depth map of Region 2

Table 2 compares the data we measured with data obtained from an Eddy Current analysis provided by the same NDT company that gave us the samples to see where our results stand against other techniques. Unfortunately we were not able to have the real depth measurement to test our data, as it would have to be taken with a destructive evaluation technique and the samples were not our own. In table 2 we will see the maximum depth data obtained from X-Ray for each of the defined regions in Figure 6.

Table 2: SCC Depth for each of the defined regions Region # Maximum Depth Maximum Depth with Corrosion Correction

1 (1.000 ± 0.090) mm (1.764 ± 0.127) mm

2 (0.981 ± 0.091) mm (1.771 ± 0.131) mm

3 (1.127 ± 0.140) mm (1.959 ± 0.186) mm

4 (1.274 ± 0.125) mm (2.102 ± 0.185) mm

5 (1.964 ± 0.145) mm (2.813 ± 0.190) mm

6 (1.289 ± 0.086) mm (2.092 ± 0.120) mm

7 (1.368 ± 0.104) mm (2.195 ± 0.135) mm

8 (1.696 ± 0.100) mm (2.525 ± 0.129) mm

3D Depth Map We were also able to produce a 3D depth map of the cracks. Since the corrosion correction is now mostly done by hand and is local to each crack, the results were rescaled for a better view and are only qualitative.

Figure 10: Overhead view of the 3D depth map

Figure 11: Underneath view of the 3D depth map DISCUSSION

Precision and Accuracy of results Comparison to other methods To verify the validity of our results we compared them to results obtained from an Eddy Current analysis. Figure 12 shows a depth map from this analysis and Table 3 shows the values for the two methods side by side. It should be noted that the Eddy Current method cannot go deeper than 3 mm, that’s why results deeper than that will be noted as 3+ mm.

Figure 12: Eddy Current analysis depth map

Table 3: Comparison between our data and Eddy Current analysis Regio This Article Eddy Current Difference n # 1 1.764 mm 1.0 mm +0.764 mm

2 1.771 mm 0.9 mm +0.871 mm

3 1.959 mm 1.3 mm +0.659 mm

4 2.102 mm 2.4 mm -0.298 mm

5 2.813 mm 3+ mm -0.187 mm (+)

6 2.092 mm 2.2 mm -0.108 mm

7 2.195 mm 2.5 mm -0.305 mm

8 2.525 mm 3+ mm -0.475 mm (+)

Error Analysis In this part, we will review the factors that impacted our results and discuss their implications.

Shape of samples An error factor that came up right at the start is the fact that the samples, coming from a pipeline, were having an ellipsoidally curved shape. While we used image stitching to reduce the effect of the cone beam divergence, we couldn’t find a stable way to place the samples so that the beam crossed the exact same amount of material in each parts of the stitched image. This is especially true when trying to correct for corrosion on the up and down edges of the plate. It can be seen as the overshoot in the data comparisons for regions 1, 2 and 3. We usually come to find shallower cracks than Eddy Current analysis.

Orientation of cracks Since X-Ray imaging is basically a measure of the path length of the beam in the sample, finding the perfect orientation of the sample for the beam to reach the bottom of the cracks was a challenge in itself. That’s why we used the cone beam in combination with the linear stage to get a series of different viewing angles to find the optimal view that would provide the deepest measurement for any given region. However, any crack that propagate at an angle that exceeds the maximum angle of the cone beam (i.e. +/- 22.5 degrees) would yield greater depth measurement errors.

Penumbra width and blurring By placing the samples near the source we had a great advantage in resolution because of the intrinsic geometrical magnification of the system. However, this has also the adverse effect of widening the penumbras of the contrast changes in the images and ultimately blurring the images. Nevertheless, the overall quality of the resulting x-ray images and the fact that we were able to distinguish very small thin cracks from the background with very sharp edges demonstrates that blurring did not have much of an effect in this case.

Thickness calibration errors While our experimental setup and image cleaning techniques enabled us to acquire very repeatable data, it is clear that the conversion from gray value to intensity to thickness is prone to errors. Shot noise amplitude variations originating from a change in temperature of the combination scintillator-detector, ray-angle distribution within a sub-image used to create the stitched image and machining errors are all examples of what could go wrong with the thickness calibration. That’s why we used the distribution of depths within a ROI and repeated measurement and analysis to get the statistics and compute the uncertainties shown above.

CONCLUSIONS In conclusion, we were able to come up with an intensity to thickness curve that was self-consistent and repeatable. Since it was based on a gray value distribution, it provided statistics for error assessment and helped us to see the effect of parameter changes. We were able to use the best parameters for this study and test our thickness curve on a known-depth sample to assess its accuracy and precision. We then analysed real pipeline samples and found out the corrosion was clouding our results. By using deterministic radiation physics equations we were able to find a way to correct the depth of the cracks that were clouded by corrosion.

Our measurements were coherent with the Eddy Current analysis but delivered much more spatial resolution and could resolve cracks that were very close together. We think that the methods shown in this article are precise and repeatable enough to pave the way towards a calibration standard for the measurement of SCC crack depths in pipelines that does not involve sample destruction. In the future, we would like to use computer vision to build an algorithm that does the corrosion correction automatically. We also want to automate the whole pipeline sample scanning procedure and analysis so that anyone could easily use this method. Once the procedure is automated we intend to proceed with thorough testing with different materials and levels of corrosion. Finally, we would like to use destructive testing on a few samples to confirm our results and potentially find details that might have been overlooked.

ACKNOWLEDGEMENTS For the equipment, the use of custom made software and the financial support we would like to thank: LynX Inspection Inc. www.lynxinspection.com

For the opportunity, the samples and the Eddy Current analysis we want to thank: EddyFi Technologies www.eddyfi.com

I want to thank all the coworkers at LynX that helped with this research: Martin Lacasse, Dominique Boutet, Roger Booto, Charles Brillon and Vincent Boulet.

REFERENCES (1) Attix F., 2008, Introduction to Radiological Physics and Radiation Dosimetry, Wiley-VCH, New York, NY. (2) Hubbell J.H., Seltzer S.M., X-Ray Mass Attenuation Coefficients, NIST Standard Reference Database Number 126, National Institute of Standards and Technology, Gaithersburg, MD, 5632, https://dx.doi.org/10.18434/T4D01F, (retrieved 03-07-2019).

CombinedCombined NDT NDT Correlation Correlation to Estimate to Estimate the Compressive the Compressive Strength Streng of thConcrete of Concrete 1 2 1 2 2 AbhishekAbhishek ChittiChitti1,, SandeepSandeep G.G. BurraBurra2,, TsuchinTsuchin P. ChuChu1,, PrabirPrabir KolayKolay2,, andand SanjeevSanjeev KumarKumar2

1Department of Mechanical Engineering and Energy Processes Southern Illinois University Carbondale, IL 62901 (618(618)) 4412-1140;12-1140; faxfax (618)(618) 453453-7658;-7658; email:email [email protected]@siu.edu

2Department of Civil and Environmental Engineering Southern Illinois University Carbondale, IL 62901 (618)(618) 412412-1263;-1263; faxfax (618(618)) 453453-3044;-3044; email:email [email protected]@siu.edu

ABSTRACT This paper deals with estimating the compressive strength of concrete specimen using the combined method of ultrasonic pulse velocity (UPV) and rebound hammer (RH) tests. Compressive strength is an important parameter to evaluate concrete structures. Generally, the destructive methods like removing ‘core sample’ from an existing structures are considered as reliable methods to assess the quality of concrete already in place. However, these test methods are expensive and can be detrimental to the structure. Nondestructive tests (NDT) like UPV and RH tests are used to overcome these disadvantages. They are widely used to assess the quality of concrete. The individual test results from UPV or RH methods may not be reliable for estimating the compressive strength, as there are different factors like aggregate size, curing age, and curing conditions that influences the measurements. In this study, several concrete samples were casted based on three different mix designs with the targeted compressive strengths of 41 MPa, 55 MPa, and 83 MPa. All of the concrete samples were cured under laboratory conditions and tested after 28 days curing period. UPV and RH tests were performed followed by crushing the cylindrical samples to evaluate the compressive strength. Effect of moisture on UPV and RH measurements was studied and the results have shown that RH measurements are significantly affected by the moisture in concrete specimens. The results from both UPV and RH tests were combined and correlated to the measured compressive strength values. This study was focused on creating a correlation curve by combining the results from UPV and RH tests and then performing multiple regression analysis between UPV, RH, and the measured compressive strengths. The accuracy of this correlation curves was determined by comparing the estimated compressive strengths to the measured compressive strengths.

Keywords: Ultrasonic Pulse Velocity (UPV), Rebound Hammer (RH), Nondestructive Testing (NDT), Combined Method, Compressive Strength

INRODUCTION Concrete is the most common material used in construction industry. The need for in-situ testing of concrete is important for quality controlling and compliance purposes. Usually to find the strength of concrete core samples are drilled out of the existing structures which will damage the structure. The location at which core has been removed should be repaired which increases cost and labor. NDT methods can be employed to overcome all this issues. Many NDT methods are capable for both assessing and determining material properties and strength of concrete specimen. Based on the requirement, whether to find defects or to determine the material property or to assess the strength of the concrete structure, NDT method can be selected.

The UPV and RH methods have been widely used to assess the quality of concrete specimen. The UPV has been successfully used to evaluate quality of concrete by detecting internal cracks and other defects like voids. Development of Pulse velocity method has begun in Canada and England at about same time. Since the 1960’s, pulse velocity methods have moved out of laboratories and to construction sites [1]. Malhotra [2] has complied an extensive list of papers published on ultrasonic testing of concrete. Bungey [3] has conducted research on validation

of UPV testing of in-place concrete for strength and concluded that detailed knowledge on relative moisture conditions under test is vital in establishing correlation. Yaman et al. [4] have made comparisons between direct and indirect wave velocities and found that indirect wave UPV is statistically similar to direct UPV if there are uniform properties along the specimen. Helal et al. [5] have identified and described the most common successful methods of NDT as applied to concrete structures. It was found that the majority of NDT methods rely on comparing tested parameters with established correlations. Al-Nu’man et al. [6] have conducted an experimental research on 880 concrete samples received from various construction projects and found that no unique relation can be established to cover all concrete specimens. Azreen et al. [7] have conducted tests on concrete slabs with different grades of concrete including ultrahigh performance concrete (UHPC). It was observed from test results that UPV for UHPC was higher than normal grade concrete specimens. Mahure [8] has conducted research on developing UPV and strength relationship curves for different mixes of concrete used in concrete structures of hydroelectric project. The estimated correlation curves are verified to be suitable for prediction of strength in health monitoring during its service. Popovics and John [9] has studied the effect of stresses on ultrasonic pulse velocity measurements and found that the ultrasonic pulse velocity in concrete is independent of the stress level.

Rebound hammer has been used to find the hardness of concrete which is further correlated to compressive strength of concrete. Mitchell and Hoagland [10] have attempted to correlate rebound number with modulus of elasticity of the concrete specimens and concluded that no general valid correlation could be made between rebound number and statistic modulus of elasticity. Aydın and Mehmet [11] have conducted a study on concrete cube specimens of 28-90 days and a number of core samples from different reinforced concrete structures have been tested. The correction factors have been suggested to be used for strength correlation for old concrete. Sanchez and Tarranza [12] have investigated the reliability on rebound hammer as a means of estimating the compressive strength of three group of concrete cube specimens. The high dispersion of data for rebound number plotted against actual compressive strength has reinforced the prior findings that the rebound hammer test is not substitute for obtaining compressive strength if used alone. Kim et al. [13] have conducted experimental research to clarify the influence of carbonation on rebound number and concrete strength evolution. A new equation was developed from this research for strength reduction coefficient, which compensates the influence of surface carbonation in the rebound hammer method.

Many investigations have applied more than one NDT method at same time to predict strength of concrete more accurately. During 1950’s, Kesler and Higuchi [14] has pioneered in this field. During this study, dynamic modulus of elasticity and damping constant were determined from resonance tests and they correlated with compressive strength of concrete and found that this method was unsuitable for in-situ measurements. The results from UPV and RH methods have been combined to predict the compressive strength of the concrete. Many factors like moisture content, aggregate size content are likely to affect the reliability, sensitivity, and reproducibility of results. There are some exceptions in those cases when a variation in properties of concrete produce opposite effects on result of each component [1]. An increase in the moisture content increases UPV value but decreases the rebound number. RILEM Technical Committees 7 NDT and 43 CND [15] has put large efforts on developing the SONREB method. This method has been developed combining both ultrasonic pulse velocity and rebound hammer measurements to create a correlation curve. These curves can be used to estimate the compressive strength but the difference between estimated and actual strength is up to 20%.

This paper presents the prediction of compressive strength of concrete by comparing the results obtained from both nondestructive tests and destructive test. UPV, RH, and compression strength tests were performed on concrete samples of different compressive strengths. The influence of moisture content on UPV and RH test results was observed and explained. Multiple regression analyses were performed by combining the results from both nondestructive (UPV and RH) tests and destructive test (compression strength test) and a combined NDT correlation was developed. The accuracy of this correlation curve was determined by comparing the estimated compressive strength values to actual compressive strengths measured.

EXPERIMENTAL STUDY Materials In the present study, the cement used was a standard QUIKRETE Portland cement Type I/II from the QUIKRETE companies, Atlanta, USA. The natural coarse aggregate used in this study, is a crushed limestone rock and was sourced from Anna Quarry, Southern Illinois. The material was class CA 11 with size ranging from4.75 and 19 mm. The initial moisture content was 0.124% and specific gravity ranged between 2.6-3.15. The natural fine aggregate was also sourced from the same quarry. The initial moisture content was 0.080%. The gradation for fine and coarse aggregates were in between the ASTM C33 upper and lower limits, The admixture used in the study is a superplasticizer named Melchem-M manufactured by GRT, Inc. The purpose of this admixture is to achieve good workability of the concrete.

Mixtures Different mixes of concrete cylinders with dimensions of 4 inches × 8 inches (Figure 1) were fabricated based on three different concrete mixtures. These mix designs were targeted to attain compressive strengths of 41 MPa (6000 psi), 55 MPa (8000 psi), and 83 MPa (12000 psi), as presented in Table 1. Two batches of samples for each type of mix were casted and each batch consisted of 9 samples. As a total, 54 samples were casted for this study. All the mix designs used for concrete were according to the Illinois Department of Transportation’s (IDOT, 2012) method of concrete mixing and ACI 211 (2002) committee report which provides guideline for the selection of different proportions for standard concrete with Portland cement. The consistency of the fresh concrete mixture was assessed using the slump test. Then the fresh concrete was poured into cylindrical molds, compacted in three layers, and kept on vibratory platform to eliminate air voids. The molded cylinders were then left in the open air for 24 hours to set. After that period the cylinders were demolded, labeled, and cured in a control tank held at a temperature of 74 ˚F (24 ˚C). Three mixes of samples were tested for compressive strength after curing for 28 days.

Table 1: Descriptions of Sample Mixes of Concrete Targeted Compressive Strength, Mix MPa (psi) Mix 1 41 (6000) Mix 2 55 (8000) Mix 3 83 (12000)

(a) (b) Figure 1. Fabricating concrete samples. (a) Molded cylinders setting in Air. (b) Curing the samples in water.

Apart from above mentioned samples, another new batch of concrete with a targeted compressive strength of 55 MPa (8000 psi) was prepared and nine samples were casted. These samples were specifically casted to study the influence of moisture content on UPV and RH measurements.

Experimental Test Ultrasonic Pulse Velocity (UPV) Test Concrete Samples from each mix (see Table 1) were removed from the immersion tank after curing for a period of 28 days. These samples were allowed to dry (surface drying) under the ambient conditions in the testing room. The tests were performed once the surface of each sample was free from moisture. UPV tests on the concrete samples were performed once the surface of each sample was free from moisture. UPV tests were performed using the direct transmission mode (the ultrasonic signal passes directly from one transducer, through material, to another transducer). Both sides of the cylindrical sample are smoothened using concrete grinding wheel to ensure that surface is smooth and transducer has maximum contact to the concrete surface. Ultrasonic couplant gel was applied on opposite ends of each horizontal cylinder (Figure 2). The couplant ensures the even contact between transducers and surface of testing cylinder by filling any air gaps between them.

Figure 2. UPV testing of a concrete sample using direct transmission mode.

The UPV values were measured on samples from all three mixes, with the distance between transducers being measured separately and used as an input to the test. UPV tests were performed according to the ASTM C 597 standard [16]. We measured the propagation of longitudinal velocity of ultrasonic waves. This value is related to intrinsic material properties as shown in Eq. (1).

(1) 𝐸𝐸(1−𝑣𝑣) where is the acoustic velocity of the longitudinal𝑐𝑐𝐿𝐿 = waves,𝜌𝜌(1 + 𝑣𝑣is) the(1− material’s2𝑣𝑣) Young’s modulus, is the material’s Poisson’s ratio, and is the mass density. √ 𝐶𝐶𝐿𝐿 𝐸𝐸 𝑣𝑣

𝜌𝜌 Rebound Hammer (RH) Test The concrete samples were tested using a rebound hammer followed by UPV test after curing period of 28 days. This test was performed according to the ASTM C805 standard [17]. Ten readings, from a various locations on the concrete cylinder, were taken, and the average of these 10 values was considered to be the rebound number of that sample. No two readings were taken at identical locations. The test locations selected on the sample were free from visible pores and discontinuities. To make sure that sample was rigidly fixed, it was clamped using the compressive strength testing machine (Figure 3), which was set to the lowest pressure that could reliably hold the sample in fixed location. Same pressure was used to hold each cylinder.

Figure 3. Rebound hammer testing of a rigidly fixed concrete sample.

Compressive Strength Testing Compressive strength tests were performed on each sample from the three different mixes after performing both nondestructive tests. This test consisted of mounting each cylinder in the hydraulic press (already used to hold the sample for RH testing), then increasing the pressure until the concrete cylinder yielded (Figure 4). That pressure is defined as the compressive strength of that specific sample. This method was the only destructive method performed on the concrete samples after a standard curing period of 28 days.

Figure 4. Determining compressive strength of concrete cylinder.

Experimental Study on Influence of Moisture Content on UPV and RH measurements A new batch of 9 nine samples with targeted compressive strength of 55 MPa (8000 psi) are used for this study. Both surfaces of these horizontal concrete cylinders were grinded to attain smoother surface for UPV measurements. These samples were tested after a curing periods of 7, 15, and 28 days. Three samples were tested at each curing period. Out of three samples, one of the samples were tested for UPV, rebound number, and crushed for measuring compressive strength on the same day after taking out from the immersion tank. Whereas the remaining two samples were placed in oven for quick drying. Both the samples were taken out after 24 hours and allowed to cool down. Then UPV and RH tests was performed on both the samples and one of these samples was crushed to find compressive strength and another sample was placed back in oven for another 24 hours. The same procedure was repeated on third sample after 24 hours of drying in oven.

RESULTS AND DISCUSSION The experimental study has been conducted on 54 concrete cylinders with three different range of compressive strengths. 24 Experimental Results from UPV, RH, and Compressive Strength Tests After a standard curing period of 28 days, compressive strength tests were performed on all of the samples from different batches using the compressive strength test machine manufactured by ELE International. The compressive strength is presented by averaging the strength values obtained by 18 samples for each type of mix (Table 2 summarizes compressive strength results). Our experimental results indicated that Mix-1(6000 psi) had an average compressive strength of 43.5 MPa (6308 psi), Mix-2 (8000 psi) had an average compressive strength of 61.2 MPa (8871 psi), and Mix-3 (12000 psi) had an average compressive strength of 75.4 MPa (10942 psi). The test results of Mix-1 and Mix-2 have exceeded their targeted strength at 28 days. Mix-3, however, did not reach the targeted strength of 83 MPa but stopped at 75.4 MPa.

Table 2: Summary of an average compressive strength of concrete cylinders from different mixes. Targeted Compressive Strength , Average Compressive Strength Obtained Mix MPa (psi) psi MPa Mix-1 41 (6000) 6308 43.5

Mix-2 55 (8000) 8871 61.2

Mix-3 83 (12000) 10942 75.4

UPV and Rh tests were the nondestructive tests performed on concrete cylinders from three different mixes. All these measurements were performed in accordance with relevant standards - ASTM C597 [16] for UPV test and ASTM C805 [17] for RH test. The results obtained from UPV, RH, and compressive strength tests performed on Mix-2 and d Mix-3 are summarized in Table 3.

Table 3: Experimental results from UPV, RH, and Compressive strength tests. Measured Compressive Mix (Design Batch UPV (m/s) Rebound Number Strength Strength) psi MPa 4926 21 8667 59.8 4975 22 8538 58.9

4975 21 9150 63.1 4927 21 8740 60.3

1 - 4926 22 8768 60.5 4976 21 8283

2 (8000 psi) 2 (8000 57.1 Batch - 4939 22 8585 59.2 Mix 4939 23 8958 61.8 4951 22 8811 60.7 4950 22 9256 63.8

compressive strength and another sample was placed back in oven for another 24 hours. The same procedure was repeated on third sample after 24 hours of drying in oven.

RESULTS AND DISCUSSION The experimental study has been conducted on 54 concrete cylinders with three different range of compressive strengths. Experimental Results from UPV, RH, and Compressive Strength Tests After a standard curing period of 28 days, compressive strength tests were performed on all of the samples from different batches using the compressive strength test machine manufactured by ELE International. The compressive strength is presented by averaging the strength values obtained by 18 samples for each type of mix (Table 2 summarizes compressive strength results). Our experimental results indicated that Mix-1(6000 psi) had an average compressive strength of 43.5 MPa (6308 psi), Mix-2 (8000 psi) had an average compressive strength of 61.2 MPa (8871 psi), and Mix-3 (12000 psi) had an average compressive strength of 75.4 MPa (10942 psi). The test results of Mix-1 and Mix-2 have exceeded their targeted strength at 28 days. Mix-3, however, did not reach the targeted strength of 83 MPa but stopped at 75.4 MPa.

Mix-2 55 (8000) 8871 61.2

Mix-3 83 (12000) 10942 75.4

4975 21 9150 63.1 4927 21 8740 60.3

1 - 4926 22 8768 60.5 4976 21 8283

2 (8000 psi) 2 (8000 57.1 Batch - 4939 22 8585 59.2 Mix 4939 23 8958 61.8 4951 22 8811 60.7 4950 22 9256 63.8

4927 22 9043 62.3 4951 22 9055 62.4 4903 21 8883 61.2

2 - 4988 22 9306 64.2 4951 21 8404 57.9 Batch 4988 23 9151 63.1 4975 23 9354 64.5 4939 23 8725 60.2 4819 28 10667 73.5 4902 26 10697 73.8 4831 28 11529 79.5

1 4819 27 11063 76.3 - 4879 28 11373 78.4

Batch 4832 28 11339 78.2

4821 26 11136 76.8 4832 26 10192 70.3 4843 27 10741 74.1

3 (12000 psi) 3 (12000 4879 27 10882 75.0

– 4831 27 10869 74.9

Mix 4832 26 10487 72.3

4819 29 11063 76.3 - 4878 28 11207 77.3

Batch 4832 27 10892 75.1 4854 27 10769 74.2 4867 28 10957 75.5 4903 28 11097 76.5

Longitudinal (p-wave) velocities of the concrete cylinders, across different mixes, were measured during this study. The average values of UPV and rebound numbers measured for samples of Mix-1 are 4919 m/s and 20 respectively. The UPV values obtained from all samples of Mix-1 were higher than expected. The reason for this was found to be higher density of the samples. As samples from Mix-1 have been targeted to achieve conventional strength. The UPV values for conventional strength concrete are usually lesser than high strength concrete. Statistical analysis performed to obtain the relation between UPV and compressive strength did not show good fit when UPV values from Mix-1 are included. To obtain a better fit and relation between UPV and compressive strength the Mix-1 values are not considered for analysis.

The average values of UPV measured from all samples of Mix-2 and Mix-3 were 4950 m/s and 4849 m/s respectively. The UPV values ranged from 4800 – 5000 m/s, for samples with compressive strength ranging from 8000 – 12000 psi. The relatively high UPV values are due to the higher densities in high strength concrete. The relationship between measured UPV values and compressive strength was obtained (Figure 5). The equation to predict compressive strength from UPV value, shown in Eq. (2), was found. The R2 fit value, adjusted for degrees of freedom, was 68.3%, with errors up to ~10%. (2) where Y1 is predicted compressive strength (psi) and X1 is longitudinal velocity (m/s) of ultrasonic pulse. 𝑌𝑌1 = 87020 − 15.739 (𝑋𝑋1)

Figure 5: Relation between measured compressive strength and UPV values.

The average values of rebound number measured from all samples of Mix-2 and Mix-3 were 22 and 27 respectively. The rebound number values ranged from 20 – 30, for samples with compressive strength ranging from 8000 – 12000 psi. The relatively high rebound number values are due to the higher surface hardness for high strength concrete samples. The relationship between rebound numbers and measured compressive strength was obtained (Figure 6). The equation to predict the compressive strength from rebound number, shown in Eq. (3), was found. The R2 fit value, adjusted for degrees of freedom, was 92.3%. (3) where Y2 is predicted compressive strength (psi) and X2 is measured rebound number. 𝑌𝑌2 = 369.91 (𝑋𝑋2) + 812.81

Figure 6: Relation between measured compressive strength and rebound number.

Combined NDT Correlation A correlation was developed by performing multiple regression analyses using spreadsheet. A bi-linear equation was developed considering UPV and rebound numbers as independent variables and estimated compressive strength as the dependent variable. A nomogram (Figure 7) was developed by plotting the curves between rebound numbers and estimated compressive strengths at different UPV values ranging from 4800 – 5000 m/s. The R2 value of the bivariate fit is 92.4%, better than the fit either UPV or RH alone and the fit errors are ~3%. A significant improvement was achieved by combining the UPV and RH values to estimate the compressive strength.

Figure 7: Combined NDT Correlation curve to estimate compressive strength of concrete.

Plotting this in 3D (Figure 8), it can be seen that while the strength is modeled as linear in both UPV and RH, the variation with RH is much greater than that with UPV over the range of measurements considered.

Figure 8: 3-D fit to predict compressive strength of concrete with UPV and rebound number.

Influence of Moisture Content on UPV and RH measurements Another new batch of 9 samples with targeted strength of 8000 psi were used to study the influence of moisture content on UPV and RH measurements. The results from 7, 15, and 28 day tests (Table 4) show that there is a significant increase in rebound number values and any change in strength values. The results do not show any conclusive relationship between the presence of moisture and UPV measurements.

Table 4: Experimental results from samples with different drying condition Curing UPV Compressive Strength, Drying Condition Rebound Number Period (m/s) psi MPa Surface dried 4907 18 6691 46.1 7- Day Oven (24- hours) 4920 23 7135 49.2 Oven (48- hours) 4933 24 7331 50.5 Surface dried 5029 20 7325 50.5 15- Day Oven (24- hours) 5000 25 7246 50.0 Oven (48- hours) 4947 25 7568 52.2 Surface dried 5069 21 8161 56.3 28- Day Oven (24- hours) 4919 25 7762 53.5 Oven (48- hours) 4912 28 7808 53.8

CONCLUSIONS The experimental study was conducted to combine two different nondestructive tests to estimate the compressive strength of concrete and the conclusions made were:  Combining UPV and RH measurements produce more reliable results than UPV or RH tests alone.  An equation for combined NDT correlation was developed by performing multiple regression analyses in conjunction with UPV and rebound values. This equation was used to develop a nomogram. This nomogram has showed highest accuracy for high strength concrete ranging from 8000-12000 psi.  The study on moisture content in sample showed that there is a significant change in rebound numbers measured but no conclusive relationship between moisture content and UPV values could be drawn.

ACKNOWLEDGEMENTS This research work was funded by a Phase-I NAVY STTR grant (Topic#: N18A-T006). This project was supported by Luminit, LLC, California.

REFERENCES [1] Malhotra, V. M and Carino, N.J., CRC Handbook on Nondestructive Testing of Concrete. 2004. [2] Malhotra, V.M., “Testing Hardened Concrete: Nondestructive Methods”, ACI Monograph 9, American Concrete Institute, 1976. [3] Bungey, J. H., “The validity of ultrasonic pulse velocity testing of in-place concrete for strength,” NDT International, vol. 13, no. 6, pp. 296–300, Dec. 1980. [4] Yaman, I., Inci, G., Yesiller, N., and Aktan, H.M., “Ultrasonic Pulse Velocity in Concrete Using Direct and Indirect Transmission”, ACI Materials Journal, 98(6), pp. 450-457, 2001. [5] Helal, J., Massoud, S., and Priyan, M., “Non-Destructive Testing of Concrete: A Review of Methods”, Electronic Journal of Structural Engineering, 14(1), 2015. [6] Al-Nu’man, B.S., Bestoon, R. A., Sabr, A. A., and Sirwan, E. K., “Compressive Strength Formula for Concrete using Ultrasonic Pulse Velocity”, International Journal of Engineering Trends and Technology, vol. 1, Aug. 2015. [7] Azreen, M.N., Pauzi, I.M., Nasharuddin, I., Haniza, M.M., Akasyah, J., Karsono, A.D., and Yen Lei, V., “Prediction of Concrete Compression Strength Using Ultrasonic Pulse Velocity”, American Institute of Physics (2016). [8] Mahure, N.V., Correlation between Pulse Velocity and Compressive Strength of Concrete. 2018. [9] Popovics, S. and John, S. P., “Effect of stresses on the ultrasonic pulse velocity in concrete,” Materials and Structures, vol. 24, no. 1, pp. 15–23, Jan. 1991. [10] Mitchell, L.J. and Hoagland G.G., “Investigation of the Impact Tube Concrete Test Hammer”, Bull. No.305, Highway Research Board, 1961. [11] Aydın, F. and Mehmet, S., “Correlation between Schmidt Hammer and destructive compressions testing for concretes in existing buildings”, vol. 5. 2010. [12] Sanchez, K. and Nathaniel, T., “Reliability of Rebound Hammer Test in Concrete Compressive Strength Estimation”, International journal of Advances in Agricultural & Environmental Engineering (IJAAEE), 1(2), 2014. [13] Kim, J., Chin-Yong, K., Yi, S., and Lee, Y., “Effect of carbonation on the rebound number and compressive strength of concrete”, Cement and Concrete Composites, 31 (2), pp. 139–144, 2009. [14] Kesler, C.E. and Higuchi, Y., “Delamination of compressive strength of concrete by using its sonic properties”, Proc. ASTM, 53, 1044, 1953. [15] “In Situ Concrete Strength Estimation by Combined Non-destructive Methods”, RILEM Committee TC 43 CND, 1983. [16] ASTM C597-16, “Standard Test Method for Pulse Velocity Through Concrete”, ASTM International, West Conshohocken, PA, 2016. [17] ASTM C805 / C805M-13a, “Standard Test Method for Rebound Number of Hardened Concrete”, ASTM International, West Conshohocken, PA, 2013.

Evaluation of Nondestructive Evaluation Methods for Applicability to Precast Evaluation of NondestructiveConcrete Deck PanelsEvaluation Methods for Applicability to Precast Concrete Deck Panels

SamanSaman Farhangdoust Farhangdoust1 and1, Armin Armin Mehrabi Mehrabi2 2

11PhD Candidate, Accelerated Bridge Construction University Transportation Center (ABC-UTC)(ABC-UTC) Department of Civil and Environmental Engineering, Florida International University Miami, FL 33174, USA ((917)917) 755755-5697;-5697; email [email protected]@fiu.edu

2 2Associate Professor, Accelerated Bridge Construction University TransportationTransportation Center (ABC-UTC)(ABC-UTC) Department of Civil and Environmental Engineering, Florida International University Department of Civil and Environmental Engineering, Florida International University Miami, FL 33174, USA (305) 348-3653;Miami, FLemail 33174, [email protected] USA (305) 348-3653; email [email protected]

ABSTRACT Accelerated Bridge Construction (ABC) is the method for building and rehabilitating bridge construction aimed at reducing on-site activities, traffic interruptions, and cost. In general, ABC uses precast elements for bridge superstructure or substructures which are fabricated on site or away, moved to the bridge site and installed in place. Precast concrete deck panels are among common elements in ABC superstructures which need to be integrated together and with the bridge using closure joints. Because of in-situ casting of the closure joints therefore there is a potential for defects to occur during construction, or develop later during the life of the structure. Despite numerous investigations on nondestructive evaluation in general, a quantitative study on the evaluation and selection of the most applicable Nondestructive testing (NDT) methods for ABC precast concrete deck panels is lacking. The most promising NDT methods and their capabilities for inspection of the ABC joints were identified by the authors and presented previously in another paper. The current paper attempts to introduce a quantitative examination and comparison among the most applicable NDT techniques in ABC taking into account the expected defects and anomalies associated with precast concrete deck panels. The investigation is focused on closure joints as the critical details in a precast concrete deck.

Keywords: Accelerated Bridge Construction, Precast Deck Panels, Closure Joints, Nondestructive Evaluation, Bridge Inspection, Structural Health Monitoring, Damage Detection, Defect Etiology.

INTRODUCTION Accelerated Bridge Construction (ABC) is defined as design, planning and construction methods to organize and arrange construction activities for repair and rehabilitating of new or existing bridges so that in-situ construction time and traffic are reduced, and public safety is enhanced (1-2). ABC uses prefabricated elements of the bridge fabricated on site or away, moved to the bridge location and installed in place. ABC addresses major disadvantages of the conventional bridge construction techniques such as delays to allow concrete curing or traffic interruptions. One of the most common prefabricated elements used in ABC are precast concrete deck panels. Closure joints normally refer to joints for connecting the bridge deck components to each other and to the girders and substructure (Fig. 1). The closure joints have the potential for defects and damages to occur initially because of the in-situ casting, curing, material incompatibility, cold joints, cavities, and steel congestion. Hence, closure joint for precast concrete deck panels may introduce a potential for weak link within ABC structures.

Figure 1: Examples of ABC Closure Joints (1-3)

According to the study reported by Farhangdoust et al (4), the ABC closure joints have been categorized into the five distinctive groups as the most common classification as it relates to their features that affects the application of NDT. The five common groups of the ABC closure joints are shown in Table 1. The first four groups are “linear” joints which refer to longitudinal and transverse joints for connecting deck component to each other and to the girders and substructure. The fifth group of the closure joints shown in the table called “blockout” are mostly used to connect deck panels to the girders. Some uncommon ABC deck joints could not be categorized into the classification, and left to be considered on case-by case basis.

Table 1: The five common groups of the ABC Closure Joints (5) Cross-section Group Brief Explanation

Usually for joining full-depth precast deck panels, generally have diamond- Type 1 shape cavity for shear transfer, and are for both longitudinal and transverse joints. Connecting full-depth precast deck panels to each other, and precast decks to precast concrete and steel girders, distinguished by straight-sided interface. Type 2 Normally contains longitudinal and transverse reinforcement. When connecting slabs to girder, includes shear reinforcement projected from girders. Normally designed to provide continuity and shear/moment transfer in linear joints or link slabs. Normally contains longitudinal and transverse reinforcement, utilized for joining butted decked precast girders, partial depth precast deck panels, and Type 3 sometimes precast slab longitudinal connection to steel girders. Two dissimilar concrete layers are formed in the depth of this type of closure joint. Has V-shape cross section, used for connecting two pre-stressed tee beams or double beams, and sometimes full or partial depth deck panels. Mechanical Type 4 joints can be included in this group. Box/rectangular shaped joint, known as blockout, connects steel girders or concrete I-beams to precast full depth deck panels. Contains shear connectors Type 5 welded or embedded in the girders.

Non-destructive testing (NDT) methods have been used for detection and localization of structural damages in many engineering applications (6-9). A variety of Non-Destructive Testing (NDT) methods have been used for inspection of precast concrete deck panels so far. However, a quantitative study on methods most applicable to the closure

joints of the precast deck panels for taking into account the most common defects of the precast concrete deck panels is lacking. Because of their critical role in the establishment of precast concrete decks and their susceptibility to damages and defects, closure joints are the focus of this investigation. In this paper, a comprehensive literature search is performed to first identify the most promising NDT techniques and their respective capabilities for application to the closure joints of the precast concrete deck panels. Consequently, a statistical analysis of the applicability of NDT methods to specific types of defects and damages is carried out. The results of the study reported in this paper have been organized so that its outcomes would allow future development of field procedures, reporting techniques, and suitability for integration into bridge health monitoring programs.

EXPECTED DAMAGES Type of damage, plays an imp[ostant role in selection of the most applicable NDT method for inspection of the ABC closure joints. Damages can affect the structural performance of the ABC closure joints and consequently the service life of the precast concrete decks. Damages in the deck joints are generally expected to follow those observed for concrete deck construction. The evaluation of NDT methods is to be performed in relation with the type of deck joints and expected damages. Damages commonly reported for precast concrete deck panels include cracking, separation and delamination, voids and/or honeycombing filled with air or water, corrosion and loss of cross-section of reinforcing bars within the joints and their vicinity, leakage of surface water through joints and cracks, abnormal appearances and roughness. Defective closure joints may include various levels of damages, one caused by the other. As an example, mixing issues may cause exccessive shrinkage that in turn would be a cause for cracking. Cracking may lead to water leakage and result in corrosion of reinforcing bars. The sequence of the damages constructed for closure joints is shown in Figure 2. The relationships shown in this figure, can be used to support the selection of the proper NDT method and is effective application.

Figure 2: The common damage sequence of ABC closure joints

NONDESTRUCTIVE EVALUATION OF ABC PRECAST CONCRETE DECK PANELS Here it is attempted to compare several non-destructive techniques for the inspection of closure joints in precast concrete deck panels using accelerated bridge construction methods. A comprehensive investigation on different NDT techniques with applicability to closure joints in precast concrete deck panels has been performed by Farhangdoust et al. (4). Among the techniques evaluated, the following seven NDT methods have been identified as the most applicable methods with better ability for damage detection and inspection of the closure joints in precast concrete deck panels (5).

Impact Echo Testing (IE) Impact Echo Testing (IE) uses mechanical wave for inspection of cracks, voids, and delamination in the concrete deck panels (10). In IE, the surface of concrete deck is impacted by an impacting device and the energy of reflected wave is obtained by a receiver which is placed on the surface near the impact location (Figure 3). Although the cost of testing and the ease of use for this technique is rated as moderate, this method has an acceptable accuracy.

Figure 3: An schematic for IE method (10)

Radiographic Testing (RT) Radiographic Testing (RT) is based on X-Ray radiation for finding the internal defects in the concrete deck panels (11). In RT, transmission of radiation through the test element will be detected by photographic films (Figure 4). RT requires significant skills by the operator for its application. Access to both sides of the component is needed for damage detection by RT.

Figure 4: Schematic layout of RT (Courtesy of Bernoullies, 2011)

Ground Penetrating Radar Testing (GPR) Ground Penetrating Radar Testing or GPR is a microwave-based method. This NDT method is normally used for investigating the location of embedded steel reinforcement and post tensioning cables in concrete deck panels (Figure 5). This NDT method is applicable to detecting the delamination and voids by using the propagation of electromagnetic waves in the deck (12). It can also be used for detection of corrosion of steel reinforcement with noticeable progress. The GPR is cost effective, and good method for ease of use and its speed of data collecting.

Figure 5: an example of GPR technique (Courtesy of Echem Consultants LLC)

Impulse Response Testing (IRT) In Impulse Response Testing or IRT, stress waves are sent through the deck panels using a hammer (13). The IRT is a good NDT technique for investigation of defects such as honeycombs, delamination, voids, and cracks in concrete structure. IRT has a good ability for evaluation of deep foundation, for speed of data collecting.

Figure 6: A set-up of IRT application (13)

Ultrasonic Testing (UT) Ultrasonic Testing (UT) is a practical Pulse Velocity Testing (PVT) method in which transmission time, amplitude and frequency are the three important parameters (14-15). UT detects internal cracks in concrete decks panels by using sound waves (Figure 7) This method works based on the reflection of the sound waves and shows the distance of any internal defect from the surface (16,17). UT is less applicable to inspection of brittle and very thin elements, and also for complex geometry's components.

Figure 7: Schematic of UT method (17)

Infrared Thermography Testing (IR) Infrared thermography testing (IR) utilizes difference in the rate of heat emissivity within the concrete deck panels to detect cracks, delamination, and voids by an infrared camera which measures the emitted infrared radiation from a structural member (18). IR method can be divided into two classes of passive and active thermography. In the passive method, IR is performed without any external cooling or heating source.

Figure 8: An application of IR method (18)

Magnetic Flux Leakage Testing (MFL) Magnetic Flux Leakage (MFL) method magnetizes the steel by a strong external magnet to detect damages such as corrosion, loss of cross section, breaks, and pitting on steel elements (Figure 9). The MFl is used for sub-surface inspection of the reinforcing steel and rebar covered by concrete deck panels (5).

Figure 9: Schematic of MFL method (Courtesy of Ghortanpour).

QUANTITAITIVE EVALUATION OF THE APPLICABILITY OF NDT METHODS TO HEALTH MONITORING OF CLOSURE JOINTS In this section, the applicability of the most promising NDT inspection methods to health monitoring of closure joints in precast concrete deck panels are statistically investigated. This is to provide for a quantitative measure to be used in selection of the most appropriate NDT method taking into account the type of defects/damages anticipated for various types of joints. Among all possible damages for closure joints of the precast concrete deck panels, visible defects such as abnormal appearance (e.g., signs of leakage and efflorescence), surface defects, surface roughness, surface cracks, spalling of concrete cover, and exposure of reinforcing bars and embedment can be best detected using visual inspection. The potential defects that are not visible therefore can be listed as: Delamination of wearing surface, Delamination of concrete cover (before cracking and spalling becomes visible), Reflective cracks (for the extent of cracking inside the joint), Voids (internal), Honeycombing (internal), Debonding at cold joints (for the extent inside the joint), Concrete material segregation, Corrosion of reinforcing bars, Corrosion

of embedded steel. These defects are categorized in four general groups as the most common defect types that may occur in deck closure joints. 1. Delamination 2. Cracks (discontinuities of various orientations including debonding) 3. Voids (including internal honeycombing and segregation as variation in density) 4. Corrosion of embedded steel (including reinforcing bars, connectors, plates, and couplers)

The NDT methods most applicable to these four defect types of each of five groups of closure joints are evaluated quantitatively in Table 2. A total of 50 sources were reviewed to determine the use of specific NDT methods and their potential for detection of defects in concrete described above as 1) delamination, 2) crack, 3) void, and 4) corrosion of embedded steel. To derive a quantitative measure for comparison the most applicable NDT methods, the results of the literature search were analyzed to find number of sources who identified a method for applicability to a defect type. The criteria for advantage in selection of each method was set as the number of citations prescribing the method for the specific defect type. Information for each defect type is summarized in Table 2.

Table 2: Statistical evaluation of applicability of NDT methods to the common defects in ABC closure joints N M IE [26][27][29][30][40][41][43][44][45][48][53][55][56][61][62][20][68] 17 34% [25][24][26][27][29][31][32][33][34][35][38][40][43][45][46][48][50][51] GPR 35 70% [52][53][55][56][57][59][60][61][62][63][64][65][23][22][20][19][66] UT [40][48][55][58][61][64][20][68] 8 16% IR [26][27][28][29][30][48] [33][41][42][51][53][54][59][60][61][63][65][22][67] 19 38%

Delamination IRT [40][48][51][21] 4 8% RT [32] 1 2% MFL --- 0 0% IE [26][56] 2 4% GPR [24] [26][27][62][31][34][37][38][44][45][48][55][20][19] 14 28%

UT [47][56] 2 4% IR [27] 1 2%

Corrosion IRT --- 0 0% RT --- 0 0% MFL [28] 1 2% IE [27][29][47][48][55][61] 6 12% GPR [34][35][52][55][19] 5 10%

UT [26][28][32][36][39][40][48][53][55][58][60][61][64][23][20] 15 30% IR [28][29][47][48][61][65][20][67] 8 16% Crack IRT [47][48][51] 3 6% RT [26][47] 2 4% MFL [49] 1 2% IE [26][28][29][47][48][53][61][62][20] 9 18% GPR [26][28][29][32][36][46][48][50][53][60][65][23][20][19] 14 28%

UT [26][39][40][53][61][23][20] 7 14% IR [26][29][42][47][48][51][60][20] 8 16% Void IRT [47][48][51][21] 4 8% RT [26][32][47][53] 4 8% MFL --- 0 0%

The table lists the four groups of expected defects for closure joints and the most promising NDT methods recognized for closure joints. The references which have identified each NDT method for applicability to certain type of defect are listed in the table in a row corresponding to the NDT method and the row corresponding to the type of defect. N and M also refer to the total number of sources (for each NDT method applicable to the type of defect) and percentage of number of sources in comparison with the total sources cited, respectively.

The results are also illustrated in Figure 10. This figure introduces the NDT method that is most appropriate for each of four damage type in ABC deck joints denoted by the higher percentage(s).

A B

C D

Figure 9: Statistical Evaluation of the NDT methods to delamination (A), corrosion (B), cracks (C), voids (D)

CONCLUSION Accelerated Bridge Construction (ABC) promises to reduce on-site construction time and mobility impact in bridge construction and rehabilitation projects by the use of special design and construction methods. Generally, it uses precast elements of the bridge fabricated on site or away, moved to the bridge location and installed in place. Precast concrete deck panels are among common elements in ABC superstructures which need to be integrated together and with the bridge using closure joints. Deck joints are normally referred to as “Closure Joints.” Because of cast-in- place nature of closure joints that are expected to go into service rapidly and problems observed for some types of closure joints, there have been some concerns about their long-term durability. Closure joints have presented themselves as potential weak link in precast concrete decks, therefore warranting application of health monitoring and NDT methods to evaluate their state immediately after construction and at regular intervals thereafter. A variety of Non-Destructive Testing (NDT) methods have been evaluated for inspection of precast concrete deck panels so far. However, a quantitative study on methods most applicable to the closure joints taking into account the most common defects for these elements was lacking. In this paper, a comprehensive review of previous work was performed to identify the most promising NDT techniques and their respective capabilities for application to the closure joints of the precast concrete deck panels. Consequently, a statistical analysis of the applicability of NDT methods to specific types of defects and damages was carried out. This evaluation provides for a much needed quantitative measure showing advantages of certain NDT methods for health monitoring of various types of joints according to their expected type of damages or defects. This results of the study reported in this paper have been

organized so that its outcomes would allow future development of field procedures, reporting techniques, and suitability for integration into bridge health monitoring programs.

ACKNOWLEDGEMENT This paper is based on a project supported by the US Department of Transportation (Grant No. DTRT13-G-UTC41) through the Accelerated Bridge Construction University Transportation Center (ABC-UTC) at Florida International University. The opinions, findings, and conclusions expressed here are those of the author(s) and not necessarily of the sponsor.

REFERENCES (1) Culmo, M. P. (2011). Accelerated bridge construction-experience in design, fabrication and erection of prefabricated bridge elements and systems (No. FHWA-HIF-12-013). (2) Culmo, M. P. (2009). Connection details for prefabricated bridge elements and systems (No. FHWA-IF-09-010). United States. Federal Highway Administration. Office of Bridge Technology. (3) Rowekamp P., “ABC-UTC Innovative Projects – Bridge over Paleface River”, Bridge Standards and Research Engineer, Minnesota Department of Transportation, Report number: MN – TH 53, page 3, 01-03-13. (4) S. Farhangdoust, A.B. Mehrabi, S.F. Al Mosawi, “NDT Methods Applicable to Health Monitoring of ABC Closure Joints,” 27th Research Symposium - The American Society for Non-destructive Testing (ASNT), Orlando, Fl, 26-29 March, 2018. (5) S. Farhangdoust, A.B. Mehrabi, “Developing Defect Etiology to Facilitate NDE of Full-Depth Precast Concrete Deck Panels in Accelerated Bridge Construction,” ASNT Research Symposium 2019 - The American Society for Non-destructive Testing (ASNT), Garden Grove, CA, 1 - 4 April 2019. (6) Farhangdoust, S., Mehrabi, A., & Younesian, D. (2019). Bistable wind-induced vibration energy harvester for self-powered wireless sensors in smart bridge monitoring systems. SPIE Smart Structures + Nondestructive Evaluation. 3-7 March 2019. Denver. Colorado. United States. (7) Tashakori, S., Farhangdoust, S., Baghalian, A., Tansel, I. N., & Mehrabi, A. (2019). Evaluating the performance of the SuRE method for inspection of bonding using the COMSOL finite element analysis package. SPIE Smart Structures + Nondestructive Evaluation. 3-7 March 2019. Denver. Colorado. United States. (8) A. Mehrabi and S. Farhangdoust, “A Laser-Based Noncontact Vibration Technique for Health Monitoring of Structural Cables: Background, Success, and New Developments”, Advances in Acoustics and Vibration, Vol. 2018, 2018. (9) Farhangdoust, S., Tashakori, S., Baghalian, A., Mehrabi, A., & Tansel, I. N. (2019). Prediction of damage location in composite plates using artificial neural network modeling. SPIE Smart Structures + Nondestructive Evaluation. 3-7 March 2019. Denver. Colorado. United States. (10) N. Gucunski, S. Nazarian, H. Wiggenhauser, and D. Kutrubes, "Nondestructive Testing to Identify Concrete Bridge Deck Deterioration," SHRP 2–FEHRL Workshop, TRA 2010 Brussels, Belgium, 2010. (11) Taskin, M., Elazig, U. C., & Turkmen, M. (2011). X-ray tests of AISI 430 and 304 stainless steels and AISI 1010 low carbon steel welded by CO2 laser beam welding. Materials Testing. 53(11-12). 741-747. (12) Parrillo, R., Roberts, R., & Haggan, A. (2006, September). Bridge deck condition assessment using ground penetrating radar. In ECNDT Conference Proceeding, Berlin, Germany (pp. 25-29). (13) System Reference Manual 2008 Slab Impulse Response,’ 2008, Manual. [Online]. Available: www.olsonengineering.com [Accessed: 10-Nov-2017]. (14) Farhangdoust, S., Younesian, D., & Esmailzadeh, E. (2017, August). Interaction of higher modes in nonlinear free vibration of stiffened rectangular plates. In ASME 2017 International Design Engineering Technical Conferences and Computers and Information in Engineering Conference (pp. V008T12A043-V008T12A043). American Society of Mechanical Engineers. (15) Hosseinkhani, A., Younesian, D., & Farhangdoust, S. (2018). Dynamic analysis of a plate on the generalized foundation with fractional damping subjected to random excitation. Mathematical Problems in Engineering, 2018.

(16) Tashakori, S., Baghalian, A., Senyurek, V. Y., Farhangdoust, S., McDaniel, D., & Tansel, I. N. (2018). Composites Bond Inspection Using Heterodyne Effect and SuRE Methods. Shock and Vibration, 2018. (17) Alobaidi, W. M., Alkuam, E. A., Al-Rizzo, H. M., and Sandgren, E. (2015). Applications of Ultrasonic Techniques in Oil and Gas Pipeline Industries: A Review. American Journal of Operations Research. 5( 274- 287). (18) Mehrabi, A. B. (2006). In-service evaluation of cable-stayed bridges, overview of available methods and findings. Journal of Bridge Engineering, 11(6), 716-724. (19) Sun, H., Pashoutani, S., & Zhu, J. (2018). Nondestructive Evaluation of Concrete Bridge Decks with Automated Acoustic Scanning System and Ground Penetrating Radar. Sensors, 18(6), 1955. (20) Lin, S., Meng, D., Choi, H., Shams, S., & Azari, H. (2018). Laboratory assessment of nine methods for nondestructive evaluation of concrete bridge decks with overlays. Construction and Building Materials, 188, 966- 982. (21) Davis, A. G. (2003). The nondestructive impulse response test in North America: 1985–2001. NDT & E International, 36(4), 185-193. (22) Washer, G., Fenwick, R., Bolleni, N., & Harper, J. (2009). Effects of environmental variables on infrared imaging of subsurface features of concrete bridges. Transportation Research Record, 2108(1), 107-114. (23) Maierhofer, C. (2003). Nondestructive evaluation of concrete infrastructure with ground penetrating radar. Journal of Materials in Civil Engineering, 15(3), 287-297. (24) Tarussov, A., Vandry, M., & De La Haza, A. (2013). Condition assessment of concrete structures using a new analysis method: Ground-penetrating radar computer-assisted visual interpretation. Construction and Building Materials, 38, 1246-1254. (25) Huston, D., Hu, J. Q., Maser, K., Weedon, W., & Adam, C. (2000). GIMA ground penetrating radar system for monitoring concrete bridge decks. Journal of Applied Geophysics, 43(2-4), 139-146. (26) "Guidebook on non-destructive testing of concrete structures." Training Course Series. International Atomic Energy Agency. Vienna, 2002. (27) Scott, M., Rezaizadeh, A., Delahaza, A., Santos, C. G., Moore, M., Graybeal, B., & Washer, G. (2003). A comparison of nondestructive evaluation methods for bridge deck assessment. NDT & E International, 36(4), 245-255. (28) Chase, S. B., & Washer, G. (1997). Nondestructive evaluation for bridge management in the next century. Public Roads, 61(1). (29) Yehia, S., Abudayyeh, O., Nabulsi, S., & Abdelqader, I. (2007). Detection of common defects in concrete bridge decks using nondestructive evaluation techniques. Journal of Bridge Engineering, 12(2), 215-225. (30) Oh, T., Kee, S. H., Arndt, R. W., Popovics, J. S., & Zhu, J. (2012). Comparison of NDT methods for assessment of a concrete bridge deck. Journal of Engineering Mechanics, 139(3), 305-314. (31) Sbartaï, Z. M., Laurens, S., Balayssac, J. P., Arliguie, G., & Ballivy, G. (2006). Ability of the direct wave of radar ground-coupled antenna for NDT of concrete structures. NDT & e International, 39(5), 400-407. (32) Büyüköztürk, O. (1998). Imaging of concrete structures. Ndt & E International, 31(4), 233-243. (33) Rhazi, J. (2000). NDT in civil engineering: the case of concrete bridge decks. CSNDT JOURNAL, 21(5), 18-25. (34) Barnes, C. L., Trottier, J. F., & Forgeron, D. (2008). Improved concrete bridge deck evaluation using GPR by accounting for signal depth–amplitude effects. NDT & E International, 41(6), 427-433. (35) Alani, A. M., Aboutalebi, M., & Kilic, G. (2014). Integrated health assessment strategy using NDT for reinforced concrete bridges. NDT & E International, 61, 80-94. (36) Kohl, C., & Streicher, D. (2006). Results of reconstructed and fused NDT-data measured in the laboratory and on-site at bridges. Cement and Concrete Composites, 28(4), 402-413. (37) Rhazi, J., Dous, O., & Laurens, S. (2007, December). A new application of the GPR technique to reinforced concrete bridge decks. In Proceeding of the 4th Middle East NDT conference and Exhibition, Manama, Kingdom of Bahrain (pp. 2-5).

(38) Rhazi, J., Dous, O., Ballivy, G., Laurens, S., & Balayssac, J. P. (2003, September). Non destructive health evaluation of concrete bridge decks by GPR and half cell potential techniques. In International Symposium on Non-Destructive Testing in Civil Engineering. (39) Maierhofer, C., Zacher, G., Kohl, C., & Wöstmann, J. (2008). Evaluation of radar and complementary echo methods for NDT of concrete elements. Journal of Nondestructive Evaluation, 27(1-3), 47. (40) Lim, M. K., & Cao, H. (2013). Combining multiple NDT methods to improve testing effectiveness. Construction and Building Materials, 38, 1310-1315. (41) Krause, M., Mielentz, F., Milman, B., Müller, W., Schmitz, V., & Wiggenhauser, H. (2001). Ultrasonic imaging of concrete members using an array system. NDT & E International, 34(6), 403-408. (42) Abdel-Qader, I., Yohali, S., Abudayyeh, O., & Yehia, S. (2008). Segmentation of thermal images for nondestructive evaluation of bridge decks. Ndt & E International, 41(5), 395-405. (43) Gucunski, N., Romero, F., Kruschwitz, S., Feldmann, R., Abu-Hawash, A., & Dunn, M. (2010). Multiple complementary nondestructive evaluation technologies for condition assessment of concrete bridge decks. Transportation Research Record: Journal of the Transportation Research Board, (2201), 34-44. (44) Arndt, R., & Jalinoos, F. (2009). NDE for corrosion detection in reinforced concrete structures–a benchmark approach. Proceedings of the Non-Destructive Testing In Civil Engineering (NDTCE'09), Nantes, France, 30. (45) Gucunski, N., Romero, F., Kruschwitz, S., Feldmann, R., & Parvardeh, H. (2011). Comprehensive bridge deck deterioration mapping of nine bridges by nondestructive evaluation technologies (No. Project SPR-NDEB (90)-- 8H-00). Iowa. Dept. of Transportation. (46) Bungey, J. H. (2004). Sub-surface radar testing of concrete: a review. Construction and Building materials, 18(1), 1-8. (47) Akhtar, S. (2013). Review of nondestructive testing methods for condition monitoring of concrete structures. Journal of construction engineering, 2013. (48) Rehman, S. K. U., Ibrahim, Z., Memon, S. A., & Jameel, M. (2016). Nondestructive test methods for concrete bridges: A review. Construction and Building Materials, 107, 58-86. (49) Krause, H-J., W. Wolf, W. Glaas, E. Zimmermann, M. I. Faley, G. Sawade, R. Mattheus, G. Neudert, U. Gampe, and J. Krieger. (2002). SQUID array for magnetic inspection of prestressed concrete bridges. Physica C: Superconductivity, 368(1-4), 91-95. (50) Belli, K., Wadia‐Fascetti, S., & Rappaport, C. (2008). Model based evaluation of bridge decks using ground penetrating radar. Computer‐Aided Civil and Infrastructure Engineering, 23(1), 3-16. (51) Bungey, J. H., & Grantham, M. G. (2014). Testing of concrete in structures. Crc Press. (52) Alani, A. M., Aboutalebi, M., & Kilic, G. (2013). Applications of ground penetrating radar (GPR) in bridge deck monitoring and assessment. Journal of Applied Geophysics, 97, 45-54. (53) Zhu, J. (2008). Non-contact NDT of concrete structures using air coupled sensors. Newmark Structural Engineering Laboratory. University of Illinois at Urbana-Champaign. (54) Omar, T., & Nehdi, M. L. (2017). Remote sensing of concrete bridge decks using unmanned aerial vehicle infrared thermography. Automation in Construction, 83, 360-371. (55) Gucunski, N., Kee, S. H., La, H., Basily, B., Maher, A., & Ghasemi, H. (2015, April). Implementation of a fully autonomous platform for assessment of concrete bridge decks RABIT. In Structures Congress 2015 (pp. 367- 378). (56) Gucunski, N., Kee, S., La, H., Basily, B., & Maher, A. (2015). Delamination and concrete quality assessment of concrete bridge decks using a fully autonomous RABIT platform. Structural Monitoring and Maintenance, 2(1), 19-34. (57) Van der Wielen, A., Courard, L., & Nguyen, F. (2010, June). Nondestructive detection of delaminations in concrete bridge decks. In Ground Penetrating Radar (GPR), 2010 13th International Conference on (pp. 1-5). IEEE. (58) Shokouhi, P., Wolf, J., & Wiggenhauser, H. (2013). Detection of delamination in concrete bridge decks by joint amplitude and phase analysis of ultrasonic array measurements. Journal of Bridge Engineering, 19(3), 04013005.

(59) Washer, G. A. (1998). Developments for the non-destructive evaluation of highway bridges in the USA. NDT & E International, 31(4), 245-249. (60) Hearn, G., & Shim, H. S. (1998). Integration of bridge management systems and nondestructive evaluations. Journal of Infrastructure Systems, 4(2), 49-55. (61) Rens, K. L., Nogueira, C. L., & Transue, D. J. (2005). Bridge management and nondestructive evaluation. Journal of performance of constructed facilities, 19(1), 3-16. (62) Huston, D., Cui, J., Burns, D., & Hurley, D. (2011). Concrete bridge deck condition assessment with automated multisensor techniques. Structure and Infrastructure Engineering, 7(7-8), 613-623. (63) Maser, K. R., & Roddis, W. K. (1990). Principles of thermography and radar for bridge deck assessment. Journal of transportation engineering, 116(5), 583-601. (64) Toutanji, H. (2000). Ultrasonic wave velocity signal interpretation of simulated concrete bridge decks. Materials and Structures, 33(3), 207. (65) Vaghefi, K., Oats, R.C., Harris, D.K., Ahlborn, T.T.M., Brooks, C.N., Endsley, K.A., Roussi, C., Shuchman, R., Burns, J.W. and Dobson, R. (2011). Evaluation of commercially available remote sensors for highway bridge condition assessment. Journal of Bridge Engineering, 17(6), 886-895. (66) Xu, Y., & Turkan, Y. (2019). Bridge Inspection Using Bridge Information Modeling (BrIM) and Unmanned Aerial System (UAS). In Advances in Informatics and Computing in Civil and Construction Engineering (pp. 617-624). Springer, Cham. (67) Chen, S., Laefer, D. F., Mangina, E., Zolanvari, S. I., & Byrne, J. (2019). UAV Bridge Inspection through Evaluated 3D Reconstructions. Journal of Bridge Engineering, 24(4), 05019001. (68) Bogue, R. (2018). Applications of robotics in test and inspection. Industrial Robot: An International Journal, 45(2), 169-174.

41 PlatformPlatform forfor TestTest Scripting and Automated Testing ofof SmartSmart MetersMeters

Vaibhav Garg

Genus Power Infrastructures Limited Jaipur,Jaipur, Rajasthan, Rajasthan India email [email protected] [email protected]

ABSTRACT We have come up with a unique and tailor-made platform for the functional testing of Smart meters. This allows test staff with no programming experience to create highly complex test cases and test the Smart Meters autonomously.

Smart meters are one of the most vital cogs of Smart and connected Grid. Smart meters are essentially Internet connected smart devices, which communicate bi-directionally with a back end server, using a basket of technologies. Add to that the requirements for various combinations of electrical data, control, anomaly and fraud detection mechanisms and operating ranges, you are staring at very large test matrix corresponding to all the different combinations.

To add to the challenge above, Smart electric meters are designed and required to be non-serviceable, such that the integrity of the device in the field is undisputed and the data cannot be repudiated. The devices are widely deployed, at every metering point; each home, office, factory, shop; each installation uniquely configured in some unique way. Needless to say, the escapes to customer are both high impact and have high likelihood given the flux of change and that very large test matrix. The embedded and application software is designed to be resilient, reliable and must maintain data integrity in all conceivable field scenarios. Any changes late in the project life cycle introduce a huge risk of regression and project cost overruns due to regression testing.

Our platform allows the test staff to create test scripts. We decided that the most intuitive way of slicing the elephant with regards to module design would be to center the task modules around interfaces to the Smart Meter. The Interfaces to a Smart meter can either be electrical (Voltages, currents, frequency, power factors, wave shapes et. el.) or logical (Reading a certain value, writing a certain value), or User interface related (Smart meter Display- interpreted using a state-of-the-art Convolutional Neural Network [1], buttons et. el.). We abstracted these interfaces out and developed modules exercising each of these interfaces. Additionally, we added some modules for imperative programming constructs such as variables and constants, delay, looping, computations, file I/O and others. The connecting glue to all this is a sequencer, which can be used to put these building blocks in any sequence to run. Essentially, the test script creator drags and drops the relevant module in the script, configures the module using a GUI, and saves the test script. The saved script can be run at any point of time in the future. The test script creates a detailed audit trail to facilitate debugging if and when defects are found.

Keywords: Smart Meters, Scripting, Deskilling, Fuzz testing, Continuous Integration, test framework, test sequencer

Smart Meters [2] Smart meters are devices which measure, communicate and control the flow of energy in a system. The energy can be in the form of electricity, gas or heat. Our focus will primarily be on the electrical smart meters. The design of an electrical smart meter is best described using the following functional block diagram.

Energy meter functional details

At the heart of the modern smart meter is the Microprocessor that has a complex firmware sitting on it. The Firmware is responsible for all the functionality of the smart meter. The other blocks of the smart meter are

1. Electrical inputs, where the voltages and current inputs come in. These inputs are also responsible for powering up the smart meter, with strong constraints on the power consumption. 2. Highly accurate ADCs- to convert the analog electrical inputs to digital domain, so that the DSP algorithms in the microprocessor can process these in real time. These need to have very low values of Integral and differential non-linearities. 3. Data Storage- A smart meter stores a full log and a large gamut of all electrical and environmental parameters to communicate the same with the head end systems/ servers. 4. An accurate Real Time clock for time synchronization of the data and performing actions that are based on the real time, such as billing, anomaly detection and logging, load profiles and others. 5. A regulated power supply- which powers up the meter’s electronics. This block also covers the battery backup systems that must support all other functionality in the absence of electrical inputs. The battery is typically non-replaceable. 6. A display- as a user interface element for the user to view the metered parameters. 7. Buttons- as user interface devices. 8. Metrology test outputs- Precise potential free, typically optical, outputs to measure the meter accuracy. 9. Tampering detection devices and sensors- which are application specific sensors and devices for the detection of various environmental and electrical tamper attempts. Complex algorithms synthesize these and other correlated electrical inputs to detect tamper events. Both type 1 and type 2 errors can have huge consequences in this area.

43 Motivation for Auto test The vision of this product is to enable automated and complete functional testing of the Energy meter, with a special focus on the firmware aspect. The overarching idea is to be able to use a completely configurable power source with scriptable software, to generate and maintain a set of tests that enable functional coverage of all aspects of the metering solution. The motivation for AutoTest solution has both been driven by the specific challenges related to the product space that we operate in, as well as operational challenges. Some of the product related challenges are 1. Smart meters are customized devices, with orthogonal customization axes designated by

o Electrical installation aspects, such as max currents, voltages, number of phases etc. o Information needs of the end user, dictating what data to be captured and retained and communicated

o Environmental constraints and variations, such as outdoor vs indoor installations o Tariff and regulatory aspects. For example, a commercial establishment cannot use the same smart meter that can be used by a domestic customer.

o Utility specific experiences and their client base o Cost constraints. This leads to a lot of flux in the design of the device, carrying the risk of regression with each change. 2. Smart meters are among the few sophisticated equipment, which are possibly installed under the custody of a malicious agent who has complete physical access to the device.

This implies some unique reliability and robustness challenges, which add to the test matrix. The embedded and application software is designed to be resilient, reliable and must maintain data integrity in all conceivable field scenarios. 3. Smart meters are non-serviceable and no-upgradable by law, once sealed in the factory cannot be opened during the entire lifetime of the device. This ensures the non-repudiation of the data in the court of law. 4. The devices are widely deployed, at every metering point; each home, office, factory, shop; each installation uniquely configured in some unique way. Acting almost like a Cash-Box for the utility, any data corruption issues with the smart meters have a direct impact on the revenue of the electrical utilities. To add to these challenges, several operational constraints dictated the design of the AutoTest solution 1. A lot of tests that we perform have been refined by our years of experience in the domain and need to be performed in a standard way to have an appropriate detectability to effort ratio. For a lot of tests, this ratio was really low, due to high effort requirements to perform the steps of the test. Our test engineers were effort constrained and could not find time to use their ingenuity to perform exploratory testing. 2. There is a set of tests that is directly dictated by the relevant standards and have to be performed at each delta change. These tests often take a lot of time to perform, but failure to comply can be catastrophic. 3. In case of a late delta change, it was often impossible to run the whole suite of tests again given the tight schedule constraints. This led to an increased risk of regression. 4. Owing to the above constraint, the nature of work for the test personnel was largely repetitive and dissatisfying. 5. It was often hard to ensure that the tests were performed in exactly the same way every time.

44 6. If a bug was detected by a certain test, the audit trail was often not adequate for the developers to reproduce the issue. 7. Test steps and test data storage of shipped releases could not have been comprehensive enough to investigate the behavior of the devices deployed in the field. 8. There was no logical separation between the personnel creating the complex test cases and the personnel performing those tests. This led to pockets of knowledge which constituted an organizational risk. 9. The maximum theoretical utilization of the working day was capped at ~30%. Most of the above challenges have been addressed to a large extent by the AutoTest Solution.

Design of auto test software Based on the challenges above, the design of the AutoTest solution was undertaken. The design philosophy behind the solution can be understood based on multiple dimensions.

Dimesion 1: Optimization of the effort required from skilled resources The typical and obvious workflow for any test regimen is to create the tests, perform the tests and to analyze the results of the tests performed. The skill and effort distribution of the steps in this workflow can be represented as the below.

Figure 2: Effort-Skill Matrix for manual testing Figure 3: Effort-Skill Matrix for AutoTest based testing When performing the tests manually (see figure 2), everything needs a high degree of skill, especially running the tests, which is both high effort and high skill job. This is due to the lack of decoupling between creating a complex set of steps for a test case and being able to run those steps with requisite fidelity. Besides, ‘running the tests’ step has be repeated for each project and each change in each project. This is the bulk of the effort required. The design philosophy of the AutoTest solution is to be able to reduce this dramatically for maximal impact.

As is evident from figure 3, by developing the AutoTest, the “Run Tests” step needs to be completely deskilled and the efforts reduced dramatically. Also, this entails a slight increase in efforts required to script the test cases and requires scripting skills. The skilled personnel can now be engaged in creating detailed test plans based on previous manually performed tests and detected issues in the field. Analyzing the test reports requires less effort now as the AutoTest solution maintains a detailed execution trail of the tests, with debug information. Dimension 2: Selection of primitives for scripting around interfaces There had been many attempts at determining the right set of basic primitives and developing the sequencer to create test cases. However, the basic challenge here is the tradeoff between the level of abstraction and the express-ability of the script being designed.

45 Erring on the side of too much abstraction led to creation of canned-test cases with very little flexibility to come up with new test sequences without going back to the original developer to add those, which doesn’t scale. Too little abstraction leads to the test engineer being forced to write their own test scripts from scratch and needs a lot of time and a separate set of coding skills. This is the challenge with the more general-purpose platforms such as National Instruments test suite. To break the conundrum, we decided to design the scripting primitives (Actions) around the smart meter’s interfaces, which are highlighted in blue in figure 2. These are, 1. Electrical Interface – which includes the Voltage, Current, Power factor, Frequency, and Wave Shape. 2. Logical Interfaces – Communications using the DLMS protocol, which is now standard across India and large parts of Europe. 3. User Interface – Display, Buttons for display navigation. 4. Optical Interface – which includes the calibration/ metrological LED outputs.

The other primitives, to complete the scriptability part of the AutoTest platform were looping, delays, registers (constants and assignable variables), Computation (with unary operators to act on one register, such as square root; and binary operators which act upon a pair of registers, such as add, subtract, multiply etc.), and compare (which gives a binary verdict). Programmers love writing comments in their code (unfortunately, not true), and we also added a component to facilitate writing comments in the test scripts. Each of these actions, or primitives, are a fully self-contained object in the OOPs paradigm. Each of these objects have attributes that are editable using a UI, both at design time and –in selected cases – at execution time. Some of the important actions are discussed in the section below. Once the action primitives are selected and configured, they are compiled into a binary blob and stored in the central database. These are retrieved, compiled and run by the execution engine JIT. • Delay: (Software Operation) Creates a software delay for the specified number of seconds. This tells the software not to send any instructions to either the bench or the meter for the specified amount of time.

• Power Source: (Write Category Operation) This function is used to pass instructions to the test bench for setting electrical parameters.

46 • Loop: (Software Operation) This function is used to repeat a certain list of instructions. This function, by default, loops all the functions preceding this function for the No. Of Repetition set on the form. The List of Component parameter can be modified to adjust the list of parameters that are to be looped.

• Read: (Read Category Operation) This function is used to request a parameter from the meter over UART. The Parameters dropdown provides list of parameters available for download. The downloaded parameter value is stored in the register specified by Output.

• Set RTC: (Write Category Operation) This operation is used to set the RTC of the meter. The RTC can be incremented by the specified minutes using Add Minutes, Custom Date and Time can be specified or a Wild Card can be used to modify the minute value, keeping the hour value unaltered. Since the smart meter has a lot of time related logical operations, with an added constraint of unforeseen power loss, this facilitates testing at time boundaries, highly accelerating the passage of time.

• Compute: (Software Operation) This operation is used for making all sorts of calculations within the software. This operation takes two registers as input namely, Operand 1 and Operand 2. The selected Operation is performed on the two operands in the form: Operand_1 (Operator)

47 Operand_2 = Output. The output generated is stored in a register specified by the user

• Result: Used for generating “Pass” or “Fail” testcase output. This operation compares two provided operands logically and generated a Pass or Fail metric based upon the output of the logical operation between the two operands.

• Note: Used for adding a statement or information that will be a part of the result log generated for the current testcase. Used for creating comments in the generated data.

Dimension 3: Supervised vs Unsupervised parts of running the tests Running the tests is the most time intensive part of the entire testing workflow, for obvious reasons. As a part of our design philosophy, we would like to ensure that the tests can be run largely unsupervised. At the same time, the challenging schedules dictate that the test engineer can be confident that when they set the tests to execute for the night or the weekend, the tests do execute to completion and are not stuck waiting for user input or for some piece of hardware to respond. The test execution part has hence been split into a Power-On Self-Test (POST) part and the actual execution part. In the POST part, the AutoTest solution does a dummy execution of the test scripts, looking for all possible user inputs and surfaces them right at the start. It also exercises all attached hardware, including the device under test and the automatic test bench, ensuring that they are properly connected and configured. It also tags the execution with the configuration details of the DUT and the identification of the test engineer to digitally sign the execution log. This ensures that once the test engineer launches the final execution, the tests do run to execution. During execution, the solution leans in the direction of being liberal in what to except, but strict in what is emitted. If a test case within a test script cannot be executed for any of the myriad of reasons, the solution just skips over the test case and marks it as “unexecuted”, to be manually validated by the test engineer. Also, being server based, the execution can be monitored remotely.

Dimension 4: Modularity Since the entire design of the AutoTest solution hinges on modular objects, or action primitives, the possibility of adding new modules is always open. Adding a new module is a non-disruptive exercise which is not likely to leak implementation abstraction over to the existing test scripts. Changing the implementation of a certain primitive also does not need to entail any changes to the existing library of test scripts as long as the method signatures, the implementation interfaces of objects, are contractually constant. In case of adding new attributes to a specific action primitive, the solution defaults to a NULLABLE attribute with the NULL being handled as the old behavior. Due to these design choices, we have been adding new modules all the time, without disruption.

Implementation Based on all the above design decision, the hardware and software implementation of the AutoTest solution is described below Hardware

48 1. Electronic Power Source. As a part of initial implementation, we selected Applied Precision’s PTE 2300A as our electronic power source, to interface with the AutoTest software. This is connected to the PC using either a serial or a USB interface. 2. Camera. A standard consumer grade web camera. 3. Sensors. Optical sensor to detect the pulse inputs. 4. The device under test- the smart meter. This is connected to the PC using its standard communication medium, wired or wireless. 5. The PC which is networked to a test server. The test server stores the scripts and results which can be accessed from anywhere on the intranet. The block diagram below is a representation of how these components come together.

Figure 4: System Block Diagram

Software The software enables all the three workflow elements of the standard test paradigm. Creating the tests, running the tests and analyzing the test results. Workflow 1: Creating the tests using a drag and drop interface In order to create a new test case, the software offers a 3-pane user interface.

Figure 5: Creating the tests

49 The library of components is on the left pane, from which you can drag and drop the components in sequence to the middle pane. The attributes of the selected action are on the right most pane, for configuration. The attributes have been illustrated in the design section for various action primitives. The test case, thus created can be saved as a part of the relevant test plan.

Workflow 2: Executing the tests Executing the tests after they have been created is as simple as selecting test cases from the library of test cases and running those. The software runs the supervised part of the workflow, also known as the Power-On Self-Test (POST). Once all attached hardware is validated and all requisite information is supplied, the test engineer can proceed to run the autonomous part of the test execution which runs unattended. While executing the software shows the current status and the real time execution log.

Workflow 3: Result analysis This mode enables the user to see reports of the tests that have been performed over time. The test execution instance can be selected based on the date of execution, the project name or the test engineer’s name or combinations of these search parameters. The test cases that had a results component in them and had passed are in green. The failed test cases are in red. The ones which do not generate a binary pass or fail are in grey. Typically, the test engineer is interested in analyzing the results for the red and the grey tests. Digging into details, the software allows Figure 6: Viewing and analyzing the results you to see each step of the test case, its respective actions and intermediate results, and even the raw time stamped byte stream of communication when each of these actions were performed.

Results The system was first deployed in the second half of 2018 and the results have been very promising. We have observed more than a 50% reduction in the test schedules, which is expected to go up further as the teams get more familiar with the scripting paradigms and the library of test cases grows to cover a larger fraction of the coverage requirements. The motivation levels of the staff have also improved, based on anecdotal evidence, as the AutoTest solution automates a large fraction of the drudgery of repetitive actions.

50 Future directions Continuous Integration Continuous integration is one of the software engineering best practices, which entails that the new implementation is built into a testable solution as frequently as possible, and the solution is tested automatically to shorten the time between implementation and getting feedback. This tightening of the feedback loop is at the heart of the Agile Methodologies. It has always been hard to do that with products that have a significant hardware component in them, smart meters being an example of the same.

Using the strategy illustrated, on a schedule –say, at 1800 hrs. – the build server gets the new bits from the configuration management server, where the code is checked in. The build servers compile and build the binary and flash the new bits to the device under test and trigger the test server. The test server already has a test sequence ready to execute, which has been created by the test team during the day, while the development team was implementing the new stuff. The test sequence runs overnight, and the results are emailed to the developers. The developers can then fix the issues thus detected the very next working day and commit the new bits to the configuration management server. All the individual interactions in the above have technically been implemented, and the process glue to tie this all together is being worked on. Probabilistic Models to simulate real world Electrical Conditions The entire execution is presently deterministic. We have determined that the various electrical parameters have very specific frequency distributions. For example, the voltage and frequency values are a gaussian distribution, while the current and power factor are uniformly distributed over the range. The probabilistic implementation of the power source action will enable automating serendipity, simulating the actual field conditions, using fuzz testing paradigm. Reading the display using a camera A low-cost web camera, along with localization networks [1] can help read the LCD, and compare it to values read from the meter’s memory. Work on this is going on and is showing great results.

REFERENCES [1] J. . Redmon and A. . Farhadi, "YOLO9000: Better, Faster, Stronger," arXiv: Computer Vision and Pattern Recognition, vol. , no. , pp. 6517-6525, 2017. [2] S. S. S. R. Depuru, L. . Wang, V. . Devabhaktuni and N. . Gudi, "Smart meters for power grid — Challenges, issues, advantages and status," , 2011. [Online]. Available: https://ieeexplore.ieee.org/document/5772451. [Accessed 5 3 2019]. [3] A. Precision, "Portable Test Equipment," Applied Precision, [Online]. Available: https://www.appliedp.com/product/pte/.

Ultrasonic Testing Beyond Flaw Detection

Ultrasonic TestingHormoz Beyond Ghaziary Flaw Detection Advanced NDE Associates, 1 2 3474Hormoz Voyager Ghaziary Circle, San and Diego, Alfred CA Haszler 92130, USA Tel: (858) 350-8630; e-mail [email protected] 1Advanced NDE Associates 3474 Voyager Circle,Alfred San Haszler Diego, CA 92130, USA (858)ALTECH 350-8630; Consulting email [email protected] GMBH Koblenz, Germany 2ALTECHa.Haszler@t Consulting-online.de GMBH Koblenz, Germany email [email protected]

ABSTRACT The use of ultrasonic testing for flaw detection has a history of about seven decades. Beyond this application ultrasonic testing offers many other opportunities in the field of material characterization in which the main interest is the passage of ultrasound through a material with respect to what it reveals of material elastic properties. Two major and measurable attributes of a sound beam passing through a material are Velocity and Attenuation both of which are directly related to material structure. The following paper describes the use of ultrasonic testing to detect and measure Residual Stress in rolled Aluminum plates that are used for machining monolithic components of aircrafts. An example is given in fig. 1. The lower bulkhead of the C17 airplane is machined in one piece out of rolled aluminum plates of up to 8” thick. Excessive residual stress in the plate can cause deflection and deformation in the material during machining operation which renders the operation useless. Aerospace industry currently specifies a destructive test method in which a number of samples are cut from various parts of a plate and tested by machining it in 0.5” steps and measuring the resulting deflection which will then be sued to calculate the amount of residual stress. The following paper describes the work of the authors in developing an ultrasonic test method which can project the degree of deflection and amount of residual stress in a nondestructive fashion.

Fig. 1: Bulkhead C17 A/C machined in monolithic form

Copyright 2019. This paper is intended for the sole use of registered attendees. No part of this publication or its contents may be copied, uploaded to the internet, or stored in any shared retrieval system. 52 RESIDUAL STRESS IN ROLLED ALUMINUM PLATES Aluminum plates used for critical components are made of heat treated Al-Zn (7050 & 7150) alloys which provides the highest strength that can be gained from aluminum alloys. The process of production is shown in fig. 2. The processes rolling, heat treatment and quenching may cause stress and warping in the plate. The process of stretching removes warping and relieves the major part of stresses. What remains in the plate is called “residual stress”.

Fig. 2 Production steps in producing high strength heat treatable aluminum alloys

Standard Residual Stress Test (Step Machining) This method has been established by the aerospace industry for the measurement of residual stress and deflection after machining. This method is shown in fig.3. The standard sample is 2” wide and 16” long. Its thickness is the ordered thickness of the plate.

Fig. 3 Step Machining Process

The sample is machined in ½” steps till half thickness is reached. At each step the sample is placed over a three point support and the amount of deflection is measured. In order to quantify the residual stress that caused the deflection the principles of beam deflection is borrowed from mechanical engineering. The model that is used is a simple supported beam loaded at its center as shown in fig. 4. As such the deflection is obtained from equation 1.

^ = (Eq.1)

𝐹𝐹𝐹𝐹 𝐿𝐿𝐿𝐿 3 Where𝛥𝛥𝛥𝛥𝛥𝛥𝛥𝛥 48 𝐸𝐸𝐸𝐸∗𝐼𝐼𝐼𝐼 F = force applied to the center of the beam, L= length of the beam, E= modulus of elasticity, I is area moment of inertia.

53 Fig. 4 Simply supported beam

ULTRASONIC MEASUREMENT OF RESIDUAL STRESS. An ultrasonic test method must simulate an external force as the cause of deflection while in reality the deflection of interest is caused by the release of internal stress during machining operation. Residual stress in rolled aluminum plates are basically misalignments in the FCC structure caused by previous external forces and frozen in the material causing stress in various regions. Machining will partially release stress which may cause warping and deflection. Stressed areas in a rolled plate are regional. Therefore a standard sample cut from a plate may contain areas with different stress level and our experiments show that stress difference is the main cause of deflection during machining a standard sample. Stress is related to material elastic properties thus areas under different stress level have different elastic moduli. Elastic module of a sample san be measured by velocity of sound traversing through its thickness. The aim of this experiment has been to establish a reliable method of scanning a plate for its elastic constant (E) and project deflection by developing a meaningful relationship between the difference in elastic Modili (ΔE) and the resulting deflection.

Calculating elastic module In order to measure the elastic module the following equation is used:

VL = (1 µ)/ (1 + µ)(1 2µ) (Eq.2)

�𝐸𝐸𝐸𝐸2 − 𝜌𝜌𝜌𝜌 − E = VL (1+µ)(1 - 2µ)/(1 -µ) (Eq.3)

ρ Where VL = longitudinal Velocity E = Elastic modulus µ = Poisson’s Ratio ρ = Density

Measuring Poisson’s ratio In equations 2 and 3, the Poisson’s ratio ρ is used as a constant but in reality it’s related to shear and longitudinal velocities of the material and in order to maintain accuracy, it must be treated as a function of longitudinal velocity VL and shear velocity VS as shown in Eq. 4.

VS = /2 (1 + ) (Eq.4)

Poisson’s�𝐸𝐸𝐸𝐸 ratio𝜌𝜌𝜌𝜌 can𝜌𝜌𝜌𝜌 then be calculated using equation 5:

2 2 µ = (VR – 2)/2(VR – 1) (Eq.5)

Where VR is velocity ratio VL/Vs

54 Test system and instrumentation Immersion technique are used in this work in which Aluminum plates are placed in an immersion tank The instrumentation used for this program included a 256 element phases array probe working in conjunction with a 256 channel pulser-receiver. The phased array probe was programed to produce ultrasound beams as shown in fig. 5.

Fig.5

A three transducer (virtual) was used. This was done by programming the firing sequence of phased array elements. The center beam is longitudinal and enters the plate at right angle. One of the two transducers on each side of the central beam transmits ultrasound beam that enters the plate at 5 degrees and generates a shear wave at 11 degrees which after reflecting from the bottom of the plate will reach the receiver transducer at the same angle arrangements. The firing sequence needed of this arrangement repeats itself from one end of the probe to the other during the scanning. The resulting scan produces two scan files containing time of flight data for longitudinal and shear wave. A typical stationary signal is shown in fig.6.

Fig.6 Data Processing A completed scan gives us two sets of time of flight data for longitudinal and shear waves. The following steps are applied to these two data sets. • Plate thickness applied to both scans to convert time of flight to velocity. • For each data point L velocity is divided by S velocity to give velocity ratio. • Velocity ratio is inserted in eq. 5 to give Poisson’s Ratio. • Poisson’s ratio is inserted in eq.3 to yield modulus of elasticity E. The above operation gives us a scan of E the modulus of elasticity. Scan after each processing step is shown in fig.7.

55 Initial scan Interim processing Scan of E Fig.8

TEST RESULTS It is important to note that for plate applications residual stress as a single data point is meaningless. Rather it refers to an area of the plate that may have the potential of warping. As such in order to get a residual strass value and a deflection potential we must select a 2” by 16” area on the plate as an imaginary sample and within that we must locate maximum and minimum values for E. The difference between high and low values of E show a direct and linear relationship with deflection after machining. Furthermore the value of residual strass calculated according to mechanical rules for deflection is very similar to respective values obtained from the UT scan. These are shown in table 1. 𝛥𝛥𝛥𝛥𝐸𝐸𝐸𝐸 Table 1. Relationship residual stress calculated from mechanical rules and residual stress as represented by E2 – E1

Deflection (mm) E2 – E1 Residual Stress (mechanical) 0.274 2.308315 2.359533 0.297 2.825014 2.561123 0.433 3.312097 3.749319 0.57 5.141422 4.937528 0.842 7.081907 7.337963 0.936 8.3644 8.149226 0.975 8.42596 8.515242

Fig.9

56 A comparison between mechanically measured i.e. by step machining is shown in fig. 9. It is important to note that the perfect linearity of the mechanical measurement is natural as it is calculated based on existing deflection. Material density, Elastic modulus and Poisson’s ratio are assumed constant whereas in ultrasonic measurement the variation of all these factors are taken into account. As it can be seen in the graph of fig.10, the relationship between deflection and difference in elastic constants in the chosen 2” by 18” area are linear and the equation given in the graph can be used to project deflection.

Fig.10

Examples for the actual tests performed in production environment Since the ultrasonic method described above is already in use we had the opportunity to validate our method in a real life situation.

Fig.11

Fig.11 shows the ultrasonic scan of an Aluminum plate and the distribution of elastic modulus E. This plate is 6 meters long, 2 meters wide and 50 mm thick. Following the process of heat treatment it was stretched for flatness and stress release as a part of production. During the process of stretching, a syncopation in stretching force occurred due to stretcher malfunction which interrupted the flow of stretching forces. The effects can be seen in the

57 scan. A small interruption in stretching forces has created significant disorder in the distribution of material constants. The black rectangle on the scan is a 2” by 16” area which plays the role of a real sample. The left side of the sample shows 535 MPa (77 KSI) as the maximum dominant E value. The right side shows 493 MPa (71.6 KSI) as dominant E value. Projected deflection can then be calculated from eq.6 as shown in fig.1.

Y = a + b*X Eq.6 Where X = deflection (mm) Y = ΔE KSI a = 0.0491 b = 8.7365 For ΔE of 5.4 KSI a deflection of 0.6237 mm is projected.

Further examples of the application Below, in fig.12, is E scan of a 7xxx Aluminum plate of 50 mm thick and 20 meters long which shows no sign of residual stress. Note that the topographic lines on the scans represent equal values of E and in fact represent the flow of material during stretching.

Fig.12

Conclusions • The method works as expected and the value of projected deflection is very close to actual deflection in the step machining method. • Best results are obtained with the use of a phased array system as it lend itself to producing scans of higher resolution. Furthermore it allows us to create virtual angle beam transducers. The simplicity of doing this and the accuracy that a phased array system provides if far more practical than using actual transducers. • It’s important to note that the described method and what it projects i.e. deflection and residual stress values, are based on the standard 2” by 16” sample. For a longer or shorter samples (both real and virtual) the system will need to be re-calibrated with new deflection values. • Equation 1 pertaining a simple load applied to the center of a beam is not the absolute standard and in fact some customers may use a uniform load distribution or a modification thereof.

REFERENCES

1. Vary, Alex,”Material property Characterization”, Nondestructive Testing Handbook, Third Edition, Volume 7, Ultrasonic testing, Columbus, OH, American Society for Nondestructive Testing 2. Vary, A, “Ultrasonic Measurement of Material Properties” Research techniques in nondestructive testing, Volume 4, London, Academic Press, 1980 3. Truell, R, C. Elbaum, and B. Chick, “Ultrasonic Methods in Solid State Physics” , Academic Press, 1969 4. Lynnworth, L, “Ultrasonic Measurements for Process Control” Academic Press, London, 1989

Guided Wave Simulations in Pipes with Non-Axisymmetric Guided Wave Simulations in Pipes with Non-Axisymmetric and Inclined and Inclined Angle Defects using Finite Element and Hybrid Modeling Angle Defects using Finite Element and Hybrid Modeling Masoud Masoumi1 and Ryan Kent Giles2 Masoud Masoumi1, Ryan Kent Giles2

1Department of Mechanical Engineering, Stony Brook University, Stony Brook, USA NY 11794-2300 1Department of Mechanical Engineering, Stony Brook University, Stony Brook, USA NY 11794-2300 email [email protected] email [email protected]

2Department of Civil Engineering, Stony Brook University, Stony Brook, USA NY 11794-4424 2Department of Civil Engineering, Stony Brook University, Stony Brook, USA NY 11794-4424 (631) 632-8601; email [email protected] (631) 632-8601; email [email protected]

ABSTRACT This work addresses mode conversion and propagation in long pipes with non-axisymmetric circumferential and inclined angle cracks using finite element (FE) modeling and a hybrid simulation approach. First, the FE method was used to simulate guided wave scattering in a 12 inch diameter pipe and to study the effects of a crack’s geometrical parameters, such as width, length and angle, on the propagating modes. A comb array transducer excited the first longitudinal mode at 120 kHz in the FE model and the simulations ran in parallel on a cluster computer. Dispersion curve calculations and the circumferential order identification approach were both implemented to identify newborn modes in the pipe as a result of defects. Each defect case is discussed in further detail. Additionally, a hybrid model, incorporating the FE method and an analytical solution, was introduced to find propagating modes at some distance along the pipe. This hybrid model uses a modal analysis-based analytical solution to find propagating signals at the intact sections of the pipe, while implementing a FE simulation for the cracked segment of the pipe. This hybrid approach dramatically reduces the time and computational power required to simulate high frequency wave scattering in long, defect-ridden pipes.

Keywords: hybrid modeling, finite element method, guided waves in pipes, circumferential order identification

INTRODUCTION The idea for the hybrid modeling approach stems from the time and computational power required to simulate high frequency guided waves in large and long structures such as electrical transmission pipes. In the case of the finite element analysis (FEA) of a high frequency propagating wave in a pipe, the required conditions for numerical solution, such as Courant-Friedrichs-Lewy (CFL), should be met [1]. Therefore, the finite element model must have a large number of elements for long pipe simulations. These large models are time consuming to solve even when using powerful computers and parallel computation approaches. This bottleneck gives an incentive to come up with hybrid models that can provide relatively fast and accurate results for guided wave modeling in long pipes with defects and geometrical irregularities.

Research published in 2011 addressed the derivation of a hybrid model to simulate wave propagation in an infinite cylinder with various defect shapes [2]. The model had two parts, where the modal expansion technique was used to find the wave patterns on both sides of the crack and a conventional finite element formulation was employed to find the solution in the cracked region. This model was basically a combination of the Semi-Analytical Finite Element (SAFE) method and the finite element (FE) modeling technique. A new hybrid model was proposed by Kirby and his colleagues to investigate coated pipes with axisymmetric uniform and nonuniform defects [3, 4]. This model also implemented SAFE along with a conventional FE approach to decrease the number of degrees of freedom. A mode matching procedure was utilized between the region with no coating, the region with a coating, and the region with the defect. The advantage of this method was that it only required a finite element discretization of the non-uniform defect and the rest of the pipe was modeled using the modal expansion technique. A pipe with an arbitrary crack was addressed using the formulation for a hybrid model consisting of SAFE and FE approaches in [5, 6]. In [5], the method was used for simulating a pipe with a partially circumferential defect with a depth of 50% of the pipe wall

thickness and a length of 2.5 mm. The pipe had an outer and an inner radius of 43.5 mm and 38 mm respectively. In [6], the pipe under study had a 39 mm inner radius, a 44.65 mm outer radius, and a 5 m length. Coating layers with thicknesses of 1.5 mm, 3 mm, and 5 mm were investigated.

This work proposes a new approach for hybrid modeling, an alternative to the procedure adopted by [3, 4, 6]. The proposed method in those works could be used to simulate guided waves in pipes with axisymmetric and non- axisymmetric cracks through a combination of SAFE modeling and an FE modeling procedure. The section of the pipe with no crack was simulated using a two-dimensional discretization approach and the cracked section was modeled via a full three-dimensional discretization. Those approaches still used discretization along the radial direction for intact sections of the pipe. The studies in [5, 6] only investigated the effects of a non-axisymmetric circumferential defect on wave scattering. The present research studies different types of non-axisymmetric as well as inclined angle defects on wave propagation. To achieve this goal, first an FE model is developed and used to study the effects of different parameters on the guided wave’s behavior. Then, this model is combined with the analytical solution to form the hybrid modeling procedure.

FINITE ELEMENT MODELING The FE model was developed using the Elmer Multiphysics package [7]. Boundary conditions were specifically written to excite the pipe at 120 kHz in its first longitudinal mode based on a comb array transducer design. The pipe is made of steel with 7925 kg/m3 density, 0.285 Poisson’s ratio, and 5100 m/s speed of sound. It also has 30.48 cm (12 in.) diameter and 0.635 cm (0.25 in.) wall thickness with a length varying between 112 cm to 142 cm depending on the case and damage size. Figure 1 shows the phase velocity and group velocity dispersion curves for the pipe covering 0-300 kHz frequency range.

(a) (b) Figure 1: (a) Phase velocity dispersion curve and (b) group velocity dispersion curve

The FE model’s time step and element size are related to each other by the Courant-Friedrichs-Lewy (CFL) condition [1], which ensures solution stability. To capture the amplitude of a propagating wave at equally-sized discrete time steps in a discrete spatial grid, the size of these time steps must be less than the time the wave needs to travel to an adjacent point in the spatial grid if an explicit method is used to solve the hyperbolic partial differential equation. A simulated comb array transducer excited the pipe in its first longitudinal mode. This type of transducer has an operating line with a slope that can be configured for the desired excitation mode at the desired frequency. The comb array generates a constant wavelength and excites modes based on the gaps between its elements and time delay values [8]. Figure 2: FE modeling procedure shows the general procedure for FE modeling and the software packages used for each step of the simulation. The model used second order hexahedron elements with 8 nodes. The mesh was divided into 224 sub-domains using ElmerGrid. Simulations were run on the SeaWulf cluster at Stony Brook University and the results were post-processed using the Python shell environment in ParaView. The model had approximately 9,500,000 elements with a 1.1×10-4 ms simulation time step to excite the pipe in its first longitudinal mode at 120 kHz. The running time varied between 40 - 60 hours depending on the case study using 8 cluster nodes with 28 CPUs per node.

Figure 2: FE modeling procedure

Although the excitation was a sine modulated signal designed to excite the pipe’s first longitudinal mode, F(1,2) mode was slightly excited as well due to the closeness of the two modes in the dispersion curve and some similarities between the mode shapes. For the intact pipe, Figure 3 shows the guided wave propagating in the pipe wall and the calculated dispersion curves and circumferential order identification results. The model and theoretical dispersion curves are well correlated. To implement circumferential order identification, time signals were measured at 18 locations around the pipe every 20 degrees and a shifting procedure was performed in the frequency domain [8].

(a) (b)

(c) (d) Figure 3: (a) Wave propagation pattern in the pipe wall at 120 kHz after 0.0154 ms, (b) dispersion curve, and circumferential mode orders at 40 cm away from the transducer for (a) 0th order mode and (b) 1st order mode

(a) (b)

Figure 4: (a) Two types of defects: (I) non-axisymmetric circumferential and (II) inclined angle and the coordinate system for the pipe, (b) schematic representation for the pipe with (1) excitation location, (2) location of defect, and (3) measurement points

Effects of Geometrical Parameters The size and shape of defects affect the newborn modes in the pipe and consequently the wave scattering pattern. Here, two types of defects are considered to study these newborn modes, including non-axisymmetric circumferential and inclined angle defects. The shape and size of these defects are defined using the parameters shown in Figure 4(a) which are listed in Table 1.

Table 1: Geometric variables for defect cases based on the parameters defined in Figure 4(a) Case δ (× wall thickness) α (°) ζ (cm) ι (cm) γ (°) 1 Case-I 2 45 2 2 90 1 Case-II 2 90 2 2 90 1 Case-III 2 120 2 2 90 1 Case-IIIN 2 120 1 1 90 1 Case-IIIW 2 120 3 3 90 1 Case-IV 2 180 2 2 90 1 Case-V 2 225 2 2 90 1 Case-VI 2 300 2 2 90 1 Case-VII 2 360 2 2 90 1 IC-I 2 45 2 4 71.52 1 IC-II 2 45 2 5.8 64.15 1 IC-III 2 45 2 20.7 30.04 1 IC-IV 2 45 2 20.7 0

For Case-I to Case-VII, dispersion curves were calculated along the pipe wall surface at four locations, including 0, 90, 180, and 270 degrees around the pipe’s cross section (labeled points 1 through 4 in Figure 4(a)) 22 cm away from the defect location. The circumferential order identification results were used to calculate the power corresponding to each mode order. The goal was to study the influence of the defect’s circumferential extent on the scattering wave. Results are shown in Figure 5 and Figure 6.

Figure 5: Comparison of the normalized-amplitude, frequency-wavenumber dispersion curves for non- axisymmetric circumferential defects (signals are measured at Point 2)

Figure 6 Comparison between signal powers calculated for mode orders 0 and 1 for circumferential defect cases with various circumferential extents using the time signal measured at 40 cm (i.e. 22 cm from defect location)

For Case-I, the main propagating waves are L(0,1) and F(1,1). However, traces of F(1,2) are observed in the dispersion curve. Increasing the circumferential extent of the crack to 90◦ for Case-II, mode L(0,1) was identified along with F(1,1) and F(1,2) modes. At 120 kHz, the group velocity difference between the F(1,1) and F(1,2) modes is not considerable. Thus, the separation between the modes cannot be observed in the circumferential order identification results for 1st order modes. However, the higher contribution of F(1,2) is seen in the dispersion curves represented in Figure 5 in values normalized with respect to the intact case’s maximum value. Case-III had a circumferential angle of 120◦, and three modes were identified: L(0,1), F(1,1), and F(1,2). The 1st order modes are still observed in dispersion curves measured at Point 2 (see Figure 4(a) for numbering). Only a trace of F(1,3) is seen in the dispersion curves calculated at Point 2. Analysis of the results for Case-IV revealed three modes similar to those in the results for prior cases as well as traces of F(1,3). Case-V, with a circumferential angle of 225◦, experienced a higher level of F(1,3) in the pipe as seen in the dispersion curves. For this case, the F(1,2) mode is identified in the dispersion curves calculated at Point 2. For case-VI, which has a 300◦ extent along the circumferential direction, there is a dominant influence of L mode on the propagating wave. The primary reason for this shift is the geometry of the crack, which is close to an axisymmetric circumferential defect. Case-VII, with a circumferential crack that is fully extended around the pipe, confirms this theory. For Case-VII, only L(0,1) was observed both in the dispersion curves and in the circumferential order identification graphs.

As shown in Figure 6, the signal power for the 0th order mode decreased continuously from the intact case to Case- VII. The passing energy was reduced due to the increase in the crack’s length and the waves’ interactions. Also, the energy contribution of the 1st order modes increased until Case-V after which it declined. Further, there is a noticeable change in the pattern of power reduction for the 0th order mode from Case-II to Case-III. This is primarily due to the difference in circumferential extent increments for these cracks. While the change from Case-II to Case-III is only 30◦, from 90◦ to 120◦, the change from Case-III to Case-IV is 60◦, from 120◦ to 180◦.

Studying Case-IIIN, Case-III, and Case-IIIW provides an insight into the effects of width on the first longitudinal mode propagation and the behavior of newborn modes. For these cases, the defect width was 1 cm, 2 cm, and 3 cm respectfully. Figure 7 presents the results for calculated dispersion curves for each case on Point 2. Figure 8 shows the power calculated for each mode order, the reflection and constructive/destructive effects of interactive waves, and the pattern observed in the measured signals.

(a) Figure 7: Comparison between the normalized amplitude frequency-wavenumber dispersion curves for circumferential defects with varying widths at Point (2)

(d)

Figure 8 (a) Schematic of the main phenomenon for longitudinal wave propagation when the width is changing, (b) comparison between signal powers calculated for mode orders 0 and 1 for intact and circumferential defect cases with various widths using the time signal measured at 50 cm, and (c) pattern of change in the 0th order mode due to changes in the crack’s width, (d) Schematic of main phenomenon in flexural wave propagation when the width is varied

Examining the wave behavior of axisymmetric and non-axisymmetric modes leads to a better understanding of the changes due to variations in the crack’s width. The axisymmetric modes’ contribution to the signal increases as the width of the crack increases, while the inverse phenomenon is observed for non-axisymmetric modes. As has been previously reported for symmetric modes in a plate [9], the symmetric modes’ contribution to the wave packet has a sinusoidal pattern due to the relation between the wavelength of the mode and the crack’s width, i.e., ζ/Λ. A simplified graphical representation in Figure 8(a) illustrates the constructive/destructive interference between the main propagating mode and its reflection from the other end of the crack. If the crack’s width is the same size as the mode’s wavelength, the reflected wave from the other end of the crack (shown in red) will be in phase with the propagating portion of the main wave (shown in a thin blue line). Therefore, the symmetric mode’s amplitude decreases due to the destructive interference between this secondary wave (the inversely reflected wave) and the main propagating wave (shown with a thick blue line). However, for a crack with 3/4 of the mode’s wavelength, destructive interference occurs between the reflected signal from the other end of the crack and the propagating portion of the main wave passing the defect. This destructive interference gives rise to a higher contribution of the longitudinal mode, moving forward along the pipe. The ratios of the crack’s width to wavelength for Case-IIIN, Case-III, and Case-IIIW are 0.556, 1.112, and 1.668, respectively. These values are displayed in a sinusoidal pattern in Figure 8(c), displaying values from ζ/Λ = 0 to ζ/Λ = 1 with repetition for larger numbers.

The flexural modes’ behavior follows a different pattern; their amplitudes decrease with an increase in the cracks’ width (see Figure 8(d)). The interference between the flexural mode and its reflections from the crack is a function of

the relationship between the crack’s width, its circumferential extent, and the mode’s wavelength. If the width of the crack is less than the mode’s wavelength minus the opposite side of the triangle formed by the circumferential extent and the helix angle, Γ, there will be no reflection of the flexural mode from the crack (as is shown in the top illustration in Figure 8(d)). Once the crack’s width becomes longer than the wavelength of the flexural mode minus the opposite side of the triangle formed by the circumferential extent and the helix angle, the propagating mode splits into two signals. The first signal has a helical pattern following the same motion as the main flexural wave, while the other signal is the reflected wave which creates an out-of-phase signal with respect to the flexural mode exiting the other end of the pipe. These interferences create a destructive effect and reduce the amplitude of the flexural wave.

Figure 9: Comparison between the normalized-amplitude, frequency-wavenumber dispersion curves for inclined angle defects at Point 2

Case IC-I to IC-IV along with Case-I were studied for defects set at an angle with respect to the axial length of the pipe. For the propagating wave in Case-I (the crack with a γ=90◦) the main contributions are from the axisymmetric mode L(0,1) and the non-axisymmetric mode F(1,1), while traces of F(1,2) are also observed. For IC-I and IC-II, similar levels of contribution from non-axisymmetric modes is observed, while L(0,1) is the only axisymmetric mode. Compared to Case-I, the contribution of the F(1,2) mode on the wave packet is slightly higher for IC-I and I-CII. Case IC-III has the smallest contribution from the axisymmetric mode, L(0,1), and the highest contribution from F(1,1) mode among all the cases. Also, Case IC-III is the only case in which F(1,3) appears in the 1st order propagating modes. Finally, IC-IV with a horizontal crack exhibits a behavior similar to the intact case, as the “disruption” the crack produces in the propagating wave is minimal. This case has only a small contribution from non-axisymmetric modes, i.e., the 1st order modes.

Overall, the signal power passing the crack decreases when the defect angle, γ, is widened or the length, l, is increased as shown in Figure 10(a). However, this pattern does not hold for case IC-III with a 30◦ angle to the horizontal axis, creating a shape like an axisymmetric mode shape in the pipe. It should be noted that the defects studied here have a higher inclined angle than the angle needed to generate these modes, i.e., the helix angle [10]. As a result, a sudden change in wave behavior cannot be contributed to these helix angles. Narrowing the defect angle and increasing the crack’s axial length results in an increase in the number of reflected waves from the second edge of the crack; it also amplifies the flexural mode as is schematically shown in Figure 10(b). This phenomenon is the case for IC-III where the crack’s axial length was increased to 20.7 cm, and the γ decreased to 30◦. IC-IV has a horizontal crack with a 2 cm circumferential extent and an axial length of 20.7 cm. This length is more than 11 times the wavelength of the first longitudinal mode. As a result, no reflection from the already reflected wave returning from the other end of the crack interferes with the propagating longitudinal mode, as the excitation signal has only 10 cycles.

(a) (b) Figure 10: (a) Comparison between signal powers for mode orders 0 and 1 for intact and inclined angle defect cases, measured at 50 cm, (b) reflections for an inclined angle defect

HYBRID MODELING A hybrid model simulates the wave propagation at a distance, while the defect or geometric irregularity is at a location far from the excitation point. Figure 11 schematically represents the general approach used to implement the hybrid modeling procedure. The propagating guided wave is found at a distance using the analytical solution as well as knowledge of the excitation signal. Then, this wave is used as an input signal for the comb array transducer in the FE model. Further, the wave scattering behavior due to any crack, defect, or irregularity is found in the time domain by running the FE simulation. Dispersion curve calculations and the circumferential order identification method are implemented for mode conversion detection, and eventually, the analytical solution is re-implemented to find the propagated mode at a distance on the pipe wall. The propagating wave packet in the pipe wall can be found by superimposing the modes at a desired distance.

Figure 11: Schematic of the hybrid modeling procedure.

When a guided wave propagates along the pipe, it spatially disperses as the amplitude decays following an exponential pattern. The decay constant for such decrements in amplitude is a function of wavenumber and the distance. Therefore, the exponential decay is more rapid for waves with higher wavenumber values. This pattern can be used to find the mode propagating at some distance away from the measurement point. The wave is measured around the pipe and the desired modes are identified from the FE simulation results through circumferential mode order identification. The propagated version of those modes can then be calculated by determining the location at which they are measured and the location along the pipe at which they need to be simulated. If the desired mode is measured and identified at a distance dm and the desired distance for the propagating version is dp measured from dm, then

− jk() d p ud(mp+= d ,) ud ( m ,) e (Eq.1)

The time representation of the propagated mode at location dp +dm can be found by taking an inverse Fourier transform of Eq.1 −1 ut() = F ud(,)mp + d  (Eq.2) @ddmp+  After confirming good correlation between a purely analytical model and an undamaged hybrid FE model, Case-I and Case-VI were used to implement the hybrid modeling procedure for wave scattering simulations in long pipes with a

defect. These two cases were selected due to the high contribution of F(1,1) mode in Case-I and of F(1,2) mode in Case-VI, which together, leads to a more accurate estimation of propagating waves using only one 1st order mode. The excitation signal for simulations was the same as in the defect analysis, except that the number of cycles was modified from 10 to 15 cycles in order to improve mode selectivity for the excitation procedure. First, the longitudinal mode was measured 10 m away from the excitation origin using the analytical solution, and then, it was used as an input for the FE model. The section of the pipe in the FE model contained the crack which caused wave scattering and mode conversion in the pipe wall. The propagating wave packet was measured at every 20◦ around the pipe wall and the circumferential mode order identification was used to detect longitudinal and flexural modes. Finally, the propagated version of each mode at the desired distance in the time domain was found using Eq.(2).

Following the above-mentioned procedure, propagating modes were calculated 10.4 m, 15.4 m, 20.4 m, 25.4 m, and 30.4 m away from the excitation point for the Case-I defect. Final wave packets at these locations were then found through superposition of the modes and are shown in Figure 12. Propagating signals for Case-VI at 10.4 m, 15.4 m, 20.4 m, 25.4 m, and 30.4 m away from the excitation point were also calculated using the hybrid model and are presented in Figure 13. For these simulations, the crack was located at 10.22 m from the beginning of the comb array transducer ring. Once the propagating wave reached the crack, the second order mode, F(1,2), was born and started scattering in the pipe along with the L(0,1) mode. The group velocity of the F(1,2) mode is higher than that of the L(0,1) mode at 120 kHz frequency (see Figure 1(b)). This higher group velocity causes the F(1,2) component of the wave to propagate faster than the L(0,1) component. Therefore, a propagating wave with two separate peaks is formed at measurement locations farther from the crack site as seen in Figure 13.

Figure 12: Wave scattering pattern at 10.4 m, 15.4 m, 20.4 m, 25.4 m, and 30.4 m along the pipe for Case-I using the full hybrid model.

Figure 13: Wave scattering pattern at 10.4 m, 15.4 m, 20.4 m, 25.4 m, and 30.4 m along the pipe for Case-VI using the full hybrid model.

CONCLUSIONS This paper discussed the modeling procedure for non-axisymmetric and inclined angle defects in a pipe using FEM and a hybrid modeling procedure, which includes analytical and FE modeling. Thirteen damage cases were studied to investigate the influence of two types of cracks and their corresponding geometrical parameters on the first longitudinal mode used as the excitation signal. A hybrid simulation method was explained and performed on two damage cases and the propagated wave packets were found at different locations along the pipe up to 30.4 m away from the excitation location. In the future, the hybrid modelling procedure can be expanded to account for multiple defects or irregularities in a pipe and even longer distances.

REFERENCES [1] Courant, R., 1956, "Partial Differential Equations [Russian translation], Mir (1972); perhaps translation of Part I of: Partial Differential Equations of Mathematical Physics, by R," Courant, K. Friedrichs, and N. Levy, New York University, New York. [2] Benmeddour, F., Treyssède, F., and Laguerre, L., 2011, "Numerical modeling of guided wave interaction with non-axisymmetric cracks in elastic cylinders," International journal of Solids and Structures, 48(5), pp. 764-774. [3] Kirby, R., Zlatev, Z., and Mudge, P., 2012, "On the scattering of torsional elastic waves from axisymmetric defects in coated pipes," Journal of Sound and Vibration, 331(17), pp. 3989-4004. [4] Kirby, R., Zlatev, Z., and Mudge, P., 2013, "On the scattering of longitudinal elastic waves from axisymmetric defects in coated pipes," Journal of Sound and Vibration, 332(20), pp. 5040-5058. [5] Duan, W., and Kirby, R., 2015, "A numerical model for the scattering of elastic waves from a non-axisymmetric defect in a pipe," Finite Elements in Analysis and Design, 100, pp. 28-40. [6] Duan, W., Kirby, R., and Mudge, P., 2016, "On the scattering of elastic waves from a non-axisymmetric defect in a coated pipe," Ultrasonics, 65, pp. 228-241. [7] Råback, P., Malinen, M., Ruokolainen, J., Pursula, A., and Zwinger, T., 2013, "Elmer models manual," CSC--IT Center for Science, Helsinki, Finland. [8] Rose, J. L., 2000, "Guided wave nuances for ultrasonic nondestructive evaluation," IEEE transactions on ultrasonics, ferroelectrics, and frequency control, 47(3), pp. 575-583. [9] Lowe, M. J. S., and Diligent, O., 2002, "Low-frequency reflection characteristics of the s 0 Lamb wave from a rectangular notch in a plate," The Journal of the Acoustical Society of America, 111(1), pp. 64-74. [10] Zhang, X., Tang, Z., and Lv, F., "Magnetostrictive helical array transducer for inspecting spiral welded pipes using flexural guided waves," Proc. AIP Conference Proceedings, AIP Publishing, p. 050002.

Ultrasonic Thickness Estimation using Multimodal Guided Lamb Waves Ultrasonic Thicknessgenerated Estimation by EMAT using Multimodal Guided Lamb Waves Generated by EMAT Joaquín García-Gómez1, Roberto Gil-Pita1, Antonio Romero-Camacho2, Jesús Antonio Jiménez-Garrido2, Joaquín García-GómezVíctor García1, Roberto-Benavides Gil-Pita2,1 ,César Antonio Clares Romero-Camacho-Crespo1, Miguel2, Jesús Aguilar Antonio-Ortega Jiménez-Garrido1 2, Víctor García-Benavides2, César Clares-Crespo 1, and Miguel Aguilar-Ortega1

11 SignalSignal TheoryTheory andand CommunicationsCommunications DepartmentDepartment University of Alcala Alcala de Henares, Madrid, Spain (34) 9191-8856751;-8856751; fax (34) 9191-8856699;-8856699; email [email protected]

22 InnerspecInnerspec TechnologiesTechnologies EuropeEurope S.LS.L Torres de la Alameda,Alameda, Madrid, Spain

ABSTRACT

The objective of this paper is to study how the selection of the coil and the frequency affects the received modes in guided Lamb waves, with the objective of analyzing the best configuration for determining the depth of a given defect in a metallic pipe with the minimum error. Studies of the size of the damages with all the extracted parameters are then used to propose estimators of the residual thickness, considering amplitude and phase information in one or several modes. Results demonstrate the suitability of the proposal, improving the estimation of the residual thickness when two simultaneous modes are used, as well as the range of possibilities that the coil and frequency selection offers.

Keywords: EMAT sensors, Lamb waves, pipeline inspection, defect sizing, coil selection, frequency selection

INTRODUCTION Defect sizing in pipeline inspection allows companies to determine when a pipe must be replaced, avoiding costly repairs in their assets. To tackle this issue, Lamb ultrasonic waves generated through Electro-Magnetic Acoustic Transducers (EMAT) allow thickness estimation without direct contact with the surface of the metallic material under investigation [1]. The use of this technology with a meander-line-coil allows generating waves in a directional way [2], which facilitates differentiating between circumferential and axial scans in Non-Destructive Testing (NDT) for pipeline inspection [3].

However, the shape of the defect changes the behavior of the ultrasonic signals when they pass through the pipeline, and it is not easy to predict the amplitude and phase of the wave in function of the residual thickness [4,5]. In recent studies the use of machine learning techniques applied to information extracted from signals sensed at different frequencies has been demonstrated to improve the accuracy of the estimation, but the use of multiple frequencies in general requires more complex sensing devices and more time. A possible way to address these disadvantages is the use of different modes sensed at a unique frequency, but in this case the selection of the coil and the inspection frequency becomes a critical aspect, since these different modes must be separable in the measurement, and this is not always the case.

This paper presents a theoretical study in which the selection of the coil and the frequency for multimodal thickness estimation are analyzed. The objective is to determine the relationship between the performance of the estimator and the configuration of the sensing system. The problem was approached from two perspectives. First, a signal processing based theoretical framework is proposed. Second, simulations obtained by a Finite Element software are considered. Results demonstrate the suitability of the proposals, improving the estimation of the residual thickness.

LAMB WAVE GENERATION USING EMAT SENSORS In this section the generation of Lamb waves through EMAT sensors will be described. These sensors are composed of a magnet and a coil wire. The current is induced in the surface of the ferromagnetic material when the alternating electrical current flow through the coil wire is placed in a uniform magnetic field near the material. When this field interacts with the field generated by the magnet, Lorentz force appears. Because of that, a disturbance affects to the material, creating an elastic wave. If the vibration is coplanar with the propagation plane, these waves are called Lamb waves. Conversely, the interaction of Lamb waves with a magnetic field induces current in the EMAT receiver coil circuit.

Lamb waves are characterized by their dispersion and sensitivity to thickness variations. Besides, they can be divided into modes: symmetric and asymmetric modes. Each mode is composed of two waves (longitudinal and transversal). They travel at different angles and with velocities and , where the latter refer to the sound velocity in longitudinal and transversal components, respectively. Considering a Lamb mode that moves in the 𝜃𝜃𝜃𝜃𝐿𝐿𝐿𝐿 𝜃𝜃𝜃𝜃𝑇𝑇𝑇𝑇 𝑐𝑐𝑐𝑐𝐿𝐿𝐿𝐿 𝑐𝑐𝑐𝑐𝑇𝑇𝑇𝑇 direction at velocity with a frequency , then the wavenumber is related to the longitudinal and transversal 𝑥𝑥𝑥𝑥 components of the wave: 𝑐𝑐𝑐𝑐𝑝𝑝𝑝𝑝 𝑓𝑓𝑓𝑓 𝑘𝑘𝑘𝑘

2 cos = cos = = (Eq. 1) 𝜋𝜋𝜋𝜋𝑓𝑓𝑓𝑓 𝑘𝑘𝑘𝑘𝐿𝐿𝐿𝐿 𝜃𝜃𝜃𝜃𝐿𝐿𝐿𝐿 𝑘𝑘𝑘𝑘𝑇𝑇𝑇𝑇 𝜃𝜃𝜃𝜃𝑇𝑇𝑇𝑇 𝑘𝑘𝑘𝑘 𝑐𝑐𝑐𝑐𝑝𝑝𝑝𝑝 where = 2 and = 2 , are the wavenumber of the longitudinal and transversal components, respectively. Furthermore, the displacement of each wave in the axis can be obtained using and , so that: 𝑘𝑘𝑘𝑘𝐿𝐿𝐿𝐿 𝜋𝜋𝜋𝜋𝑓𝑓𝑓𝑓⁄𝑐𝑐𝑐𝑐𝐿𝐿𝐿𝐿 𝑘𝑘𝑘𝑘𝑇𝑇𝑇𝑇 𝜋𝜋𝜋𝜋𝑓𝑓𝑓𝑓⁄𝑐𝑐𝑐𝑐𝑇𝑇𝑇𝑇

𝑧𝑧𝑧𝑧 𝛼𝛼𝛼𝛼𝐿𝐿𝐿𝐿 𝛼𝛼𝛼𝛼𝑇𝑇𝑇𝑇 1 1 = sin = 2 (Eq. 2)

𝛼𝛼𝛼𝛼𝐿𝐿𝐿𝐿 𝑘𝑘𝑘𝑘𝐿𝐿𝐿𝐿 𝜃𝜃𝜃𝜃𝐿𝐿𝐿𝐿 𝜋𝜋𝜋𝜋𝑓𝑓𝑓𝑓� 2 − 2 𝑐𝑐𝑐𝑐1𝐿𝐿𝐿𝐿 𝑐𝑐𝑐𝑐𝑝𝑝𝑝𝑝1 = sin = 2 (Eq. 3)

𝛼𝛼𝛼𝛼𝑇𝑇𝑇𝑇 𝑘𝑘𝑘𝑘𝑇𝑇𝑇𝑇 𝜃𝜃𝜃𝜃𝑇𝑇𝑇𝑇 𝜋𝜋𝜋𝜋𝑓𝑓𝑓𝑓� 2 − 2 𝑐𝑐𝑐𝑐𝑇𝑇𝑇𝑇 𝑐𝑐𝑐𝑐𝑝𝑝𝑝𝑝 Considering that the wave is reflected in the surfaces and applying the boundary conditions, we can get an equation related to the dispersion of the Lamb modes. Equation (4) refers to the symmetric modes and equation (5) refers to asymmetric ones.

4 sin cos + sin cos ( ) = 0 (Eq. 4) 2 2 2 2 2 𝐿𝐿𝐿𝐿 𝑇𝑇𝑇𝑇 𝑇𝑇𝑇𝑇 𝐿𝐿𝐿𝐿 2 2 2 𝐿𝐿𝐿𝐿 𝑇𝑇𝑇𝑇 𝛼𝛼𝛼𝛼 ℎ 𝛼𝛼𝛼𝛼 ℎ 𝛼𝛼𝛼𝛼 ℎ 𝛼𝛼𝛼𝛼 ℎ 𝑇𝑇𝑇𝑇 4𝑘𝑘𝑘𝑘 𝛼𝛼𝛼𝛼 𝛼𝛼𝛼𝛼 cos� � sin � � + cos� � sin � � (𝛼𝛼𝛼𝛼 − 𝑘𝑘𝑘𝑘 ) = 0 (Eq. 5) 2 2 2 2 2 𝛼𝛼𝛼𝛼𝐿𝐿𝐿𝐿ℎ 𝛼𝛼𝛼𝛼𝑇𝑇𝑇𝑇ℎ 𝛼𝛼𝛼𝛼𝑇𝑇𝑇𝑇ℎ 𝛼𝛼𝛼𝛼𝐿𝐿𝐿𝐿ℎ 2 2 2 𝑘𝑘𝑘𝑘 𝛼𝛼𝛼𝛼𝐿𝐿𝐿𝐿𝛼𝛼𝛼𝛼𝑇𝑇𝑇𝑇 � � � � � � � � 𝛼𝛼𝛼𝛼𝑇𝑇𝑇𝑇 − 𝑘𝑘𝑘𝑘 From the previous equations, it can be derived that there exists a relation between the excited frequency , the thickness of the pipe and the phase velocity . In particular, each mode travels at different depending on the 𝑓𝑓𝑓𝑓 other above-mentioned parameters. We get a similar relation with the group velocity , defined in equation (6). ℎ 𝑐𝑐𝑐𝑐𝑝𝑝𝑝𝑝 𝑐𝑐𝑐𝑐𝑝𝑝𝑝𝑝

𝑐𝑐𝑐𝑐𝑔𝑔𝑔𝑔

= −1 (Eq. 6) 2 𝜕𝜕𝜕𝜕𝑐𝑐𝑐𝑐𝑝𝑝𝑝𝑝 𝑐𝑐𝑐𝑐𝑔𝑔𝑔𝑔 𝑐𝑐𝑐𝑐𝑝𝑝𝑝𝑝 �𝑐𝑐𝑐𝑐𝑝𝑝𝑝𝑝 − 𝑓𝑓𝑓𝑓ℎ � 𝜕𝜕𝜕𝜕𝑓𝑓𝑓𝑓ℎ Solving the previous equations for different values of frequency and thickness we obtain the phase and group velocity for each propagating mode.

Now we will consider how signals are generated and received in the pipeline. The EMAT system consists of a meander-line-coil which generates two signals per loop in the system (one per meander). These waves are characterized by their wavelength which depends on the separation of the meanders. The following equations are valid for one mode and then we will iterate for all the modes which appear at a given frequency. Thus, we have to set the wave equation depending on the group and phase velocities. Considering as the excited frequency, the transmitted signal propagating in the axis will be generated according to equation (7). 𝑓𝑓𝑓𝑓

𝑥𝑥𝑥𝑥 ( , ) = sin 2 (Eq. 7) 𝑥𝑥𝑥𝑥 𝑠𝑠𝑠𝑠 𝑥𝑥𝑥𝑥 𝑡𝑡𝑡𝑡 � 𝜋𝜋𝜋𝜋𝑓𝑓𝑓𝑓 �𝑡𝑡𝑡𝑡 − 𝑝𝑝𝑝𝑝�� 𝑐𝑐𝑐𝑐 Please note here that the velocity will depend on the frequency and the thickness of the pipe. In a real case, the transmitted signal includes an envelope ( ) that generates the transmitted wave packet ( , ). This envelope 𝑐𝑐𝑐𝑐𝑝𝑝𝑝𝑝 limits the transmission time, and allows controlling the length of the transmitted pulse. Typically, the length of this 𝑤𝑤𝑤𝑤 𝑡𝑡𝑡𝑡 𝑝𝑝𝑝𝑝 𝑥𝑥𝑥𝑥 𝑡𝑡𝑡𝑡 envelope is described in function of , the number of cycles included in the wave packet. This envelope will travel at an average velocity of , and in general its shape will change with the distance due to dispersion effects. So, once 𝐶𝐶𝐶𝐶 the envelope is considered, the transmitted wave packet ( , ) will be expressed using equation (8). 𝑐𝑐𝑐𝑐𝑔𝑔𝑔𝑔

𝑝𝑝𝑝𝑝 𝑥𝑥𝑥𝑥 𝑡𝑡𝑡𝑡 ( , ) = sin 2 (Eq. 8) 𝑥𝑥𝑥𝑥 𝑥𝑥𝑥𝑥 𝑝𝑝𝑝𝑝 𝑥𝑥𝑥𝑥 𝑡𝑡𝑡𝑡 � 𝜋𝜋𝜋𝜋𝑓𝑓𝑓𝑓 �𝑡𝑡𝑡𝑡 − 𝑝𝑝𝑝𝑝�� 𝑤𝑤𝑤𝑤� �𝑡𝑡𝑡𝑡 − 𝑔𝑔𝑔𝑔� 𝑐𝑐𝑐𝑐 𝑐𝑐𝑐𝑐 From this point, instead of using the transmitted envelope ( ) we will use using ( ), which changes its shape in function of the distance due to dispersion effects. It is also necessary to consider that under EMAT technology the 𝑤𝑤𝑤𝑤 𝑡𝑡𝑡𝑡 𝑤𝑤𝑤𝑤� 𝑡𝑡𝑡𝑡 excitation signal is generated in a set of loops of a coil, separated by a distance , which will generate the propagation wave ( , ) using equation (9). 𝑁𝑁𝑁𝑁 𝐿𝐿𝐿𝐿

𝑦𝑦𝑦𝑦 𝑥𝑥𝑥𝑥 𝑡𝑡𝑡𝑡 + + 2 2 ( , ) = 2𝑁𝑁𝑁𝑁 ( 1) sin 2 (Eq. 9) 𝐿𝐿𝐿𝐿 𝐿𝐿𝐿𝐿 𝑚𝑚𝑚𝑚 𝑥𝑥𝑥𝑥 𝑚𝑚𝑚𝑚 𝑥𝑥𝑥𝑥 𝑚𝑚𝑚𝑚 𝑦𝑦𝑦𝑦 𝑥𝑥𝑥𝑥 𝑡𝑡𝑡𝑡 � − � 𝜋𝜋𝜋𝜋𝑓𝑓𝑓𝑓 �𝑡𝑡𝑡𝑡 − 𝑝𝑝𝑝𝑝 �� 𝑤𝑤𝑤𝑤� �𝑡𝑡𝑡𝑡 − 𝑔𝑔𝑔𝑔 � 𝑚𝑚𝑚𝑚=1 𝑐𝑐𝑐𝑐 𝑐𝑐𝑐𝑐 Each loop generates two signals (one per meander), and the sign of their contribution to the propagation wave ( , ) is included in the term ( 1) . Besides, the measure is sensed at a distance , in another set of loops separated by a distance . So, the received𝑚𝑚𝑚𝑚 signal ( ) will be expressed using equation (10). 𝑦𝑦𝑦𝑦 𝑥𝑥𝑥𝑥 𝑡𝑡𝑡𝑡 − 𝐷𝐷𝐷𝐷 𝑁𝑁𝑁𝑁

𝐿𝐿𝐿𝐿 𝑧𝑧𝑧𝑧 𝑡𝑡𝑡𝑡 + ( + ) + ( + ) 2 2 ( ) = 2𝑁𝑁𝑁𝑁 2𝑁𝑁𝑁𝑁 ( 1) sin 2 (Eq. 10) 𝐿𝐿𝐿𝐿 𝐿𝐿𝐿𝐿 𝑚𝑚𝑚𝑚+𝑛𝑛𝑛𝑛 𝑥𝑥𝑥𝑥 𝑚𝑚𝑚𝑚 𝑛𝑛𝑛𝑛 𝑥𝑥𝑥𝑥 𝑚𝑚𝑚𝑚 𝑛𝑛𝑛𝑛 𝑧𝑧𝑧𝑧 𝑡𝑡𝑡𝑡 � � − � 𝜋𝜋𝜋𝜋𝑓𝑓𝑓𝑓 �𝑡𝑡𝑡𝑡 − 𝑝𝑝𝑝𝑝 �� 𝑤𝑤𝑤𝑤� �𝑡𝑡𝑡𝑡 − 𝑔𝑔𝑔𝑔 � The signal received from𝑛𝑛𝑛𝑛 each=1 𝑚𝑚𝑚𝑚 =mode1 ( ) has different values of𝑐𝑐𝑐𝑐 and , as it was concluded𝑐𝑐𝑐𝑐 from equations (4), (5) and (6). Thus, each mode arrives at the receiver with different amplitude and envelope, depending on the 𝑧𝑧𝑧𝑧 𝑡𝑡𝑡𝑡 𝑐𝑐𝑐𝑐𝑝𝑝𝑝𝑝 𝑐𝑐𝑐𝑐𝑔𝑔𝑔𝑔 attenuation of each mode and the difference of phase when the signal is received in the coil. Therefore, the amount of energy of the received signal will vary in function of the frequency.

In order to find out more about the behavior of the modes, a frequency sweep has been carried out between 0 and 800 kHz with one coil and = 4 cycles per wave packet. Figure 1 shows the phase velocity (left) and group velocity (right), where black color means the energy is maximum at that frequency. Dispersion has been taken into 𝐶𝐶𝐶𝐶 account to carry out these experiments, since the signal ( , ) has been decomposed with the envelope window

𝑝𝑝𝑝𝑝 𝑥𝑥𝑥𝑥 𝑡𝑡𝑡𝑡

( ) through the Fourier Transform, and different velocity has been applied to each frequency component. These graphs correspond to a steel pipe with the following parameters: Young’s modulus = 210 10 , Poisson’s 𝑤𝑤𝑤𝑤� 𝑡𝑡𝑡𝑡 ratio = 0.3 and density = 7800 . 9 2 𝐸𝐸𝐸𝐸 ∙ 𝑁𝑁𝑁𝑁⁄𝑚𝑚𝑚𝑚 3 𝜈𝜈𝜈𝜈 𝜌𝜌𝜌𝜌 𝑘𝑘𝑘𝑘𝑘𝑘𝑘𝑘⁄𝑚𝑚𝑚𝑚

(a) (b)

(e) (f)

Figure 1: Phase velocity (a) and group velocity (b) in function of the product frequency by thickness, using coils with different value.

𝑳𝑳𝑳𝑳

The coil used in the experiments has the following parameters: distance between loops ranging from 0.3 to 0.5 inches, and = 3 loops. It can be observed that the same coil could be used to excite other frequencies, even if it 𝐿𝐿𝐿𝐿 has been designed to get the maximum energy in a given frequency. Furthermore, if we want to analyze the behavior 𝑁𝑁𝑁𝑁 of the modes in a deep way, we could change the length L of the coil. In Figure 1 it is observed that the points and areas of maximum energy vary significantly from one coil to another.

FREQUENCY AND COIL SELECTION FOR MULTIMODAL FEATURE EXTRACTION The modeling of the pipeline by means of the ultrasound waves is a non-trivial problem. The changing shape of the defects makes difficult to draw general conclusions about the relation between the defect and the received signals. The distortion caused by the defects over the different modes strongly varies with the shape of the mode [4,5]. For instance, the amplitude of the signal, the time of arrival (group velocity ) and the phase velocity of the wrap- around signal vary with the dimension and shape of the defect. 𝑐𝑐𝑐𝑐𝑔𝑔𝑔𝑔 𝑐𝑐𝑐𝑐𝑝𝑝𝑝𝑝

Thus, it is necessary to analyze how the different modes are going to be represented in the received signal, in order to look for the best configuration (frequency and size of the coil) that allows a better representation of the different modes over the same signal.

As it was stated, a meander-line-coil is used to generate the ultrasonic signals that are analyzed once they wrap the pipeline. It allows us to know the condition of the pipes depending on the different modes and wrap arounds received. In order to investigate how the behavior of the modes changes according to the length of the coil, a sweep of experiments has been carried out with coils from 0.30 inches to 0.55 inches, in steps of 0.01 inches. The most relevant results are shown in Figure 2, where we show where the energy of the different modes is located in a time- frequency representation. Both asymmetric (A0, A1, A2, A3) and symmetric (S0, S1, S2, S3) modes are plot in different colors. In each of the modes, curves indicate the area where the energy of the mode is higher or lower.

Results from this figure show that as we change the length of the coil, the parameters related to the modes (frequency of appearance, area of maximum energy, etc.) are not the same. The final effect is that the modes “move” in frequency and time. For instance, as the length of the coil in higher, some of the modes appear at lower frequencies. That is the case of A0, S0, A1 and S1 modes. Other modes disappear from the observed window, such as the A2 mode (pink), whose second wrap around went away from 0.40 inches to 0.45 inches. First wrap around disappear from 0.45 inches to 0.50 inches.

(b) (a)

(e) (f) Figure 2: Energy localization of the different modes in function of the frequency.

However, the usefulness of these graphs is that we can set a frequency depending on the modes or wrap arounds we are interested in, particularly when the objective is not to use just one mode. For instance, if we want to focus on modes A1 and S1 (dark blue and yellow), it can be seen that 0.30 and 0.35 inches are not the suitable lengths because both modes will appear mixed in the received signal. We should choose a higher value, such as 0.45 inches, where first wrap-around of A1 mode as well as first and second wrap around from S1 mode are well separated in time between 400 and 600 kHz approximately, so they will not be overlapped. Other option would be to choose a value of 0.50 inches, where these modes are almost completely separated, but including the second wrap-around of both modes or even the third one from the S1 mode. Again, it is clear that 0.55 inches is not a suitable value because these modes start to appear together again.

SMART SOUND PROCESSING FOR SIZING ESTIMATION If we want to solve the problem of pipeline sizing, it is necessary to apply a pattern recognition system, which is composed of two stages. In the first stage, useful information is extracted from the signals in the form of features. Later, in a second stage, a predictor tries to learn a model which will be useful for predicting the defects presented in the pipeline.

To extract useful information from the received signal is very important in the process, since it will be the “raw material” that the predictor will use. Analyzing a set of signals from real pipelines and the state of the art [6], we observe that the following features could be useful for the problem at hand:

• Average echo energy (dB), which represents the average energy of the echo received. • Peak wrap-around energy (dB), which represents the maximum energy of the pulse. We have considered ±30 µs around , the maximum of the signal in the case of absence of defect, to look for the maximum of each signal. 𝑡𝑡𝑡𝑡0 • Average wrap-around energy (dB), which represents the average energy of the pulse. We have considered ±30 µs around , the maximum of the signal in the case of absence of defect. • Wrap-around phase delay (µs), which represents how much time has passed between the pulse was sent and 𝑡𝑡𝑡𝑡0 it was received in the same point of the pipeline. It is determined measuring the time difference between and its closest maximum in ( ). Please note that a delay larger than 1 2 causes uncertainty, which conditions the usefulness of this measurement. 𝑡𝑡𝑡𝑡0 𝑧𝑧𝑧𝑧 𝑡𝑡𝑡𝑡 ⁄ 𝑓𝑓𝑓𝑓 • Wrap-around group delay (µs), denoted . In order to estimate this measurement, we consider the centroid of the average energy of the pulse around , with equation (11). 𝑡𝑡𝑡𝑡�𝑔𝑔𝑔𝑔

𝑡𝑡𝑡𝑡0 ( ) −5 = 𝑡𝑡𝑡𝑡0+3∙10 (Eq. 11) −5 2 𝑡𝑡𝑡𝑡=𝑡𝑡𝑡𝑡0−3∙10 ( ) ∑ −5 𝑡𝑡𝑡𝑡 𝑧𝑧𝑧𝑧 𝑡𝑡𝑡𝑡 𝑔𝑔𝑔𝑔 𝑡𝑡𝑡𝑡0+3∙10 𝑡𝑡𝑡𝑡� −5 2 ∑𝑡𝑡𝑡𝑡=𝑡𝑡𝑡𝑡0−3∙10 𝑧𝑧𝑧𝑧 𝑡𝑡𝑡𝑡

Figure 3: Model of the simulated defects.

Once we have obtained the features, we need to apply a nonlinear predictor to get the final profile of the pipeline and to know the performance of the developed model. Neural Networks have been applied, specifically the Multi Layer Perceptron (MLP) [7]. In this paper MLPs with a hidden layer of twenty neurons have been trained using the Levenberg-Marquardt algorithm [8].

From the results presented in Figure 2, we will select and = 450 and a coil with = 0.52 inches. So, in the case of using just the main mode we will consider 5 features, and in the case of considering two modes we will have 9 𝑓𝑓𝑓𝑓 𝐿𝐿𝐿𝐿 features (4 wrap-around features for each mode plus the echo energy).

To study the relationship between these parameters and the shape of the defects, we have used the Finite Element Method (FEM) included in the Partial Differential Equations Toolbox of Matlab. With these simulations we have generated a database with several different defects. The defects have been characterized with three parameters: length ( ), depth ( ) and slope ( ). Figure 3 describes the meaning of these parameters in a real pipeline. The thickness of the pipe used is = 7.8 mm, and the distance to the receiver is = 0.7 m. For simplicity, we have not 𝑙𝑙𝑙𝑙 𝑑𝑑𝑑𝑑 𝑠𝑠𝑠𝑠 modeled the width of the defect, that is to say, we have not considered the dimension of the pipeline. ℎ 𝐷𝐷𝐷𝐷

𝑦𝑦𝑦𝑦 Table 1: RMSE (mm) in the estimation of the residual thickness using an MLP with 20 neurons in the hidden layer for different number of used modes. S1 mode S1 and A1 modes

5 features 9 features RMSE (mm) 10.50 mm 9.72 mm

In a second approach, we have developed an experiment using a synthetic database for estimating the residual thickness of the pipeline. The database consists of 384 signals generated with defects of different shape. The length of the defect ( ) ranged from 10 to 100 mm, the depth ( ) from 0 to 9 mm, and the slope ( ) from 1 to 100 mm.

𝑙𝑙𝑙𝑙 𝑑𝑑𝑑𝑑 𝑠𝑠𝑠𝑠 To obtain the prediction results, -fold cross validation was applied in the generated database, being = 5. This method consists in dividing the database in groups so that the full process is repeated times, using one group of 𝑘𝑘𝑘𝑘 𝑘𝑘𝑘𝑘 signals as test subset and the remaining 1 groups as training subset. Results are then averaged to obtain the Root 𝑘𝑘𝑘𝑘 𝑘𝑘𝑘𝑘 Mean Square Error (RMSE) of the estimation of the depth. The advantage of this method is that the obtained results 𝑘𝑘𝑘𝑘 − are generalizable to defects different from those used in the database.

The objective is to know how well estimated is the received signal at different frequencies, so different experiments have been considered. First, we have considered the use of only one frequency, and we have also studied what happens when both frequencies are used at the same time. Concerning the features, the usefulness of each feature has been studied, and the inclusion of all three features has also been considered. Table 1 shows the RMSE in function of the features and the frequencies. In all the cases above it is clear that as we get better results when two modes are used.

CONCLUSIONS Pipeline inspection problem can be approached in many different ways. Lamb wave generation through EMAT sensors proves to be a very effective and useful one. However, the amount of information provided by the wrap- around signals needs to be processed by advanced techniques, such as smart sound processing algorithms. Thanks to them, it is feasible to get good estimation results of the pipeline defects, in both real and simulated signals.

In this paper we establish tools for determining the frequency and dimensions of the coil in order to be able to analyze two modes with an unique scan. We study how the behavior of the modes change when the length of the used coil is different, demonstrating its interest for multimodal approaches. Studies of the size of the damages with all the extracted parameters have been used to propose estimators of the residual thickness, considering amplitude and phase information. Results with two modes demonstrate the suitability of the proposal, improving the estimation of the residual thickness.

ACKNOWLEDGEMENTS This work has been funded by Innerspec Technologies Europe S.L through the “Chair of modeling and processing of ultrasonic signals” (CATEDRA2007-001), and by the Spanish Ministry of Economy and Competitiveness- FEDER under Project TEC2015- 67387-C4-4-R.

CONFLICTS OF INTEREST The authors declare that there is no conflict of interest

REFERENCES (1) Green, R.E., 2004. “Non-contact ultrasonic techniques”. Ultrasonics, 42, 9–16. (2) Zhai, G., Jiang, T., Kang, L., 2014. “Analysis of multiple wavelengths of Lamb waves generated by meander- line coil EMATs”. Ultrasonics, 54, 632–636. (3) Salzburger, H.J., Niese, F., Dobmann, G., 2012. “EMAT pipe inspection with guided waves”. Welding in the world, 56, 35–43. (4) Demma, A., 2003. The interaction of guided waves with discontinuities in structures. PhD thesis, University of London. (5) Cobb, A.C., Fisher, J.L., 2016. “Flaw depth sizing using guided waves”. AIP Conference Proceedings. AIP Publishing, Vol. 1706, p. 030013. (6) García-Gómez, J., Bautista-Durán, M., Gil-Pita, R., Romero-Camacho, A., Jimenez-Garrido, J.A., Garcia- Benavides,V., 2018, “Smart Sound Processing for Residual Thickness Estimation using Guided Lamb Waves generated by EMAT”. 27th ASNT Research Symposium, 99-105. (7) Weisz, L., 2016. “Pattern Recognition Statistical Structural And Neural Approaches”. Pattern Recognition, 1, 2. (8) Hagan, M.T.; Menhaj, M.B., 1994. “Training feedforward networks with the Marquardt algorithm”. IEEE transactions on Neural Networks, 5, 989–993.

77 MagnetostrictiveMagnetostrictive Cold ColdSpray Spray Sensor Sensor for Long for Long-Term-Term or Harsh or HarshEnvironment EnvironmentUltrasound Ultrasound

S.S. W. W. GlassGlass11,, J.J. P.P. Lareau1, K. S. Ross1, S. AliAli22,, F.F. HernandezHernandez22,, andand B.B. LopezLopez22

1 1Pacific NorthwestNorthwest NationalNational LaboratoryLaboratory P.O. Box 999 Richland,Richland, WAWA 99352 USA (509)(509 372-6190;) 372-6190 email; bill.glass [email protected]@pnnl.gov

2 2Innerspec Technologies, Inc. 2940 Perrowville Rd. Forest,Forest, VAVA 24551 USA ABSTRACT Ultrasound sensors are frequently used to generate acoustic waves capable of detecting cracks, pits, erosion, inclusions, etc. One problem with piezoelectric transducers, particularly in harsh environment applications, is the difficulty to achieve coupling between the transducer and the structure. This work explores the behavior of a magnetostrictive (ms) cold-spray patch on a stainless steel inspection target, and compares it to the performance of a standard adhesively bonded ferrous-cobalt (FeCo) ms strip. Cold spray is a process where powdered metal is accelerated to 2–3 × speed of sound and impacted on the surface forming a metallurgically bonded coating. If the powder is nickel or cobalt with high ms coefficients, this surface can serve as the base of an SH-0 or A-0 Lamb wave mode EMAT sensor suitable for crack or pitting corrosion damage monitoring that is not subject to temporal or environmental degradation. Guided-wave ultrasound edge reflection signals from adhesively applied FeCo strips were compared to nickel or nickel alloy cold-spray coatings applied with various process parameters. The ms coefficient of nickel is less than half of FeCo so some reduction in amplitude was expected. Cold-spray responses ranged from a moderate increase to > 40 dB reduction. The signal/noise ratios though were > 26 dB. Based on the edge reflection amplitudes and signal/noise, the inspection sensitivity is inferred to be manageable by instrument gain adjustments and, therefore, the cold-spray patch would be a viable alternative to an adhesive strip sensor that is not subject to coupling or adhesive degradation.

Keywords: cold spray, molten salt reactor, MSR, tungsten cold spray

INTRODUCTION The concept of magnetostrictive ultrasound in general and ms guided-wave ultrasound in particular is well understood by practitioners of the art and has been addressed by numerous researchers in the literature [1, 2]. Two principal configurations of patch ms sensing are shown in Figure 1. The sensor system consists of an electrically conductive meander shape, referred to as the coil, placed over the ms layer (either a FeCo adhesive strip or a cold- spray patch for the purpose of this publication). A perturbation in the magnetic field acting on an ms material with its magnetic domains directionally aligned gives rise to a stress wave. The nature of the stress wave (primarily shear- horizontal [SH-0] or lamb [A-0]) is dependent upon the relative alignment of the magnetic field perturbation and the domain orientation.

The acoustic wave is generated in the ms layer by inducing a high-current pulse or sinusoidal burst of one to several wavelengths at the desired inspection frequency that induces a transient magnetic field change. The acoustic wave transmits equally in both directions. Frequency selection is a trade-off between propagation distance, temporal/spatial resolution, and natural modes that may be enhanced or attenuated depending on material thickness, speed of sound, attenuation, etc. The amplitude of the acoustic wave is a function of the ms coefficient of the ms layer, the biasing field strength, the coil pulse current, lift-off between the coil and the ms layer, and a number of

Copyright 2019. This paper is intended for the sole use of registered attendees. No part of this publication or its contents may be copied, uploaded to the internet, or stored in any shared retrieval system. 78 other factors that are beyond the scope of this paper. The focus of this study was simply to compare the efficacy of a nickel (Ni) cold-spray ms patch layer compared to a commercial FeCo strip. The study examines both the SH-0 shear-horizontal wave mode (where coils are aligned with the magnetic bias field) and the A-0 Lamb wave mode (where coils are transverse to the magnetic bias field).

Figure 1: (left) SH-0; (right) Lamb [A-0] wave ms sensor. In either case, the sensors may be used singly as both transmit and receive units or in tandem in a pitch-catch mode.

MAGNETOSTRICTIVE DOMAIN ALIGNMENT Magnetostrictive domain alignment can be achieved by passing or swiping a strong permanent magnet over the ms material. This alignment is transitory in that the coil-generated magnetic pulse or burst will disrupt the preferential alignment and the signal response will degrade after minutes or seconds of exposure. This is not a problem for a “one-shot” field inspection supported by an operator who can incorporate the magnet swipe into the inspection procedure. The signal is typically acquired within milliseconds, well before the magnetic domain alignment is compromised. For a permanently installed continuous or periodic monitoring application where it is impractical to manually swipe the ms material, alternate approaches are required. Permanent magnets that bridge the coils can be conveniently and simply applied [3]. These magnets will be strongly attracted to the ms material even if the underlying substrate is a non-magnetic material such as stainless steel, aluminum, or even a non-metal-like carbon- fiber reinforced plastic. Most permanent magnets however do not perform well at high temperatures. Samarium cobalt (SmCo) magnets’ maximum recommended use temperature is 250–550°C with a Curie temperature of 700– 800°C. Neodymium magnets’ maximum recommended temperature is 230°C with a Curie temperature of 310°C. All permanent magnets loose some of their magnetic strength as temperature increases; and once temperatures exceed the Curie temperature, they lose all magnetic properties. Electromagnets may also be used and can be engineered to withstand significantly higher temperatures; however, their use requires wire/cable attachments and power to generate the electromagnet coil current.

MAGNETOSTRICTIVE MATERIAL The performance of the ms sensor is primarily governed by its ms coefficients. Ms coefficients related to the Joule effect [4] are quantified by the change in dimension as a material is subjected to a change in its magnetic field. The Joule magnetostriction coefficient is defined as λ = δL / L as the material passes from a magnetic field of zero to saturation. This is a complex measurement because the dimension change must be addressed in six dimensions, plus there is the consideration of the orientation of the imposed magnetic field. In practice, λ is only measured along the principal field axis. Coefficients of λ may be either positive or negative but, for sensor performance, only the absolute value is important. Larger is preferred. The saturation magnetostriction, λs, is calculated from the

difference between the maximum magnetostriction with the field parallel to a given direction (λs||) and that with the field perpendicular to the given direction (λs ). Assuming for simplicity that the medium is isotropic, the saturation magnetostriction is given by Equation 1 [5].⊥

79 (λssP) −λ( ⊥ ) =(3 2) λ s and is commonly expressed in dimensionless units of ( 3 2)λ s (Eq. 1)

The inverse ms effect or the Villari effect relates to the change of magnetic susceptibility of a material when subjected to a mechanical stress [6]. For high-temperature applications, one must also consider the Curie temperature above which all magnetic properties are lost. The ms coefficients and Curie temperatures of common or candidate ms materials are shown in Table 1.

Table 1: Candidate materials for ms sensors (from [5] except where otherwise noted). ms Coefficient Curie Material Notes (mostly related to cold spray) (3/2) λs × 10-6 Temperature Nickel (Ni) −50 354 Corrosion-resistant; commonly cold-sprayed

Cobalt (Co) −93 1120 Can be cold-sprayed; may need environmental control

Iron (Fe) −14 770 Can be cold sprayed; corrosion-susceptible Ferrous Cobalt 87 500 Standard for adhesive strip alloyed 50%/50% (FeCo) Galfenol >200 670 No cold-spray history No cold-spray history but similar to ceramics that have Met-Glass 60 370 been sprayed

Iron-cobalt (FeCo) material is most commonly used in an adhesive strip configuration for ms sensors. In some cases, this material is annealed within a strong magnetic field to preferentially orient the magnetic domains and make them more easily aligned. This is the configuration used as a reference to compare to the cold-spray patch described below.

COLD-SPRAY MAGNETOSTRICTIVE SENSOR Cold-spray coatings are deposited by accelerating 10–100 micron powders to 2–3 × speed of sound to impact against a substrate. Typical substrates include carbon steel, stainless steel, Hastelloy, aluminum, zirconium, and carbon- fiber-reinforced plastic. A typical spraying configuration is shown in Figure 2. The spray can be applied either robotically or manually. Typically each spray pass deposits 1–3 mills (0.2–0.7 mm) of material, but repeated passes can build up more than a centimeter of material. Moreover, the coating is virtually completely in compressive stress. Even though the powder is heated to 300+ °C, thermal stresses are minimal compared to thermal or plasma spray

Figure 2: (left) Schematic of cold spray system and (right) image of manual application cold-spray system. Courtesy of VRC Metal Systems. coatings. This is one reason why cold-spray coatings are favored for corrosion resistance and particularly stress corrosion cracking.

First Round CPNi Cold-Spray Patch vs. FeCo Adhesive Strip Comparison To compare the performance of CPNi cold-spray patch sensors80 to the adhesive FeCo strip sensors, two 0.25 in. × 2 ft. × 4 ft. (6.35 mm × 610 mm × 1219 mm) plates and one 0.5 in. ×2 ft. × 4 ft. (12.7 mm × 510 mm ×1219 mm) plate with cold-spray patches by Vendor A and FeCo adhesive strips were prepared. Vendor A’s process parameters are shown in Table 2. The sensor coil and magnet were then applied similarly to both the cold-spray patches and the FeCo strip to compare the edge-wall reflections of the two conditions. The ms magnetic bias was achieved and examined in three ways: (1) by swiping a permanent magnet (~ 0.35 tesla at magnet face), (2) by biasing with a stationary permanent magnet (~0.52 tesla @ magnet face), and (3) by biasing with a stationary electromagnet (assumed to be similar strength to permanent magnet at magnet face). Initial results were essentially identical between the electromagnet and the permanent magnet so continuing studies only examined the swipe magnet and the permanent magnet. The swipe magnet responses were ~ 10–25% of the permanent magnet bias case. These comparisons are more fully discussed in [7] and [8] and will not be further addressed here.

Table 2: Cold spray process parameter configurations from Vendors A, B, and C. Company Vendor A Vendor B Vendor C High-Pressure High-Pressure Low-Pressure High-Pressure Low-Pressure Spraying Parameter Cold Spray Cold Spray Cold Spray Cold Spray Cold Spray

Process Gas Nitrogen (N2) N2 Air N2 N2 or Air

Pressure (bar) 55 40 7 60 6

0.85–2.5 (N2) Gas Flow Rate N/A Not available 793 l/min. 324 l/min. m3/min. Powder Feed Rate (kg/h) N/A 13.5 4.5 4.08 1.2 Ni (60%) + Particle Composition Ni (99.7%) Ni (99.7%) Ni (99.7%) Ni (pure) Al2O3 (40%) Particle Diameter (µm) 15–45 5–50 5–50 15–45 5–20 Gun Traverse Speed N/A 10–25 10–25 8.33 2.5 (mm/s)

Data was acquired using an Innerspec PowerBox H with a medium-range ultrasonic testing instrument and, in all cases, the coil was an Innerspec RF coil (205C0469) with a pitch of 0.5 in., thereby producing an SH-0 wavelength of 1.0 in. The excitation waveform was a 2-cycle pulse with 600 volts. Displays are gained for the plate edge reflections to be ~ 80% of full scale to graphically show the signal and noise amplitude. The relative response amplitudes (relative response = measured response/max-response [from permanent magnet biased FeCo strip]) are shown in the bar chart of Figure 3 for SH-0 wave responses and Figure 4 for Lamb wave responses.

Under all mag-bias cases with the ¼ in. plate, the FeCo strip produced clean SH-0 wave back-wall reflections (Figure 5, left). The initial cold-spray patch-generated reflections were not as strong but were still clearly evident. The CPNi cold-spray ms patch thicknesses were 1 mm or more. Signals from these thick cold-spray patches were contaminated by noise that seemed to be associated with a ringing echo within the CPNi cold-spray patch (Figure 5, right). To mitigate this, the CPNi cold-spray patch was machined to ~ 0.5 mm (0.020 in.) and subsequently to 0.25 mm (0.010 in.). Neither of the thinned CPNi patches exhibited the ringing noise within the ms patch (Figure 6). coatings. This is one reason why cold-spray coatings are favored for corrosion resistance and particularly stress corrosion cracking.

First Round CPNi Cold-Spray Patch vs. FeCo Adhesive Strip Comparison To compare the performance of CPNi cold-spray patch sensors to the adhesive FeCo strip sensors, two 0.25 in. × 2 ft. × 4 ft. (6.35 mm × 610 mm × 1219 mm) plates and one 0.5 in. ×2 ft. × 4 ft. (12.7 mm × 510 mm ×1219 mm) plate with cold-spray patches by Vendor A and FeCo adhesive strips were prepared. Vendor A’s process parameters are shown in Table 2. The sensor coil and magnet were then applied similarly to both the cold-spray patches and the FeCo strip to compare the edge-wall reflections of the two conditions. The ms magnetic bias was achieved and examined in three ways: (1) by swiping a permanent magnet (~ 0.35 tesla at magnet face), (2) by biasing with a stationary permanent magnet (~0.52 tesla @ magnet face), and (3) by biasing with a stationary electromagnet (assumed to be similar strength to permanent magnet at magnet face). Initial results were essentially identical between the electromagnet and the permanent magnet so continuing studies only examined the swipe magnet and the permanent magnet. The swipe magnet responses were ~ 10–25% of the permanent magnet bias case. These comparisons are more fully discussed in [7] and [8] and will not be further addressed here.

Process Gas Nitrogen (N2) N2 Air N2 N2 or Air

Pressure (bar) 55 40 7 60 6

81 Figure 3: (left) SH-0 relative amplitude near-edge response of FeCo strip plus various cold-spray patch configurations. (right) Corresponding signal-to-noise ratios.

Figure 4: (left) Lamb wave relative amplitude near-edge response of FeCo strip plus various cold-spray patch configurations. (right) Corresponding signal-to-noise ratios.

Figure 5: (left) ¼ in. plate with coil over FeCo adhesive strip. (right) ¼ in. plate with coil over 1 mm CPNi patch and with stationary permanent magnet.

82 Figure 6: (left) ¼ in. plate with coil over 0.5 mm thick CPNi patch with stationary permanent magnet. (right) ¼ in. plate with coil over 0.25 mm CPNi patch and with stationary permanent magnet.

Edge reflections from the Lamb wave mode were also clear in all cases and the response from the FeCo strip was comparable to the cold-spray patches. The Lamb wave amplitude response from Vendor A was surprisingly larger than from the FeCo strip (Figure 7). Positioning and configurations of the magnet significantly influenced the Lamb wave mode response. Amplitude responses reported were based on a magnet configuration where the magnet was offset-spaced 1 inch above the coil and ms material. Optimization of the Lamb wave mode magnet position may increase the amplitude responses but such an optimization was not part of this investigation.

Figure 7: (left) Lamb wave response with FeCo strip. (right) Vendor A CPNi cold-spray patch Lamb wave response.

Second Round CPNi Cold-Spray Patch vs. FeCo Adhesive Strip Comparison Following the initial tests of cold spray and adhesive strip comparisons, a second round of cold-spray patches were tested to assess different cold-spray process parameters. Two other vendors were chosen to produce plates similar to those explored with Vendor A under the first round of tests. Following the guidance from the first-round tests, all patches were ~0.25 mm thick. The cold-spray process parameter configurations for both round one and two tests are shown in Table 2. Note that there are other process parameters not reported that may also be as important or more important. These parameters are simply meant to generally describe the different processes.

83 Figure 6: (left) ¼ in. plate with coil over 0.5 mm thick CPNi patch with stationary permanent magnet. (right) ¼ in. plate with coil over 0.25 mm CPNi patch and with stationary permanent magnet.

Figure 7: (left) Lamb wave response with FeCo strip. (right) Vendor A CPNi cold-spray patch Lamb wave response.

CONCLUSIONS • From the Round 1 effort focusing on SH-0 mode plate edge reflections, the thick (1–2+ mm) CPNi cold-spray patch produced excessive noise from the acoustic wave reflecting within the patch from the patch edges. Thinning the patch to 0.5 mm and subsequently to 0.25 mm substantially eliminated this patch-edge noise. • Thinning the CPNi patch from 0.5 mm to 0.25 mm increased the amplitude response by ~6 to 10 dB. The general conclusion from this effort was that a coating specification could be fairly forgiving since coatings from 0.2–0.5 mm produced patch-edge-reflection noise-free responses. • From the Round 1 and 2 tests, the FeCo strip had the highest amplitude responses for all SH-0 mode tests. The responses of the cold-spray patches ranged from 35% to less than 1% of the FeCo strip responses. The cold-spray configuration and differences among cold-spray vendors caused significant variations in the amplitude responses for both the SH-0 and Lamb (A-0) wave modes.

With some outlying exceptions, the highest responses were generally from the high-pressure N2 spray configurations and the lowest responses were from the low-pressure air configurations.

• The highest Lamb wave amplitude response was measured on Vendor A’s high-pressure N2 patch. The highest response Lamb wave reflections were produced with a significant liftoff between the magnet and the plate. Little or no effort was made to optimize this configuration; however, it was noted that improvement in the Lamb wave signal response may be expected if the test configuration were optimized (liftoff, coil spacing, magnet configuration, etc.). • Since most inspection regimes do not depend on response amplitudes as long as a response is detectable, it is noted that the instrument used has a wide dynamic range; and in virtually all cases, both near- and far-edge responses were detectable from both the SH-0 and Lamb wave mode. Moreover, the signal-to-noise ratios (with some outlying exceptions) were in excess of 26 dB or greater than a factor of 20. • The practice of swiping a permanent magnet across the strip or the cold-spray patch produced a significantly lower response than with a permanent magnet bias. Moreover, if the coil was energized/pulsed, the edge response diminished rapidly. Re-swiping the ms strip or patch returned the amplitude response to a high level. This swiping practice is convenient for manual testing but not practical for permanent-monitoring test configurations as would likely be used for the cold- spray patch configuration. • Based on the edge reflection amplitude responses and signal-to-noise ratios, it is inferred that sensitivity to pit or crack flaws would be similarly detectable and, therefore, the cold-spray patch would be a viable alternative to a FeCo adhesive strip sensor that is not subject to adhesive degradation.

ACKNOWLEDGMENTS Funding provided by the U.S. Department of Energy’s Technology Commercialization Fund. This work was performed collaboratively between Innerspec Technologies and Pacific Northwest National Laboratory (PNNL) to verify feasibility of a ms cold-spray sensor following a PNNL US patent application (62/430,093). The technology has been exclusively licensed to Innerspec Technologies.

REFERENCES [1] Kim, Y. Y., and Kwon, Y. E., 2015, "Review of Magnetostrictive Patch Transducers and Applications in Ultrasonic Nondestructive Testing of Waveguides," Ultrasonics, 62, pp. 3-19. [2] Thompson, R. B., 1979, "Generation of Horizontally Polarized Shear Waves in Ferromagnetic Materials Using Magnetostrictively Coupled Meander-Coil Electromagnetic Transducers," App. Phys. Lett., 34(2), pp. 175-177. [3] Vinogradov, S., Duffer, C., and Light, G. M., 2014, "Magnetostrictive Sensing Probes for Guided Wave Testing of High Temperature Pipes," Mater. Eval., 72(6), pp. 803-811. [4] Joule, J. P., 1847, "XVII. On the Effects of Magnetism upon the Dimensions of Iron and Steel Bars," The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science, 30(199), pp. 76-87. 84 • Thinning the CPNi patch from 0.5 mm to 0.25 mm increased the amplitude response by ~6 to 10 dB. The general conclusion from this effort was that a coating specification could be fairly forgiving since coatings from 0.2–0.5 mm produced patch-edge-reflection noise-free responses. • From the Round 1 and 2 tests, the FeCo strip had the highest amplitude responses for all SH-0 mode tests. The responses of the cold-spray patches ranged from 35% to less than 1% of the FeCo strip responses. The cold-spray configuration and differences among cold-spray vendors caused significant variations in the amplitude responses for both the SH-0 and Lamb (A-0) wave modes.

With some outlying exceptions, the highest responses were generally from the high-pressure N2 spray configurations and the lowest responses were from the low-pressure air configurations.

REFERENCES [1] Kim, Y. Y., and Kwon, Y. E., 2015, "Review of Magnetostrictive Patch Transducers and Applications in Ultrasonic Nondestructive Testing of Waveguides," Ultrasonics, 62, pp. 3-19. [2] Thompson, R. B., 1979, "Generation of Horizontally Polarized Shear Waves in Ferromagnetic Materials Using Magnetostrictively Coupled Meander-Coil Electromagnetic Transducers," App. Phys. Lett., 34(2), pp. 175-177. [3] Vinogradov, S., Duffer, C., and Light, G. M., 2014, "Magnetostrictive Sensing Probes for Guided Wave Testing of High Temperature Pipes," Mater. Eval., 72(6), pp. 803-811. [4] Joule, J. P., 1847, "XVII. On the Effects of Magnetism upon the Dimensions of Iron and Steel Bars," The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science, 30(199), pp. 76-87. [5] Dapino, M., 2004, "On Magnetostrictive Materials and Their Use in Adaptive Structures," Struct. Eng. Mech., 17(3-4), pp. 303-329. [6] Villari, E., 1865, "Ueber die Aenderungen des magnetischen Moments, welche der Zug und das Hindurchleiten eines galvanischen Stroms in einem Stabe von Stahl oder Eisen hervorbringen," Annalen der Physik. [7] Glass III, S. W., Lareau, J. P., Ross, K. A., Ali, S., Hernandez, F., and Lopez, B., 2018, "Magnetostrictive Cold Spray Sensor for Harsh Environment and Long-Term Condition Monitoring," Proc. 45th Annual Review of Progress in Quantitative Nondestructive Evaluation, ASNT. [8] Glass III, S. W., Lareau, J. P., Ross, K. A., Ali, S., Hernandez, F., and Lopez, B., 2018, Magnetostrictive Cold Spray Sensor Feasibility Assessment, Pacific Northwest National Laboratory, Richland, WA.

MWM-Array and MR-MWM-Array Eddy Current Testing for Piping and Vessels MWM-Array and MR-MWM-Array Eddy Current Testing for Piping and N. Goldfine, T. Dunford, A. VesselsWashabaugh, S. Chaplan, and K. Diaz N. Goldfine T. Dunford, A. Washabaugh, S. Chaplan, K. Diaz

JENTEK Sensors, Inc. 121 Bartlett Street Marlborough, MA 01752 (781)(781 373-9700;) 373-9700 fax; fax (781) (781 642-7525;) 642-7525 email; [email protected] [email protected]

ABSTRACT The MWM®-Array and MR-MWM-Array are eddy current sensor arrays with single or dual rectangle drive configurations and linear arrays of inductive or magnetoresistive sensing elements. This presentation describes a range of applications for these sensors including stress/strain imaging/monitoring, corrosion imaging (internal and external) with and without coatings, insulation or fireproofing for piping, vessels and structures, and crack detection and depth measurement for fatigue and stress corrosion cracking. These solutions use multivariate inverse methods with precomputed databases called HyperLattices that enable estimation of multiple unknown properties of interest such as magnetic permeability, lift-off, pipe wall thickness, insulation thickness and weather jacket properties. The unique approach improves robustness to variations in field conditions (e.g. insulation sagging that causes substantial variation in insulation thickness around a pipe, or temperature variations that make inspection during operation challenging, or surface conditions that make crack or pit depth measurement more difficult). Handheld and larger portable instruments are also described that are uniquely capable of providing reliable imaging of multiple unknown properties (as required for reliable corrosion imaging through coatings or difficult surface conditions). These instruments provide simultaneous measurement of all channels in the arrays (no multiplexing) to enable improved image reliability and even rescaling of crack responses for cracks falling between channels. Furthermore, impedance values at up to three frequencies are recorded simultaneously at every sensing element. This is essential for reliable correction for uncontrolled variations, such as changes in lift-off (sensor proximity to the surface) and/or magnetic permeability, to provide accurate pipe wall thickness measurement for example. Basic methods are described along with practical application results.

Keywords: MWM, Eddy Current, Inverse Methods, Pipes, Vessels, Coatings, Arrays

BACKGROUND ON USE OF MODEL-BASED MULTIVARIATE INVERSE METHODS (MIMS) MWM-Array and MR-MWM-Array Eddy Current Testing The primary innovations that make MWM-Array and MR-MWM-Arrays unique and generally more capable than other magnetic field nondestructive testing (NDT) methods are (1) the use of precomputed databases of sensor responses (called Grids, Lattices and Hyper-lattices) to perform Multivariate Inverse Methods (MIMs); and (2) the design of sensors and instrumentation with the first objective being improvement of signal-to-noise not only through signal enhancement, but often most importantly through reduction of unmodeled behavior contributions to the sensor responses. This sounds theoretical, but it has huge practical implications and enables improved performance for eddy current testing (ET) of layered material constructs such as pipelines, coating systems and aircraft structures.

First, in order to use this method it is necessary to meet two difficult challenges: (1) design an ET sensor winding construct that can be accurately modeled to enable accurate prediction of the sensor response over the range of Material Under Test (MUT) properties of interest; and (2) develop an instrument that not only provides extremely accurate and sufficiently precise measurements over the frequency and impedance range of interest for the ET sensor and the MUT properties, but also does this for multiple parallel channels and multiple frequencies simultaneously. All aspects of this second requirement are not generally met by other ET systems, since these other systems do not typically attempt to rescale sensor responses for variability in operating conditions or MUT

variability. To provide robust and reliable performance an NDT system should provide the same defect response with substantial variations in MUT or other operating conditions such as lift-off (proximity of the sensor windings to the nearest conducting or magnetic material), coating/insulation thickness (sometimes but not always part of effective lift-off), or position of a defect relative to the multiple elements within an array or relative to an edge or other geometric feature.

One sensor construct that meets the first requirement is the MWM-Array (with inductive sensing elements) or the MR-MWM-Array (with magnetoresistive sensing elements). These sensors both use rectangular drive conductors with long linear drive segments and a row of sensing elements placed at a selected distance (defined as λ/4) from one of the linear drive segments. Note that when the sensing elements are placed at the center of a simple dual rectangle MWM-Array drive construct, then λ is the spatial wavelength of the primary mode of the magnetic field created by this drive winding construct. This spatial wavelength is proportional to the magnetic field depth of penetration at lower frequencies, while the depth of penetration is inversely proportional to the drive current frequency at higher frequencies. For more on MWM sensor constructs refer to references [1-7]. Note that the original MWM sensors used a meandering (square wave shaped drive), thus the original name Meandering Winding Magnetometers. The name magnetometer was used instead of eddy current sensors, since early work included magnetic fluids and other nonconducting magnetic materials for which no eddy currents were induced. For the aperiodic MWM-Array constructs and even the enhanced MWM sensors a square wave (“meandering”) drive is not used. New periodic drives in the MWM outperform a simple square wave, and MWM-Arrays use single or dual rectangle drive constructs. Even circular MWM and MWM-Array formats exist that take advantage of the same requirements and MIMs using appropriate models. Note some MWM-Arrays use multiple rows of sensing elements at more than one distance from the drive conductor (drive-to-sense gap). This allows estimation of more unknowns for complex heterogeneous materials, such as composites. These “segmented field” arrays are not addressed in this paper.

Material Under Test (MUT) Layups This paper considers a progression of increasingly complex MUT layups, beginning with simple infinite half-spaces and foils. Figure 1 shows this progression, with simple schematics that represent the sensor (shown above the MUT as a dashed line with a lift-off gap between the sensor and the MUT). The caption of Figure 1 lists all the MUT layups considered in this paper. Note that each of these layups is assumed to be comprised of layers with uniform electrical conductivity and magnetic permeability as a function of depth within the individual layer. Note that it is also possible to assume a linear variation in these properties or other more complex property variations, but it is typically not necessary for most NDT applications. Furthermore, for pipes, hole inspection, and other such curved features cylindrical coordinates are used to account for behavior variations and generate the sensor response databases (HyperLattices).

One type of damage that is of broad interest is general corrosion, which reduces the thickness of a material layer over a large surface area compared to the sensor’s sensing element footprint (note that the footprint is larger than the actual size of the individual sensing elements and defines the region that each sensing element is sensitive to on the surface of the MUT). For damage such as general corrosion, a simple uniform thickness layer model provides a sufficient representation of the damage. However, for local damage, such as small corrosion pits and cracks, the damage is not well represented by a typical uniform thickness, layered media model for the underlying physics. Thus, for small local damage, as described below, the damage will be represented as a perturbation (deviation or change) from the simple uniform layer model. This perturbation approach is often sufficient for detection and even sizing of damage, if a correlation relationship is developed between the perturbed sensor response and the damage characteristic of interest, such as crack depth. For improved defect sizing, it is also valuable to develop signature libraries (either empirically or using 3 dimensional numerical modeling tools that are empirically validated). These libraries are then searched using the actual lift-off or other defining parameter, to identify the correct signature for filtering of MWM-Array data to both enhance defect responses and suppress inconsequential responses.

MWM-Array MWM-Array Cladding MWM-Array MWM-Array h s , D , μ h s D c c c h (lift off) h Coating 87 s D µ (permeability) Coating s D g sV, µV s (conductivity). . Coating layer . Cladding (a) (b) (c) (d)

MWM-Array MWM-Array MWM-Array MWM-Array

h h h ho o µ µ D hi mesh µ, i D s s wj, wj s µ , s µV, sV D P P DP V

Near Side Corrosion Far Side Corrosion Insulation and Insulation and Weather Jacket Weather Jacket (e) (f)

Figure 1: Schematics for (a) infinite half-space; (b) conducting, non-magnetic foil; (c) single layer conducting and non-magnetic coating on a thick conducting and non-magnetic base material; (d) single layer conducting and slightly magnetic coating with a disbond or blister between the coating and a magnetic and conducting base material- such as for cladding overlay on an internal vessel surface; (e) corrosion, wall loss, for with an insulating, non-conducting, coating on a conducting and magnetic single layer – such as a pipe or a vessel; (f - left) corrosion, or wall loss, for a magnetic and conducting layer, with insulation and a weather jacket – such as for refiner piping; (f - right) corrosion, or wall, loss for an concrete layer with internal wire mesh - such as for underwater pipeline weight coat or fireproofing for vessel skirts.

Multivariate Inverse Methods (MIMs) Precomputed databases have been generated for each of the configurations in Figure 1, using the simple uniform layer modeling method. These databases are generated off line and used by the JENTEK GridStation software to perform the MIMs. The GridStation software uses a rapid solver (essentially an intelligent database search method) to find the solution for multiple unknown MUT properties of interest for each position of the sensing elements along the inspected or monitored surface of the MUT. Figure 2 shows examples of Grids and Lattices for the two and three unknown layups shown in Figure 1. Note that Grids are for two unknowns and look like a curved sheet of graph paper; Lattices are for three unknowns and are simply a set of Grids. Each Grid in the Lattice represents the same range for two selected unknown variables. For each Grid, the third unknown variable is held constant. This third unknown is then varied from Grid to Grid. Hyper-lattices are difficult to represent visually, since they are used to solve problems with four or more unknowns.

libraries are then searched using the actual lift-off or other defining parameter, to identify the correct signature for filtering of MWM-Array data to both enhance defect responses and suppress inconsequential responses.

MWM-Array MWM-Array Cladding MWM-Array MWM-Array h s , D , μ h s D c c c h (lift off) h Coating s D µ (permeability) Coating s D g sV, µV s (conductivity). . Coating layer . Cladding (a) (b) (c) (d)

MWM-Array MWM-Array MWM-Array MWM-Array

h h h ho o µ µ D hi mesh µ, i D s s wj, wj s µ , s µV, sV D P P DP V

Near Side Corrosion Far Side Corrosion Insulation and Insulation and Weather Jacket Weather Jacket (e) (f)

(a)

(b) (c) (d)

(e)

(f) Figure 2: Grids and Lattices corresponding to schematics from Figure 1, (a) through (f). Grids (a-left) and (b-left) were generated in 1989 [4, 7]. The (b-left) grid is for conductivity and thickness. Before this grid is used, a higher frequency measurement provides the lift-off in a “hierarchical” 3-unknown inverse method.

The MIM can be performed on all unknowns simultaneously solving for them by searching the Grid, Lattice, or Hyper-lattice for the best solution, using a single frequency method or a multiple frequency method. Note that each frequency provides a magnitude and phase measurement (or complex real and imaginary part measurement) for the impedance (sensing element voltage divided by the primary current). Thus, for two unknown problems a single frequency is sufficient, since the phase and magnitude measurements each provide the equivalent of one equation that can be solved. But for three or four unknowns, two frequencies are needed since we need either the same number or more equations (measurements) than unknowns – as in simple linear algebra problems. Note that because the physics equations are inherently nonlinear, we must use a search routine to find the solution – thus, by precomputing the solution spaces and searching them rapidly, we can solve complex problems very quickly. For problems with more than four unknowns, three frequencies or more are needed. In the following sections, some brief approaches are described for a few important problems – including imaging of Stress Corrosion Crack colonies through coatings, imaging of corrosion under insulation (CUI) for piping or vessels, sizing of hydrogen blisters under overlay cladding inside vessels, and imaging of corrosion under fireproofing with wire mesh [5-9].

Stress Corrosion Cracking (SCC) Problem Description Figure 2a (middle and right) above provides two different representations of a pipeline SCC colony mapping method. In Figure 2a (middle) the MUT has been represented by a thick layer with a constant magnetic permeability and the effective electrical conductivity is allowed to vary. This is a two-unknown problem with the unknowns being lift-off and electrical conductivity. Figure 2a (right) instead, assumes the electrical conductivity is constant and allows the magnetic permeability to vary as one of the two unknowns, along with lift-off as the second unknown. Goldfine [4] showed that the magnetic permeability and the electrical conductivity cannot be measured independently at relatively high frequencies (without use of a variable bias field); thus, either of these representations is sufficient for use of the perturbation method for crack detection and sizing, described earlier. In a perturbation method, the crack shows up as a change in one of the unknowns. This works best if the change is primarily in one unknown of the uniform layer representation. However, for cracks, the estimated lift-off value is also affected for relatively open cracks. Tight closed cracks have a smaller effect on lift-off. As shown in Figure 3, the magnetic permeability response can also be correlated very well with crack depth and this can be displayed as a C-Scan image of depth for each crack or as a B-Scan for the individual channels. Figure 4 shows examples of crack maps that match very well with the actual crack morphology shown also in the figure. Note that a similar method can be used for pit depth imaging, where the lift-off is the primary unknown of interest and a smooth layer (e.g. a thin but hard plastic) is used to enable the flexible MWM-Array to be scanned at relatively constant lift-off away from the pits. Note that fully automated systems have been developed for full circumferential scanning of pipes of variable diameter for SCC crack depth mapping.

Figure 3: (left) Data showing magnetic permeability response can also be correlated very well with crack depth; (right) displayed as a C-Scan image of depth for each crack or as a B-Scan for the individual channels.

Figure 4: (left, top and bottom) MWM-Array crack response C-Scan images; (right, top and bottom) corresponding enhanced photographs of the crack morphology.

Corrosion under Insulation, Internal and External (with and without a weather jacket) Figure 5 and Figure 6 provide results for corrosion under insulation imaging without and with a weather jacket, respectively. For applications without a weather jacket, this is a simple three unknown problem, if we assume the electrical conductivity of the pipe wall layer is constant. If we add the weather jacket, then this becomes a five- unknown problem. When we solve five-unknowns, we typically use a hierarchical method, where we solve for a subset of the unknowns first.

Figure 5: Images of lift-off (which is the same as the coating thickness in this case) and magnetic permeability for a riser with a relatively insulating coating. Both the lift-off and the magnetic permeability images show the defect, even when the insulating coating thickness is varied. Note that the use of lift-off for the defect response in this method requires careful maintenance of relatively constant sensor proximity to the coating, since any variation will show up in the lift-off response.

Figure 6: Pipe wall thickness images are shown above for imaging of corrosion through weather jacket and insulation. (top) Insulation thickness is 2 inches (50mm), the weather jacket is 0.020inches (0.5 mm) thick aluminum, the pipe walls are 0.25 inches (6.35mm) and the corrosion is external. (Bottom) Insulation thickness is 2 inches (50mm), the weather jacket is 0.020 inches (0.5 mm) thick aluminum, the pipe walls are 0.5 inches (12.7mm) and the corrosion is internal.

Hydrogen Blister Imaging and Crack detection with Overlay Cladding One challenging application is the imaging of disbonds between overlay cladding and the base material for large vessels. To solve this problem the disbond is modeled as a simple gap and this gap is estimated as one of the unknowns. Details for this application are provided in a complementary paper at this conference [5]. This problem can be solved as a three unknown problem for lift-off, cladding thickness and gap if we assume that the cladding conductivity and magnetic permeability are constant and known, and that the base material conductivity and permeability are also constant and known. For the more complex problem where the goal is to detect cracks in the base material, the base material magnetic permeability can be added as a fourth unknown.

Corrosion Under Fireproofing (CUF) or Weight Coat CUF imaging or corrosion imaging through weight coat for underwater pipelines is also of interest. This problem adds the complexity of having a magnetic layer (the wire mesh) inside an insulating (non-conducting) layer of concrete. This problem can be solved by simply estimating92 the magnetic permeability of the wire mesh and then treating it as a constant, if we can assume we know the approximate position of the mesh within the concrete layer. One way to accomplish this is to assume that the nominal vessel or pipe wall thickness is known at a few locations and then use these locations to estimate the wire mesh properties and position, first. This has been demonstrated successfully both in the laboratory and in the field, and is also described in a complimentary paper at this conference [6].

SUMMARY This paper provides a brief overview of the MWM-Array and MR-MWM-Array technology with a focus on the MIMs using Grids and Lattices for complex multiple unknown problems. Schematic representations have been described for problems that range from a simple foil, to a pipe with insulation and weather jacket or a vessel skirt covered by concrete with wire mesh. The key is to identify the unknowns of interest and to solve them rapidly with a MIM. The MIM relies on having a sensor that has a response that can be accurately predicted over the range of MUT properties of interest and an instrument that can provide accurate measurement of impedance at multiple frequencies over impedance range of interest and at all array channels simultaneously. This method has numerous practical applications, such as SCC crack depth measurement, internal and external corrosion imaging, and hydrogen blister volume estimation. This method is in use with the MWM-Array and MR-MWM-Array sensors, providing reliable services for a range of applications.

REFERENCES (1) Denenberg, S.A., T.M. Dunford, N.J. Goldfine, and Y.K. Sheiretov, 2013, Method and Apparatus for Inspection of Corrosion and Other Defects through Insulation, US 20130124109 A1. (2) Schlicker, Darrell E.; Goldfine, Neil J.; Washabaugh, Andrew P.; Walrath, Karen E.; Shay, Ian C.; Grundy, David C.; Windoloski, Mark.; “Test Circuit Having Parallel Drive Segments and a Plurality of Sense Elements,” US Patent # 7,049,811 B2, May 23, 2006. (3) Goldfine, Neil J.; Zilberstein, Vladimir A.; Schlicker, Darrell E.; Grundy, David C.; Shay, Ian C.; Washabaugh, Andrew P.; “High Resolution Inductive Sensor Arrays for Material and Defect Characterization of Welds,” US Patent # 6,995,557 B2, Feb. 7, 2006. (4) Goldfine, N. J., “Uncalibrated, Absolute Property Estimation and Measurement Optimization for Conducting and Magnetic Media Using Imposed ω-k Magnetometry,” Sc.D. thesis, Department of Mechanical Engineering, MIT, September, 1990. (5) Goldfine, N., Sheiretov, Y., Manning, B., Thomas, Z., Dunford, T., Denenberg, S., Al Rushaid, R., Haught, F., Bayangos, J., Minachi, A., “Inspection of Steel Vessels with Cladding Overlay, using MWM-Array Technology” 7th Middle East NDT Conference & Exhibition, Manama, Kingdom of Bahrain, September 13 - 16, 2015. (6) Goldfine, N., Manning, B., Thomas, Z., Sheiretov, Y., Denenberg, S., Dunford, T., Haque, S., Al Rushaid, R., Haught, F., “Imaging Corrosion under Insulation and under Fireproofing, using MR-MWM-Arrays,” 7th Middle East NDT Conference & Exhibition, Manama, Kingdom of Bahrain, September 13 - 16, 2015 (7) Goldfine, N., “Magnetometers for Improved Materials Characterization in Aerospace Applications,” Materials Evaluation, Vol. 51, No. 3, p. 396-405; March 1993.

permeability are also constant and known. For the more complex problem where the goal is to detect cracks in the base material, the base material magnetic permeability can be added as a fourth unknown.

Corrosion Under Fireproofing (CUF) or Weight Coat CUF imaging or corrosion imaging through weight coat for underwater pipelines is also of interest. This problem adds the complexity of having a magnetic layer (the wire mesh) inside an insulating (non-conducting) layer of concrete. This problem can be solved by simply estimating the magnetic permeability of the wire mesh and then treating it as a constant, if we can assume we know the approximate position of the mesh within the concrete layer. One way to accomplish this is to assume that the nominal vessel or pipe wall thickness is known at a few locations and then use these locations to estimate the wire mesh properties and position, first. This has been demonstrated successfully both in the laboratory and in the field, and is also described in a complimentary paper at this conference [6].

93 The use of the pull-off test method to characterize the performance of a concrete repair system

Evan Karunaratne, Dr. Julie Ann Hartell, and Dr. Norbert Delatte

The Use of the Pull-OffOklahoma Test State UniversityMethod to Characterize the PerformanceCivil and Environmentalof a Concrete Engineering Repair System 320 Engineering South Evan Karunaratne, Dr. Julie Ann Hartell, and Dr. Norbert Delatte Stillwater, OK 74078 (405) 436Oklahoma-5234; [email protected] University Civil and Environmental Engineering (405) 744320-5222; Engineering [email protected] South Stillwater, OK 74078 (405) 744-5189; [email protected] (405) 436-5234; email [email protected] (405) 744-5222; email [email protected] (405) 744-5189; email [email protected]

Abstract The pull-off test or "bond-test" is considered as a nondestructive testing method generally used in the concrete repair industry to measure the adhesion between a coating and a concrete's surface. The method allows for better quality control of bond performance when applying coatings or overlays on concrete, in addition to providing a better understanding of the applied coating and substrate strength. In particular, the strength values obtained from this test let the user know if there is an issue regarding the coating testing or the substrate surface preparation leading to disbondment. For example, the moisture content of a concrete material before the application of a coating causing insufficient adhesion. For this study, differences between concrete surface preparations and coating materials are investigated to understand the field performance of the concrete repair system better and propose adequate repair methods to ensure optimal performance. Through testing of various epoxy and polyaspartic polyuria based coatings, the pull-off test was used to better understand the bonding performance between the coating and the surface of the concrete. In addition to this, the influence of internal moisture and surface moisture conditions along with coating layer thickness were evaluated to see how these affected the bond strength between the concrete and the coating. Experimental data that the internal moisture of concrete can affect the bond performance of the coating material even if the surface moisture was low at the time of application. Also, the test may not be suitable for determining the performance of a coating if its thickness is a parameter of concern. Introduction

As the repair and assessment of concrete structures have been changing, so have the ways of testing them. Non- destructive testing methods have grown in increasing popularity as they do not damage the structure as much regarding serviceability or even aesthetics. The test method discussed in this paper is a partially destructive testing method (Nepomuceno and Lopes 2017). In addition to only being a partially destructive test method, this method is a quick and easy to run, permitting testing at close locations (Ghavidel, Madandoust, and Ranjbar 2015). ASTM Standard D7234 has been used to test the adhesive strength if coatings to concrete. This test method involves the bond strength of a coating applied to concrete. What makes this standard unique is the variability of results that can point towards different issues regarding the coating and concrete substrate. The bond of a particular coating by the adherence of a coating to a substrate. The bond is through the microcracking from surface preparation and the porosity of the substrate itself (Sadowski 2013).

Copyright 2019. This paper is intended for the sole use of registered attendees. No part of this publication or its contents may be copied, uploaded to the internet, or stored in any shared retrieval system. 94 Recent research conducted for the applicability of this test method includes the application of a coating under SSD or dry conditions. The result of this has been that a dry coating application is generally better than an SSD condition, which has been shown to decrease the bond strength. When a coating bonds to a substrate, the substrate draws in the moisture from the coating to help the coating penetrate deeper into the substrate binding to the rough surface of the concrete (Bentz et al. 2018). In addition to research for wet or dry substrate applications, additional studies have analyzed data on the variability of results from this test method. This test method has a high variability between samples, but if there are enough samples, it is still representative of the concrete and coating mechanism. With these results, it is possible to get data representative of the structure (Ramos et al. 2012). Typical applications of this test are determining the performance if a new coating material, the quality of the repair, and the long term performance of a coating as well. What the different failure mechanisms tell us is the quality of the bond between a coating and concrete, the quality if the coating material itself, and different mechanisms affecting the coating itself. The main failure types discussed in this are the failure of the adhesive, the coating, the substrate, and a combination of the coating and the substrate. The failure of the adhesive is due to improper curing or the application of the adhesive itself. A failure of the coating can suggest that something is happening to the coating itself. The substrate failure means that the coating bond is at least as strong as the substrate itself. The results from running these tests can help give an idea of what is happening to the bond between a coating and a substrate based on the failure type and variation between samples (Mitchell et al. 2010). The limitations of the test are the coefficients of variation between samples, the load rate of the device, surface characteristics of the concrete, and adhesive/coating thickness variance. In particular, the misalignment of the testing device is a source of error (Genty et al. 2017). This misalignment is due to the rough surface, the uneven adhesive layer, and the uneven layer of the coating itself. The misalignment of any and or all of these contribute to an uneven stress distribution across the surface of the testing device leading to an inaccurate bond strength result (Genty et al. 2017). Bond performance between coatings themselves is quite variant. However, this test method can help define the problem existing in the coating. Through different failure types discussed further in the paper, they can be a result of the bond between a coating and a substrate, a result of the adhesion, disbondment of the coating itself, and even to asses coating repairs. This testing is small scale and localized in such a way that it creates a high variability between samples taken from the same area. The interpretation of the results is that much more important if they are to be used to characterize the bond strength between a coating and a substrate (Mata and Atadero 2014). Research Methodology

Manufacturer and Product Selection

For the first part of the study, four manufacturers of polyaspartic polyuria coating systems for concrete participated.

Concrete Surface Preparation ACI and ICRI guidelines were instrumental in defining adequate surface preparation before application of the coating. The surface must be clean, free of dust and debris. The surface moisture content must be below 4% (concrete scale) before the application of the primer. The use of a concrete surface moisture meter (TRAMEX) helped identify this value.

For this study, a needle scaler was the chosen surface preparation method. This removal method required the removal of approximately 1/8" to 1/4" of concrete from the sample's surface exposing the coarse aggregate. The concrete surface profile (CSP) achieved is approximately 6 to 8 CSP (ICRI).

Product Application Note: In laboratory environmental conditions (71℉, 50% RH), the time between each coating layer varied between 45 and 90 minutes. The pot life for each product once mixed varied; however, the products were in a workable state for 30 to 60 minutes. To dispose of unused product and rollers, they were kept in their respective containers until they fully cured (several hours to harden). Uncured products are controlled materials and must follow a disposal process according to their material safety data sheet recommendations and state regulations.

95 Concrete Materials and Sample Preparation For this study, there is a single concrete mixture preparation (0.45 water-to-cement ratio). The concrete mixtures contained a # 57 crushed limestone concrete aggregate and natural sand for the fine aggregate proportion. The mix contained a type-I cement manufactured in Oklahoma. The chemical composition of the cement is in Table 1. The mixtures contain an air-entraining admixture. Mixture proportions are in Table 2.

Table 1: Chemical composition of Portland cement Chemical composition (% by weight) MgO CaO SO3 SiO2 Al2O3 Fe2O3 1.9 62.9 3.3 19.4 5.1 3.4

Table 2: Mixture design details Coarse Water Cement Fine Agg. Air Ent. Mixture w/cm Agg. Paste (%) (kg/m3) (kg/m3) (kg/m3) (oz) (kg/m3) 1 0.45 163.2 362.5 1088.7 709.0 16.0 29.7

Materials were batched and mixed in a temperature-controlled environment and samples were cast respecting standard methods of preparing concrete samples in a laboratory environment (ASTM C 192). For this testing regimen, approximately 33 cylinders (Ø100 mm x 200 mm), 48 concrete blocks (6 in x 6 in x 6in) and 109 prisms (3 in x 4 in x 12 in) were prepared and demolded after 24 hours. After demolding, the samples moved to a moist curing room maintained within ASTM limits for 28 days.

After curing, the samples were placed in a controlled dry room (73 ° F, 50% RH) to allow internal moisture to evaporate and achieve field conditions. Meanwhile, the preparation of the surfaces is as described previously to obtain the required surface texture and physical characteristics determined before product application.

Pull-off testing

Pull-off testing as per ASTM D7234 assessed the bond performance of the products to the surface of the concrete. The samples consisted of 6”x6”x6” blocks for that purpose. Two test surfaces evaluated are a “dry” and a “wet" surface. Twenty-four hours before the application of the products, the blocks were either kept in a dry environment (50% RH) or placed in a wet environment (immersed in tap water) to simulate different environmental conditions encountered in the field and the influence of moisture intake. Next, the "dry" and "wet" samples were removed from their environment and allowed to dry until they reached a surface reading of 3% to 4% moisture content (concrete scale) measured with a (TRAMEX) moisture meter.

In addition to having samples made and tested this way, slab samples dimensioned 12”x9”4.5” were tested using the same process after being exposed to chloride ion penetration by ponding per ASTM C1543. These casting of these slabs were the same as the cubes.

Another influence is the layer thickness of the coating material on its bond strength. The evaluation of three thicknesses consisted of applying 1, 2 and three coats of the product for the cube samples. For each specimen type, each block consisted of two test replicates (Figure 1).

Figure 1: Example of Pull-off Testing

Results and Discussion

The pull-off test (ASTM D7234) was performed to assess the bond performance of each coating system. Here, the influence of layer thickness and moisture of the concrete at the time of application on bond performance were the main variables. The presence of moisture within the substrate may affect the bond performance. The types of coatings evaluated generally do not exhibit disbonding and are suitable systems to use because of their resistance to moisture. However, they can experience poor substrate adherence if exposed to water and moisture in the air during the coating process. (Tator 2015) Primer coats are essential to ensure a higher bond strength and reduce the risk for disbondment. (Ha 2013) Still, post-application moisture transport and vapor transport may also affect bond performance by the creation of surface blisters. (Zhang 2012) (Ha 2013) The experimental regimen devised evaluated these principles. The average results and corresponding coefficients of variation obtained for each specimen type are in the following tables along with a bar type graph demonstrating two standard deviations (2s) from the mean to aid in the comparative analysis.

Failure types

Bond strength of a particular coating can be variable. The failure type can indicate the bond. For this testing regimen, there were four different failure modes.

 The first failure type is between the adhesive and the coating. This failure type is due to improper curing of the adhesive.  The second type is the failure of just the coating.  The third is the failure of the concrete substrate.  The last failure type is a combination of failure due to the coating and the substrate.

(a) (b)

Figure 2: failure types consisted of: (a) adhesive failure, (b) coating failure, (c) substrate failure, and (d) a combination of coating and substrate failure

Coating A

The results for this coating product are in Figure 3. It would seem that there is a slight gain in performance for the coatings applied on dry samples in comparison to that of the wet samples. However, the high variability in results obtained for the wet replicates cannot validate this statement. For dry and wet replicates with a three layers coating system, the dry replicates both exhibited partial failure within the concrete; however, one of the wet replicates exhibited failure in the concrete while the other failed at the bond interface. The latter resulted in a low recorded average and a high coefficient of variation. Similar fracture patterns also occurred for the 1- and 2-layer coating systems which contributed to the high variability in measurements. The observed disintegration of the coating may cause this with time combined with the influence of moisture transport post-application.

98 800 700 600 500 400 300 200 100

Bond Strength (psi) 0 -100 -200 -300

Dry 1 Dry 2 Dry 3 Wet 1 Wet 2 Wet 3

Figure 3: Coating A: Pull-off test result comparison between dry and wet samples and coating thicknesses.

Coating B

Overall, bond testing for the product was not successful. It would seem that the coating disbonded with curing time which resulted in coating bond failures for all of the samples. There are no distinguishable trends for the effects of moisture nor layer thickness on bond strength (Figure 4).

500

400

300

200

100

Bond Strength (psi) 0

-100

-200

Dry 1 Dry 2 Dry 3 Wet 1 Wet 2 Wet 3

Figure 4: Coating B: Pull-off test result comparison between dry and wet samples and coating thicknesses.

Coating C

First, a few noteworthy observations about product performance during the experimental investigation. Thickness readings were not achievable for this product due to the presence of excessive bubbles entrapped in the coating.

99 These bubbles appeared during the curing of the coating and not at the time of application. The presence of such voids could diminish the effective performance of the coating; however, this was not observable while performing the bond-test. On average, the bond-strength for this product seemed adequate as the majority of the failures occurred in the concrete material. Here, consistency in fracture types resulted in a lower recorded variability in comparison to that of the other two products discussed previously.

Here, coefficients of variation vary between 2.5% and 33.1% (average of 12.1%), which might be due to the scaled uneven profile of cube surfaces which may have caused loading eccentricities. Moreover, the scaling process to remove the concrete layer may have caused micro-fissures at the surface of the concrete which weakens the bond interface as well. In general, fracture occurred within the concrete material and not at the bond interface, demonstrating that the bond strength is superior to the tensile strength of the concrete. Similarities in measurements are due to this principle; they reflect the tensile strength properties of the concrete material at its surface.

In Figure 5, there are no observables trends for values obtained for both the dry and wet samples types as well as for the number of coating layers. The results are relatively consistent for all measurements which are attributable to the failure type. Therefore, the actual bond performance between the different sample types is not useful because of the concrete failure in tension.

1000 900 800 700 600 500 400

Bond Strength (psi) 300 200 100 0

Dry 1 Dry 2 Dry 3 Wet 1 Wet 2 Wet 3

Figure 5: Coating C: Pull-off test result comparison between dry and wet samples and coating thicknesses.

Coating D

The last product evaluated is Coating D. This product did not demonstrate any surface features of concern. Moreover, it performed well for the pull-off testing regimens. Similarly, to Coating C, there are no significant differences in performance between the dry and wet sample types. Also, there are no noticeable trends between pull-load and layer thickness. Again, the low coefficients in variation calculated are due to the failure type being in the concrete layer. They vary between 2.9% and 46.8% (average of 14.3%).

The pull-load values obtained for this product are among the highest recorded for this study. These differences are not significant due to the inherent variability of the test method and fracture type.

100 1000 900 800 700 600 500 400 300 Bond Stgrength (psi) 200 100 0

Dry 1 Dry 2 Dry 3 Wet 1 Wet 2 Wet 3

Figure 6: Coating D: Pull-off test result comparison between dry and wet samples and coating thicknesses.

Comparison of Salt Scaled Coatings

In addition to the bond test (ASTM 7234) on cube samples, slab samples after having undergone a salt ponding test (ASTM C1543) underwent the bond test as well. These samples have a one coat layer system as previously described in the test method. The results of this pull off test are shown in Figure 8. Here we have a high variability for all coating types. Coating B had the least variability, shown in the one layer dry coating tested in Figure 4. However, the variability of this coating increases as the number of coating increased as well as wet condition applications. The chart below demonstrates that as a polyuria polyaspartic coating is affected by chloride ion penetration, the uneven distribution of salt concentrations can lead to an increase in the variability of bond test measurements.

900 800 700 600 500 400 300 Bond Strength (psi) 200 100 0 A B C D Coating Materials

Figure 8: Pull-off test result comparison between salt scaled samples with one coat

101 Conclusion

The pull-off test (ASTM 7234) is a test method that has high variability in results. However, this variability can be used to characterize the bond between a coating and a concrete substrate. The coatings experiencing flaking or disbondment (A and B respectively) experienced a higher variability between samples as the coating itself diminished or disbonded from the substrate. In addition to this, bond strength can be used to determine if the concrete was in a wet or dry condition before a coating was applied. This variation reduces the bond strength as well.

In conclusion, even though this test method has a high variability between samples, it can be used to help understand the degradation mechanisms of the coating to the concrete bond. This variability helps to point out specific issues regarding the bond like wet application, disbondment from the substrate, and deterioration of the coating itself. References

1. ASTM Standard C192, 2016a, "Standard Practice for Making and Curing Concrete Test Specimens in the Laboratory," ASTM International, West Conshohocken, PA, 2018, DOI: 10.1520/C0192_C0192M-16A, www.astm.org 2. ASTM Standard C511, 2013, "Standard Specification for Mixing Rooms, Moist Cabinets, Moist Rooms, and Water Storage Tanks Used in the Testing of Hydraulic Cements and Concretes," ASTM International, West Conshohocken, PA, 2018, DOI: 10.1520/C0511-13, www.astm.org 3. ASTM Standard C1543, 2010a, "Standard Test Method for Determining the Penetration of Chloride Ion into Concrete by Ponding," ASTM International, West Conshohocken, PA, 2018, DOI: 10.1520/C1543- 10a, www.astm.org 4. ASTM Standard D7234, 2012, "Standard Test Method for Pull-Off Adhesion Strength of Coatings on Concrete Using Portable Pull-Off Adhesion Testers," ASTM International, West Conshohocken, PA, 2018, DOI: 10.1520/D7234-12, www.astm.org 5. Ha, S. (2013). Bond characteristics of sprayed FRP composites bonded to concrete substrate considering various concrete surface conditions. Composite Structures, 100, 270-9. Retrieved from https://www- sciencedirect-com.argo.library.okstate.edu/science/article/pii/S0263822313000147?via%3Dihub 6. Hughes, Mark E., and Carl R. Bischof. Concrete Repair Manual. 4th ed., vol. 1, American Concrete Institute, 2013. 7. Hughes, Mark E., and Carl R. Bischof. Concrete Repair Manual. 4th ed., vol. 2, American Concrete Institute, 2013. 8. Petrie, Edward M., (2011). Metal Finishing, Volume 109, Issue 6 - Osmotic Blisters in Coatings and Adhesives. Pages 28-30, ISSN 0026-0576. Retrieved from http://www.sciencedirect.com/science/article/pii/S0026057613700217 9. S. Mondal & J. L. Hu (2006) Segmented shape memory polyurethane and its water vapor transport properties, Designed Monomers and Polymers, 9:6, 527-550, DOI: 10.1163/156855506778944028. Retrieved from https://doi.org/10.1163/156855506778944028 10. Tator, Kenneth B.. (2015). ASM Handbook, Volume 05B - Protective Organic Coatings - 7.2 Epoxy Resins. ASM International. Retrieved from https://app.knovel.com/hotlink/pdf/id:kt00UPY7M1/asm-handbook-volume-05b/epoxy-resins 11. Tator, Kenneth B.. (2015). ASM Handbook, Volume 05B - Protective Organic Coatings - 12.2 Polyurethane Chemistry Basics. ASM International. Retrieved from https://app.knovel.com/hotlink/pdf/id:kt00UPYBF2/asm-handbook-volume-05b/polyurethane-chemistry 12. Tator, Kenneth B.. (2015). ASM Handbook, Volume 05B - Protective Organic Coatings - 13.2 A Brief History of Polyurea Development. ASM International. Retrieved from https://app.knovel.com/hotlink/pdf/id:kt00UPYBV1/asm-handbook-volume-05b/brief-history-polyurea

102 13. Tator, Kenneth B.. (2015). ASM Handbook, Volume 05B - Protective Organic Coatings - 14.2 Polyaspartic Esters and Their Chemistry. ASM International. Retrieved from https://app.knovel.com/hotlink/pdf/id:kt00UPYC72/asm-handbook-volume-05b/polyaspartic-esters-their 14. Tator, Kenneth B.. (2015). ASM Handbook, Volume 05B - Protective Organic Coatings - 41.1 Variability within a Properly Applied Coating Layer. ASM International. Retrieved from https://app.knovel.com/hotlink/pdf/id:kt00UPYZD4/asm-handbook-volume-05b/variability-within-properly 15. Tator, Kenneth B.. (2015). ASM Handbook, Volume 05B - Protective Organic Coatings - 44.2 Adhesion Failure. ASM International. Retrieved from https://app.knovel.com/hotlink/pdf/id:kt00UPZ0S1/asm-handbook-volume-05b/adhesion-failure 16. Zhang. (n.d.). Interior Relative Humidity of Normal- and High-Strength Concrete at Early Age. Journal of Materials in Civil Engineering,24(6), 615-622. Retrieved from https://ascelibrary- org.argo.library.okstate.edu/doi/abs/10.1061/(ASCE)MT.1943-5533.0000441 17. Bentz, Dale P., Igor De la Varga, Jose F. Muñoz, Robert P. Spragg, Benjamin A. Graybeal, Daniel S. Hussey, David L. Jacobson, Scott Z. Jones, and Jacob M. LaManna. 2018. “Influence of Substrate Moisture State and Roughness on Interface Microstructure and Bond Strength: Slant Shear vs. Pull-off Testing.” Cement and Concrete Composites 87 (March): 63–72. https://doi.org/10.1016/j.cemconcomp.2017.12.005. 18. Genty, Sébastien, Jean-Baptiste Sauvage, Philippe Tingaut, and Maëlenn Aufray. 2017. “Experimental and Statistical Study of Three Adherence Tests for an Epoxy-Amine/Aluminum Alloy System: Pull-Off, Single Lap Joint and Three-Point Bending Tests.” International Journal of Adhesion and Adhesives 79 (December): 50–58. https://doi.org/10.1016/j.ijadhadh.2017.09.004. 19. Ghavidel, Reza, Rahmat Madandoust, and Malek Mohammad Ranjbar. 2015. “Reliability of Pull-off Test for Steel Fiber Reinforced Self-Compacting Concrete.” Measurement 73 (September): 628–39. https://doi.org/10.1016/j.measurement.2015.06.013. 20. Mata, Oscar R., and Rebecca A. Atadero. 2014. “Evaluation of Pull-Off Tests as a FRP–Concrete Bond Testing Method in the Laboratory and Field.” Practice Periodical on Structural Design and Construction 19 (2): 04014001. https://doi.org/10.1061/(ASCE)SC.1943-5576.0000170. 21. Mitchell, M. R., R. E. Link, Troy Eveslage, John Aidoo, Kent A. Harries, and William Bro. 2010. “Effect of Variations in Practice of ASTM D7522 Standard Pull-Off Test for FRP-Concrete Interfaces.” Journal of Testing and Evaluation 38 (4): 102682. https://doi.org/10.1520/JTE102682. 22. Nepomuceno, Miguel C. S., and Sérgio M. R. Lopes. 2017. “Analysis of Within-Test Variability of Non- Destructive Test Methods to Evaluate Compressive Strength of Normal Vibrated and Self-Compacting Concretes.” IOP Conference Series: Materials Science and Engineering 245 (October): 032025. https://doi.org/10.1088/1757-899X/245/3/032025. 23. Ramos, N.M.M., M.L. Simões, J.M.P.Q. Delgado, and V.P. de Freitas. 2012. “Reliability of the Pull-off Test for in Situ Evaluation of Adhesion Strength.” Construction and Building Materials 31 (June): 86–93. https://doi.org/10.1016/j.conbuildmat.2011.12.097. 24. Sadowski, Łukasz. 2013. “Non-Destructive Evaluation of the Pull-off Adhesion of Concrete Floor Layers Using Rbf Neural Network.” Journal of Civil Engineering and Management 19 (4): 550–60. https://doi.org/10.3846/13923730.2013.790838.

103

TFM Acoustic Region of Influence TFM Acoustic Region of Influence Chi-Hang Kwan, Guillaume Painchaud-April, and Benoit Lepage Chi-Hang Kwan1, Guillaume Painchaud-April1, and Benoit Lepage1 Olympus NDT Canada 1 3415Olympus Rue Pierre-Ardouin NDT Canada 3415Québec Rue QC,Pierre G1P-Ardouin 0B3 (418) 263-9621; emailQuébec [email protected] QC, G1P 0B3 (418) 263-9621; [email protected]

ABSTRACT In this paper, we introduce a newly developed semi-analytical model to predict the TFM amplitude sensitivity map for both non-directional and directional flaws. For complicated acoustic paths which involve multiple interface interactions and wave-mode conversions, a knowledge of the Acoustic Region of Influence (AROI) enables an inspector to refine the scan plan to maximize the signal-to-noise ratio of the resultant TFM image and increase the probability of flaw detection. The accuracy of this new acoustic model was tested and validated by experiments using test blocks that contain side-drilled holes and flat-bottom holes. Results from the validation experiments show good agreement between the empirical TFM amplitude maps and theoretical AROI maps. The results also indicate that the model can be used to guide the selection of the optimal TFM inspection mode.

Keywords: Total Focusing Method, Acoustic Model, Multi-Mode Imaging

INTRODUCTION The Total Focusing Method (TFM) is a synthetic aperture beam forming technique that has been under active development in the NDT industry over the past decade [1]. By applying appropriate transmission and reception delays to A-scan data collected in a Full Matrix Capture (FMC) dataset, TFM can electronically focus on every location within an inspection region. Since every point is electronically focused, TFM can provide better resolution compared to conventional phased array ultrasound inspection techniques. In addition, by calculating and applying the time of flights of multiple acoustic modes, multi-mode TFM that can provide additional information about the specimen being inspected [2].

Despite the advantages listed above, TFM also has limitations governed by physical laws. An inspection area may have poor sensitivity due to the effects of interface interaction, beam forming limitations, and propagation path attenuation. Owing to the novelty of TFM, the lack of inspection codes, and the complexity of multi-mode TFM imaging, inspectors are generally unaware of the physical limits of TFM and, therefore, cannot define an optimal scan plan that maximizes the Signal-to-Noise Ratio (SNR) and the probability of detection. Consequently, there is a need to introduce a tool that estimates the acoustic sensitivity map for a given TFM inspection scan plan.

ACOUSTIC REGION OF INFLUENCE The Acoustic Region of Influence (AROI) is a theoretical acoustic amplitude sensitivity map for a given TFM inspection scan plan. In general, the AROI map differs for directional and non-directional flaw scatterers. NDT examples of non-directional scatterers include slag and porosity in welds, while examples of directional scatterers include lack of fusion in welds and various cracks. The directional scattering response of a flaw is an important parameter that is often neglected in the modeling of phased array transducer systems.

To calculate the AROI, we have developed a semi-analytical, ray-based acoustic model that calculates the two-way pressure response of pulse-echo, self-tandem, and double-skip TFM inspection modes. This acoustic model takes into account the effects of transmission and reflection coefficients, geometric beam spread, and material attenuation.

In addition, in our model we also used the Rayleigh-Sommerfeld integral [3] to model the far-field scattering response for a Flat Bottom Hole (FBH). The FBH scattering response is used to simulate directional flaws.

VALIDATION EXPERIMENTS To examine the acoustic model’s accuracy, we conducted validation experiments to compare experimentally obtained TFM amplitude maps with theoretically calculated TFM AROI maps. Results obtained from two validation experiments are presented in this section. The first validation experiment was conducted on a test block that contains small-diameter Side-Drilled Holes (SDH), which simulate the scattering response of nondirectional flaw scatterers. The second validation experiment was conducted on a test block that contains FBHs, which simulate the scattering response of directional flaws.

For the results presented in this paper, the x-axis is defined positive to the right of the first transducer element, and the z-axis is defined positive below the surface of the test sample. A schematic diagram of this coordinate system is shown in Figure 1.

Figure 1: Coordinate system used in this paper.

Side-Drilled Hole Validation The SDH validation experiment was conducted on a NAVSHIPS metric 1018 steel test block that contains six 1.2 mm diameter SDHs at depths from 6.25 mm to 37.5 mm in 6.25 mm increments. By turning over the test block, it is possible to examine SDHs at depths from 6.25 mm to 68.75 mm. For this experiment, we used a 32-element 5L32- A31 probe with a center frequency of 5 MHz and an element pitch of 0.6 mm. The probe was coupled to a 36.1º SA31-N55S-IHC Rexolite wedge. A schematic drawing of the experimental setup is shown in Figure 2.

Figure 2: Schematic diagram of the SDH validation experiment. Note that only the top scan orientation is shown.

105

By translating the probe along the surface of the test block, we obtained scattering echoes from the SDHs at different positions relative to the probe. FMC datasets were gathered at each scan position for post-processing to generate empirical TFM amplitude maps. A description of the post-processing algorithm is described in the following subsection.

Generating the Empirical TFM Amplitude Map The main steps for generating the empirical TFM amplitude map are: 1. For a given flaw at a fixed scan position, use a depth (z-direction) gate to obtain an amplitude line along the width of the Amplitude map. 2. Repeat step 1 for different scan positions to obtain a composite amplitude line for a given flaw. 3. Repeat steps 1 and 2 for all other flaws to obtain composite amplitude lines at different z-positions. 4. Interpolate the composite amplitude lines in the z-direction to obtain the final TFM amplitude map.

Step 1 is illustrated in Figure 3. Figure 3 shows that we first form a TFM strip along the width of the amplitude map at depths specified by the z-gate. The position of the z-gate is chosen based on the known depth of the flaw. At every x-position along the TFM strip, the maximum amplitude is taken along the z-direction to obtain the amplitude line shown at the bottom of Figure 3.

z-gate

Figure 3: Procedure to obtain an amplitude line for a given flaw at one scan position.

To form the composite amplitude line for a given flaw, we compare all amplitude lines obtained at different scan positions and record the maximum amplitude values. This procedure is illustrated in Figure 4.

106

Figure 4: The procedure to form composite amplitude lines at different scan positions.

After forming the composite amplitude line for a given flaw, the process is repeated for all flaws at different depths. Figure 5 shows the pulse-echo TT mode composite amplitude lines obtained for the SDHs present in the NAVSHIPS test block (both top and bottom orientations). In Figure 5 and all other experimental TFM-derived figures presented in this paper, the amplitudes of the TFM images are not normalized. Since a 12-bit digitizer is used in the acquisition electronics and the probe contains 32 elements, the theoretical maximum amplitude in the TFM image is 2097152 (212 ÷ 2 × 32 × 32).

Note that the composite amplitude lines for SDHs at depths of 6.25, 62.5, and 68.75 mm were not included in Figure 5. Due to the proximity of these SDHs to the lateral limits of the test block, it was not possible to obtain complete composite amplitude lines along the entire width of the amplitude map.

Figure 5: The composite amplitude lines of SDHs present in the NAVSHIPS test block.

Comparing Empirical TFM Amplitude Maps with AROI Maps By performing z-direction interpolation on the composite amplitude lines shown in Figure 5, we obtained the empirical TFM amplitude map shown in Figure 6 (a). Figure 6 (a) shows that this TFM scan plan has poor sensitivity both at low (<30º) and high (>70º) steering angles. The poor sensitivity at low steering angles is caused by small transmission coefficient values from the Rexolite wedge to the steel test block [4]. In contrast, the poor sensitivity at high steering angles is due to poor focusing caused by large effective F-numbers [5]. These findings are consistent with the recommended steering angle guidelines for conventional phased array angled inspections [6].

107

(a) (b)

Figure 6: (a) Empirical SDH amplitude map and (b) theoretical SDH AROI map for pulse-echo TT mode. 30º and 70º steering angle guidelines (from the midpoint of the active aperture) were added.

The corresponding theoretical SDH AROI map is shown in Figure 6 (b). Comparing the two figures, it is evident that the acoustic model can accurately predict the area within the scan plan that has the optimal sensitivity. The discrepancies between the two figures can be attributed to small variations in coupling pressure as the probe is translated along the test block’s surface. Note that the amplitude of the theoretical AROI map is in arbitrary units because it is extremely difficult to model the exact magnitude of the received voltage signals from the acquisition system. However, since consistent arbitrary units are used for different AROI maps, it is still possible to compare the TFM acoustic sensitivities of different scan plans and different acoustic modes.

Flat-Bottom Hole Validation To test the accuracy of the acoustic model for predicting the amplitude sensitivity for directional flaws, we conducted validation experiments on a custom-machined test block. The test block has a thickness of 20 mm and contains FBHs that were drilled to match the profile of a typical J-bevel weld. For this study, we use the 5 FBHs whose bottom surface normal vectors are oriented 3º below horizontal. A photograph of the test block with indications of the scan axes is shown in Figure 7.

Figure 7: Custom-machined FBH test block showing the scan axes.

108

For this experiment, we used a 32-element 5L32-A32 probe with a center frequency of 5 MHz and an element pitch of 1 mm. The probe was coupled to a 36.1º SA32-N55S-IHC Rexolite wedge. Since the orientations of the bottom surfaces of the FBHs are near vertical, the acquired FMC datasets were processed in self-tandem (single skip) modes. A schematic diagram of the scan plan is shown in Figure 8.

Figure 8: A schematic diagram of the validation experiment showing the self-tandem TFM mode.

Comparing Empirical TFM Amplitude Maps with AROI Maps The empirical FBH amplitude map and the theoretical FBH AROI map for self-tandem TTT mode are respectively shown in Figure 9 (a) and (b). Comparing the two plots, it is evident that the acoustic model provided an accurate estimate of the relative acoustic sensitivity within the scan region. Figure 9 suggests that self-tandem TTT is better suited for detecting vertical flaws located near the bottom of the test sample.

(a)

(b)

Figure 9: (a) Empirical FBH amplitude map and (b) theoretical FBH AROI map for self-tandem TTT mode.

The empirical FBH amplitude map and the theoretical FBH AROI map for self-tandem TLT mode are respectively shown in Figure 10 (a) and (b). Once again, it is evident that the acoustic model provided an accurate estimate of the relative acoustic sensitivity within the scan region. The oscillations of the empirical amplitude map from z = 25 mm to z = 40 mm are caused by interference from other acoustic modes that have similar travel times.

109

In addition, comparing Figure 9 with Figure 10, we see that the ratios of maximal amplitudes between the two self- tandem modes are approximately 3.3 (13800/4200) for the empirical amplitude maps and approximately 3.4 (1.23/0.36) for the theoretical AROI maps. The similarity of amplitude ratios suggests that the acoustic model can also be used to predict the relative acoustic sensitivity across different TFM imaging modes.

(a)

(b)

Figure 10: (a) Empirical FBH amplitude map and (b) theoretical FBH AROI map for self-tandem TLT mode.

EXAMPLE APPLICATION To further demonstrate the utility of the acoustic model, we present an example of a real-world application where the theoretical AROI map is used to guide our selection of the TFM inspection mode. For this example, we inspected a V-bevel weld sample with a known lack of fusion defect. The weld angle is approximately 35º, and we used the same 5L32-A32 probe and SA32-N55S-IHC wedge that were used for the FBH validation experiment. A schematic diagram of the experimental setup is shown in Figure 11.

Figure 11: Schematic diagram for lack of fusion inspection.

In the theoretical model, the lack of fusion defect is simulated by a 5 mm diameter FBH defect that has a bottom surface oriented at 35º away from the vertical. The corresponding theoretical AROI maps for self-tandem TLT mode and double-skip TTTT mode are shown in Figure 12.

110

(a) (b)

Figure 12: Theoretical AROI maps for a lack of fusion inspection plan in (a) self-tandem TLT mode and (b) double-skip TTTT mode.

Figure 12 shows that the TLT mode AROI map is more irregular compared to the double-skip mode AROI map. Consequently, it would be more difficult to obtain a robust assessment of the size of the lack of fusion defect using TLT mode. In addition, the expected amplitude from the TLT mode is 3 orders of magnitude lower than the double- skip mode. Using these theoretical AROI maps, we predict that the double-skip TTTT mode is the preferred TFM imaging mode. The corresponding experimental TFM images are displayed in Figure 13.

(a) (b)

Figure 13: TFM images of a lack of fusion defect in (a) self-tandem TLT mode and (b) double-skip TTTT mode.

Figure 13 demonstrates that the double-skip TFM image has good a good SNR and provides a clear assessment of the size of the lack of fusion defect. In contrast, the self-tandem TFM image has a poor SNR and contains isolated echoes that are difficult to interpret. The isolated echoes are likely the diffracted echoes from the sharp tips of the lack of fusion defect. Nevertheless, the dimension and type of defect are difficult to assess in self-tandem TLT mode.

The poor SNR of the self-tandem TLT mode TFM image corroborates with the low amplitude shown in the theoretical AROI map in Figure 12 (a). However, it should be noted that the ratio of echo amplitudes for the two modes in Figure 13 is lower than the amplitude ratio predicted by the theoretical AROI maps in Figure 12. Since the geometry of the lack of fusion defect is different from the FBH model used to simulate the flaw, the amplitudes of the diffracted echoes from the sharp tips of the lack of fusion defect might be underestimated in the theoretical model.

111

CONCLUSIONS We have demonstrated an acoustic model that can accurately predict the TFM amplitude map for both nondirectional and directional defects. For a given inspection mode, the model can be used to adjust the scan plan (aperture, scan frequency, probe location, etc...) to optimize the SNR and the probability of detection. Since the model provides a comparison of relative amplitude across different acoustic modes, it can also be used to select the optimal TFM reconstruction mode. In the future, we plan to extend the model to more complex geometries and include more flaw scattering models to increase the utility of the model.

REFERENCES [1] C. Holmes, B. W. Drinkwater, and P. D. Wilcox, “Post-processing of the full matrix of ultrasonic transmit– receive array data for non-destructive evaluation,” NDT E Int., vol. 38, no. 8, pp. 701–711, Dec. 2005. [2] K. Sy, P. Bredif, E. Iakovleva, O. Roy, and D. Lesselier, “Development of methods for the analysis of multi- mode TFM images,” J. Phys. Conf. Ser., vol. 1017, p. 012005, May 2018. [3] L. W. S. Jr, Fundamentals of Ultrasonic Nondestructive Evaluation: A Modeling Approach, 2nd ed. Springer International Publishing, 2016. [4] Foundations of Biomedical Ultrasound. Oxford, New York: Oxford University Press, 2006. [5] S. I. Nikolov, J. Kortbek, and J. A. Jensen, “Practical applications of synthetic aperture imaging,” in 2010 IEEE International Ultrasonics Symposium, San Diego, CA, 2010, pp. 350–358. [6] E. A. Ginzel and D. Johnson, “Phased-Array Resolution Assessment Techniques,” p. 13.

112

Automated Non-Destructive Evaluation of Spot Welds Automatedusing the Imagingnon-destructive Analyses evaluation of the Residual of spot Magnetic welds using Flux the Density imaging analyses of the residual magnetic flux density Christian Mathiszik, Jörg Zschetzsche, and Uwe Füssel Christian Mathiszik, Jörg Zschetzsche, Uwe Füssel Technische Universität Dresden Chair Technischeof Joining Technology Universität andDresden Assembly Chair of JoiningDresden, Technology Germany and Assembly +49 351 463 34346; emailDresden, [email protected] Germany +49 351 463 34346; [email protected]

ABSTRACT Resistance spot welding is due to its high reliability and economy one of the most widely used welding methods in the automotive and railway industries. The high quality assurance requirements in these areas call for reliable nondestructive testing (NDT) methods of the spot welds. The nugget is localized between the sheet metals and therefore not directly measurable from outside. This aspect continues to pose a major challenge for non-destructive quality assurance, especially due to the increasing diversity of used steel alloys in car body manufacturing. At the Technische Universität Dresden, Germany, the imaging analyses of the residual magnetic flux density has been developed for NDT of spot welds. Up until now, the manual evaluation of the measurement results is time- consuming and subjective. In order to achieve greater reliability and to minimize evaluation times, an algorithm for automated evaluation has been developed. In order to evaluate the quality of the algorithm, the results are compared with NDT and destructive measurements of the same samples. For these samples, typical automotive steel combinations for spot welding were used. This talk will show the measuring concept of the imaging analyses of the residual magnetic flux density for NDT of spot welds and presents the high potential of using an automated non-destructive evaluation algorithm for NDT of spot welds.

Keywords: spot welding, RSW, NDT of spot welds, magnetic testing, imaging analysis

INTRODUCTION AND OBJECTIVE Resistance welding processes are established joining methods used widely in industrial manufacturing. In particular, resistance spot welding is used in power and rail vehicle construction, in metal construction and in equipment technology. The processes are characterized by a very high efficiency. The current flow path is determined essentially by the electrode shape [1]. Commonly two or three sheets are welded together. Ideally, a lenticular welded joint, called nugget, is created between the sheets as illustrated in Figure 1. Thus, the weld joint is not visible from the outside. Hence, the nugget diameter , which is the most important quality criteria, cannot be measured directly. This still poses a major challenge for non-destructive testing (NDT) of spot welds. Current test methods are 𝑑𝑑𝑛𝑛 sometimes inadequate. The technical bulletin DVS 2916-5 [2] describes the individual NDT methods for resistance spot welding. Ultrasonic testing, using the pulse-echo technology, has prevailed so far in industrial mass production. Critical influencing factors for this test method include a rough surface topography, extreme material-sheet- thickness combinations combined with high deviations in repetitive measurements. The imaging analysis of the residual magnetic flux density shows high potential to extend the current application limits for NDT of spot welds [3]. However, the evaluation of the measurements is currently still subjective and time-consuming. The main objectives of this study is to increase productivity and, as a result, the cost-effectiveness of resistance spot welding overall. In order to achieve an increase in productivity, both the test time and the subjectivity of the test procedure for quality assurance are reduced.

Figure 1: Illustration of a spot weld in cross-sectional view

EXPERIMENTAL DESIGN Methodology to develop the algorithm for an automated evaluation The methodology to improve the algorithm for an automated non-destructive evaluation is shown in Figure 2. The procedure is divided into two phases. In the first phase, spot welds are welded and divided into training and test sets. The variable parameters of the algorithm are optimized based on the training sets. The algorithm is validated by the test set for functionality without further adaptation of the variable parameters. The amount of discrepancies between the results of the training and test sets provide information about the current quality of the developed algorithm. This procedure is run through several times to optimize the algorithm. After training, the algorithm is applied to an third independent set of samples in the second phase. Here again, the quality of the algorithm is checked and validated.

Figure 2: Methodology for improving the algorithm for an automated non-destructive evaluation of the nugget diameter

Material combinations and welding parameters Four material combinations (MC) were selected for the investigations. MC2, MC3 and MC4 pose great challenges for the common resistance spot welding process. These combinations are characterized by at least one of the following properties:  sheet thickness ratio > 3/1 (MC2, MC4)  distinctly different specific electrical resistances (MC3)  significantly different material strength (MC2, MC3, MC4) The larger the sheet thickness ratio, the more difficult it is to weld the thinner sheet. The different material resistances lead to the fact that the nugget starts to grow in the thicker sheet metal and not in the JP as usual. Referring to ISO 14373 [4], the maximum sheet metal ratio is limited to 3/1 for common resistance spot welding processes. Distinctly different specific electrical resistances lead to the same problem. The nugget has its origin in the sheet metal with the higher specific electrical resistance, and grows into the JP of both sheets. If the process parameters are set unfavorably or the process stability is very low, e.g. caused by electrode wear, the sheet with the lower electrical resistance will not be properly welded. Significantly different material strengths lead to uneven electrode indentations and may cause deterioration of the overall stiffness due to tapered sheet thicknesses.

114

Table 1: Material combinations used for the investigations MC MC1 MC2 MC3 MC4 upper sheet HX340LAD+Z, 1 mm DC04+ZE, 0.65 mm DC04+ZE, 0.65 mm DC04+ZE, 0.65 mm middle sheet 22MnB5+AS, 2 mm HX340LAD+Z, 3 mm lower sheet HX340LAD+Z, 1 mm HX340LAD+Z, 3 mm 22MnB5+AS, 2 mm HX340LAD+Z, 2 mm

The welding current is adjusted in six steps for each material combination to generate different qualities of the spot welds. The electrode force and welding time are constant for each spot weld within one material 𝐼𝐼𝑤𝑤 combination. As a result, insufficiently spot welds, spot welds with spatters and spot welds with appropriate quality 𝐹𝐹𝑒𝑒 𝑡𝑡𝑤𝑤 are generated. The assignment of the suitable welding current intervals of the respective material combination and the derived welding parameters are shown in Figure 3. This method ensures that the algorithm to be developed will differentiate between an appropriate quality and a non-appropriate quality of spot welds.

Figure 3: Assignment of the suitable welding current intervals of the respective material combination (MC) and joint plane (JP) and the derived welding parameters

NDT-Method and methods for evaluation of the NDT results The imaging analysis of the residual magnetic flux density consists of two steps. Figure 4 shows both steps. In the first step, the sample is located with its spot weld coaxial between two coils oriented in the same direction and is magnetized. The magnetic field strength is adjusted so that the material reaches its magnetic saturation. The residual magnetic flux density is measured in the second step flat on the surface of the sample using hall sensors. During the investigations, an elastic and flexible hall sensor attachment was developed besides a rigid one. The flexible 𝐵𝐵𝑟𝑟 version showed distinct advantages in comparison with the rigid version in [5]. It was used for the measurements of the residual magnetic flux density after its testing. This applies in particular to the measurements of the lower sides of the samples of all investigated material combinations. The result of the measurement can be analyzed using intensity diagrams and data processing algorithms described below (Figure 4).

Figure 4: Test procedure of the imaging analysis of the residual magnetic flux density for spot welds

115

Every measurement of the independent set is measured manually by calculating the second derivative of the magnetic flux density visualized in intensity plots. The diameters are determined according to ISO 14329 [6], wherein the average diameter is calculated from the largest diameter and the one perpendicular to it (Figure 5, right side). The spot weld diameters are measured for each joint plane separately. The upper side measurements always correspond to the diameter evaluation of the JP1. This applies also to the lower side measurements for two-sheet combinations. For three-sheet combinations, the lower measurements refer to the spot weld or nugget diameter of JP2 (compare to Figure 1). The torsion test is used to evaluate the NDT results. This test method has the advantage over the chisel test that the destroyed spot welds can be measured very well according to ISO 14329 [6] and Figure 5, right side. The self- developed torsion test stand of the Chair of Joining Technology and Assembly of the Technische Universität Dresden was used (Figure 5). The construction of the test stand guarantees an automatic alignment of the rotation axes of the drive shaft, nugget and the lower sample clamping. The automated destruction of spot welds is done with a constant and controlled angular velocity and a continuous recording of torsional moment and angle . The samples are destroyed at an angular velocity ω = 100°/min. Additionally, cross-sections of the spot welds are 𝑀𝑀𝑇𝑇 𝜑𝜑 analyzed. The plane for the cross-section was chosen perpendicular to the orientation of the welding gun, as Figure 14 illustrates.

Figure 5: Torsion test stand of the Chair of Joining Technology and Assembly of the Technische Universität Dresden and the measuring concept for determining the spot weld diameter manually [6]

𝒅𝒅𝒑𝒑

Figure 6: Illustration of the cutting plane for the cross-section samples

DEVELOPMENT OF THE AUTOMATED EVALUATION ALGORITHM Theoretical approach The size of the nugget influences the magnetization process. This can be proved by using the finite element method (FEM) to simulate the magnetization process. Figure 7 shows the results of such FEM simulations done with ANSYS® Academic Research Mechanical, Release 18.1. The simulation demonstrates the magnetization process by

116

comparing a small nugget with a big one. It shows that the size of the nugget influences the magnetization process. This effects the amount of the resulting residual magnetic flux density of the spot weld, which is measured after magnetization. As a result, the amount of measured flux density in the welded and not welded area is significantly different. At the edge of the nugget, the amount increases sharply. The first derivative of the magnetic flux density shows the maxima of this increase. Where the second derivative disappears is the edge of the weld (Equation 1). This encloses an area, which is assumed to be the welded area as illustrated in Figure 8. This theoretical approach is implemented in the software for the evaluation algorithm to enable an automatic evaluation of the measurement results of the imaging analysis of the residual magnetic flux density. Thus, the ratings of the welds are objective and independent of the user.

Figure 7: FEM-simulation of the magnetization process comparing a small with a big nugget

(Eq. 1) 2 2 = 0 ∂ 𝐵𝐵𝑧𝑧 ∂ 𝐵𝐵𝑧𝑧 2 2 Δ𝐵𝐵𝑧𝑧 = ∂x + ∂y where is the z-component of the magnetic flux density is direction in x 𝐵𝐵𝑧𝑧 in direction in y 𝑥𝑥

𝑦𝑦

Figure 8: Measured flux density as raw data (left), second derivative (middle) and an overlay of second derivative and destructive test (DT) true to scale (right)

Implementation of the theoretical approach into the algorithm The implementation of the theoretical approach showed, that a reference plane as threshold is necessary. This threshold is compared with a mean value of the second derivative to distinguish between spot welds with inappropriate and appropriate quality. If the mean value is lower than the threshold, the algorithm rates the sample as insufficient. The threshold value of the reference plane depends on the material combination. So far, it has been empirically determined. The reference plane is also used as a threshold to filter out smaller increases in the measured data. If = 0 between the two maxima above the reference plane, the location of the maximum can be deduced, as shown in Figure 9. This results in an annular area, shown as a contour ring Figure 9. The inner area of the contour 𝑧𝑧 corresponds∆𝐵𝐵 to the welded area or the nugget. The results of the implementation confirm the theoretical approach.

117

Figure 9: Second derivative according to Equation 1 with a determined contour ring for the evaluation of the spot weld (left) and the profile with the reference plane to determine the contour ring (right) [7]

RESULTS AND DISCUSSION In the following evaluations, the results from phase one and phase two are discussed together. First of all, the achieved quality of phase one is assessed. Subsequently, the results from phase two are compared with results from the manual evaluations. Boxplots are used to statistical evaluate the quality of the algorithm by comparing the absolute deviations.

MC1 Figure 10 shows a typical not-welded sample in cross-section on the left. The heating was not enough to generate a spot weld. Only the zinc coatings of both sheets were soldered together. Thus, the weld is classified as a solder joint and insufficient. The cross-sections on the right side in Figure 10 shows an appropriate spot weld. The evaluation of the training sets can be estimated very well with a deviation of -0.9% ± 2.9% measured on the upper side and -1.0% ± 6.3% on the lower side of the sheet Figure 11. The determined nugget diameters correlate strongly with the destructively determined spot weld diameters in the entire parameter range. The measurement deviation and the correlation of the test results differ only slightly from the training set. Analyzing the results of phase two, large deviations of the automatically determined diameters by the algorithm appear. When evaluating the results from the upper side, the algorithm determines inappropriate welds incorrect. The reason for this is the different measurement concepts as described above. The upper sides were measured with the rigid hall sensor mount, whereas the lower sides were measured with the flexible attachment of the hall sensor. The incorrect evaluations from the lower side are much less. The absolute deviation and the variation of the manually measured diameters also is lower. This shows, that an automated evaluation has an improvement potential for the algorithm, because the determinations are based on the same measurements.

Figure 10: Cross-section of MC1: left: no spot weld ( = 4.8 kA, = 2.7 kN, = 240 ms), right: appropriate weld quality ( = 8.8 kA, = 2.7 kN, = 240 ms) 𝑰𝑰𝒘𝒘 𝑭𝑭𝒆𝒆 𝒕𝒕𝒘𝒘

𝑰𝑰𝒘𝒘 𝑭𝑭𝒆𝒆 𝒕𝒕𝒘𝒘

as shown in Figure 9. This results in an annular area, shown as a contour ring Figure 9. The inner area of the contour corresponds to the welded area or the nugget. The results of the implementation confirm the theoretical approach.

Figure 10: Cross-section of MC1: left: no spot weld ( = 4.8 kA, = 2.7 kN, = 240 ms), right: appropriate weld quality ( = 8.8 kA, = 2.7 kN, = 240 ms) 𝑰𝑰𝒘𝒘 𝑭𝑭𝒆𝒆 𝒕𝒕𝒘𝒘

𝑰𝑰𝒘𝒘 𝑭𝑭𝒆𝒆 𝒕𝒕𝒘𝒘

118

Figure 11: Comparison between NDT and destructive testing of MC1 and statistical comparison of manually and automatically evaluated spot welds

MC2 A sheet thickness ratio > 3/1 and widely differing strengths of the materials characterize MC2:  DC04: yield strength = 141.1 MPa, tensile strength = 284.8 MPa [8]  HX340LAD: yield strength = 356.5 MPa, tensile strength = 434 MPa [8] 𝑅𝑅𝑝𝑝0.2 𝑅𝑅𝑚𝑚 The challenge welding this material combination is to achieve a proper weld to the upper sheet, without creating a 𝑅𝑅𝑝𝑝0.2 𝑅𝑅𝑚𝑚 too deep electrode indentation. The welding tests show that no zinc coating solder joints were realized. In other words, in the case of the weld joint, it can be observed in the cross-section, that the weld nugget grows only very slightly into the upper sheet at low welding currents (Figure 12, left side). At higher welding currents, the weld nugget grows very far into the upper sheet (Figure 12, right side). It was not possible to achieve soldered joints of only the zinc coatings. Either the sheets were not connected at all after the welding process or a nugget was formed. Figure 13 shows the results of the determined nugget diameters. The measured values of MC2 in phase one spread more measured from the upper side of the sample than those of the lower side. With a measurement deviation of 1.5% ± 2.8%, the algorithm is suitable for an automated evaluation for this material combination. When comparing the results of phase two, the manually and automatically analyzations differ not very much. Only the automatically determined diameters from the upper side show incorrect evaluations, where spot weld with an appropriate quality are assessed as inappropriate. Again, this might be a result of the different hall sensor mounts. The evaluations of the lower side in contrast show only slightly differences and correlate very well in both cases for manually measured and automatically determined diameters.

Figure 12: Cross-section of MC2: left: nugget with a marginal nugget penetration into the upper sheet = 5.6 kA, = 2.0 kN, = 600 ms), right: appropriate weld quality ( = 9.6 kA, = 2.0 kN, = 600 ms) (𝑰𝑰𝒘𝒘 𝑭𝑭𝒆𝒆 𝒕𝒕𝒘𝒘 𝑰𝑰𝒘𝒘 𝑭𝑭𝒆𝒆 𝒕𝒕𝒘𝒘

119

Figure 13: Comparison between NDT and destructive testing of MC2 and statistical comparison of manually and automatically evaluated spot welds

MC3 Strongly differing strengths and specific electrical resistances characterize MC3:  DC04: yield strength = 141.1 MPa, tensile strength = 284.8 MPa [8], specific electrical resistance = 0.13 Ωmm2/m [9] 𝑅𝑅𝑝𝑝0.2 𝑅𝑅𝑚𝑚  22MnB5: yield strength = 300 – 550 MPa, tensile strength = 500 – 700 MPa [10], specific 𝑅𝑅 electrical resistance = 0.214 Ωmm2/m [11]; 0.25 Ωmm2/m [12] 𝑅𝑅𝑝𝑝0.2 𝑅𝑅𝑚𝑚 Figure 14 shows the challenge of reliably welding this material combination. Only with welding currents above 8.8 𝑅𝑅 kA a weld of the upper sheet could be achieved. The nugget penetration depth into the upper sheet is very low. The cross-section in Figure 14, left side, shows that there was a structural change in the upper sheet, but no welded joint has formed in JP1. With increasing welding current, very deep electrode indentations occur on both sides of the samples. Liquid material is squeezed out of the welding zone during the welding process and more welding spatter occur at higher welding currents from 8.8 kA upwards. In phase one, MC3 shows small errors of 1.0 ± 2.6% measured on the upper side of the test set samples. The results of the lower side scatter more. Nevertheless, the training and test sets have similar deviations. The results of the lower sides show good suitability of the algorithm for this material combination. The evaluations of the second phase show the same characteristic as MC2. The upper side measurements scatter more. Again, the algorithm misjudges spot welds with an appropriate quality. However, it detects all inappropriate spot welds, which is not the case in the manual evaluation.

Figure 14: Cross-section of MC3: left: nugget with no fusion to the upper sheet ( = 5.6 kA, = 3.5 kN, = 1000 ms), right: spot weld with deep electrode indentation ( = 8.0 kA, = 3.5 kN, = 1000 ms) 𝑰𝑰𝒘𝒘 𝑭𝑭𝒆𝒆 𝒕𝒕𝒘𝒘 𝑰𝑰𝒘𝒘 𝑭𝑭𝒆𝒆 𝒕𝒕𝒘𝒘

120

Figure 15: Comparison between NDT and destructive testing of MC3 and statistical comparison of manually and automatically evaluated spot welds

MC4 MC4 is characterized by differing strengths and a large sheet thickness ratio > 3/1:  DC04: yield strength = 141.1 MPa, tensile strength = 284.8 MPa [8],  HX340LAD (3 mm): yield strength = 356.5 MPa, tensile strength = 434 MPa [8] 𝑅𝑅𝑝𝑝0.2 𝑅𝑅𝑚𝑚  HX340LAD (2 mm): yield strength = 375.3 MPa, tensile strength = 440.2 MPa [8] 𝑅𝑅𝑝𝑝0.2 𝑅𝑅𝑚𝑚 Again, the cross-sections illustrate the great challenge of reliably welding this material combination. The nugget 𝑅𝑅𝑝𝑝0.2 𝑅𝑅𝑚𝑚 starts to grow in the center of the material combination. The welding of the upper and middle sheets, is to be regarded as very critical. In the cross-sections of Figure 16, a significant heat influence of the sheets of JP1 can be observed. Local welds appear. Contrary to this, the test results of the torsion test do not show any samples with welds between the sheets of JP1. Those destroyed torsion samples were difficult to evaluate. Additional measurements by means of energy dispersive X-ray spectroscopy (EDX) of the fracture surfaces should provide information. However, the comparison of the amount of zinc in the torsion fracture surface on sheet 1 of these critical samples and a well-welded sample in Figure 17 shows no significant differences of distribution of zinc. Due to the slight change in the residual magnetic flux density in the area of the nugget on the surface of the welded samples, it is more difficult for an automated evaluation. Therefore, the measurement deviation is calculated to - 7.0% ± 7.0% in phase one. In contrast, the lower side of this material combination is better suited for automated evaluation. The change in the measured flux density in the area of the nugget leads to good measurement results. The same applies to the evaluation of the results of phase two. The manually determines diameters show smaller absolute deviations. The results of the automatically determined spot welds show better results measured form the lower side, again.

121

Figure 16: Cross-section of MC4: left: nugget with no fusion to the upper sheet ( = 6 kA, = 2.5 kN, = 400 ms), right: spot weld with deep electrode indentation ( = 6.8 kA, = 2.5 kN, = 400 ms) 𝑰𝑰𝒘𝒘 𝑭𝑭𝒆𝒆 𝒕𝒕𝒘𝒘

𝑰𝑰𝒘𝒘 𝑭𝑭𝒆𝒆 𝒕𝒕𝒘𝒘

Figure 17: EDX analysis with representation of the Zn –amount of the torsion fracture surfaces of sheet 1

Figure 18: Comparison between NDT and destructive testing of MC4 and statistical comparison of manually and automatically evaluated spot welds

122

CONCLUSION AND OUTLOOK In this article, the development of an algorithm for automated non-destructive evaluation of spot welds using the imaging analyses of the residual magnetic flux density was presented. With the help of training and test sets, the evaluation algorithm could be developed so far that an automated NDT evaluation of spot welds is possible. This allows objective assessments of resistance spot welds to be made without relying on test personnel and their experience. This is a crucial aspect for fully automated and production-accompanying NDT of resistance spot welding. In addition, the imaging analyses of the residual magnetic flux density works contactless and without couplants. The goal of reducing the test time and the subjectivity of the measurement data evaluation is achieved by an automated evaluation of the measurement results. The application of the algorithm is in principle conceivable for measurement data of other measuring systems, such as a magnetic field camera. This requires an adjustment of the decisive variables in the algorithm. The correlation between NDT and destructive testing depends on the material combination. Unexpectedly, the developed algorithm works very well on the challenging material combinations MC2, MC3, and MC4 compared to MC1. To improve the quality of the algorithm, further investigations should concentrate on the physical background of the threshold. The authors assume that the value of the threshold depends on the material and its magnetic properties, especially of the nugget. So far it is a big challenge to measure the magnetic properties of the nugget without influencing the microstructure or mechanical properties, which influence their magnetic behavior.

ACKNOWLEDGMENTS Das IGF-Vorhaben IGF 19.208 B/DVS-Nummer 04.070 der Forschungsvereinigung „Forschungsvereinigung Schweißen und verwandte Verfahren des DVS, Aachener Straße 172, 40223 Düsseldorf“ wurde über die AiF im Rahmen des Programms zur Förderung der industriellen Gemeinschaftsforschung und -entwicklung (IGF) vom Bundesministerium für Wirtschaft und Energie aufgrund eines Beschlusses des Deutschen Bundestages gefördert. Wir bedanken uns bei den Mitgliedern des projektbegleitenden Ausschusses, unseren Studenten und allen anderen Mitwirkenden für die Unterstützung bei der Durchführung des Vorhabens.

The authors thank AiF for funding the IGF-Project IGF 19.208 B of the Research Association on Welding and Allied Processes of the DVS, which was part of the program to support cooperative industrial research (Industrielle Gemeinschaftsförderung (IGF)) by the Federal Ministry of Economic Affairs and Energy, on the basis of a decision by the German Bundestag. Equal thanks go to all companies and participants, who contributed their support and knowledge to the project.

REFERENCES (1) Deutscher Verband für Schweißen und verwandte Verfahren e.V., 2000, “Widerstandspunkt-, Buckel- und Rollennahtschweißen von Stahlblechen bis 3 mm mit metallischen Überzügen.”, (Technical Bulletin), DVS 2920. (2) Deutscher Verband für Schweißen und verwandte Verfahren e.V., 2017, “Prüfen von Widerstandspressschweißverbindungen Zerstörungsfreie Prüfung.”, (Technical Bulletin), DVS 2916-5. (3) Füssel, U., Mathiszik, C., and Zschetzsche, J., 2019, “Zerstörungsfreie Charakterisierung der Anbindungsfläche beim Widerstandspress-schweißen durch bildgebende Analyse der Remanenzflussdichte.”, (Non-destructive characterization of spot welds by imaging analysis of the residual magnetic flux density) Schlussbericht. Technische Universität Dresden, Dresden. (4) DIN Deutsches Institut für Normung e.V., 2015, “Widerstandsschweißen – Verfahren zum Punktschweißen von niedriglegierten Stählen mit oder ohne metallischem Überzug.”, (Resistance welding – Procedure for spot welding of uncoated and coated low carbon steels), ISO 14373. (5) Reinhardt, T., 2018, “Erarbeitung eines Algorithmus zur automatisierten Auswertung der Messergebnisse der bildgebenden Analyse der Remanenzflussdichte.”, (Development of an automated evaluation method for measurement results of the imaging analysis of the residual magnetic flux density) Diplomarbeit. Technische Universität Dresden, Dresden. (6) DIN Deutsches Institut für Normung e.V., 2003, “Widerstandsschweißen - Zerstörende Prüfung von Schweißverbindungen - Brucharten und geometrische Messgrößen für Widerstandspunkt-, Rollennaht- und Buckelschweißungen.”, (Resistance welding - Destructive tests of welds - Failure types and geometric measurements for resistance spot, seam and projection welds (ISO 14329:2003)), ISO 14329. (7) Mathiszik, C., Reinhardt, T., Zschetzsche, J., and Füssel,123 U., 2018, “NDT of spot welds by imaging analysis of the residual magnetic flux density – Investigation on the influence of electrode indentation on the measurement results.”, Materials Testing Vol. 60 No. 12 2018: pp. 1179–1183. DOI 10.3139/120.111262. (8) voestalpine Stahl GmbH, 2017, “chemische Analyse und mechanische Kennwerte der gelieferten Versuchswerkstoffe für IGF 19.208.”, (chemical analysis and mechanical characteristics of the supplied experimental materials for IGF 19.208). (9) Metall Jobst, 2019, “Unlegierter Stahl DC04 W.-Nr. 1.0338: Techn. Info unleg. Stahl.”, URL https://www.metall-jobst.de/media/pdf/17/d1/6d/dc04.pdf. (10) Salzgitter Flachstahl GmbH, 2014, “22MnB5 Borlegierte Vergütungsstähle.”, Salzgitter. (11) Kaars, J., Mayr, P., and Koppe, K., 2018, “Determining Material Data for Welding Simulation of Presshardened Steel.”, Metals Vol. 8 No. 10 2018: p. 740. DOI 10.3390/met8100740. (12) Wink, H.-J. and Krätschmer, D., “Charakterisierung und Modellierung des Bruchverhaltens von Punktschweißverbindungen in pressgehärteten Stählen: Teil II - Simulation des Schweißprozesses.”. 11. LS- DYNA Forum.

measurement results of the imaging analysis of the residual magnetic flux density) Diplomarbeit. Technische Universität Dresden, Dresden. (6) DIN Deutsches Institut für Normung e.V., 2003, “Widerstandsschweißen - Zerstörende Prüfung von Schweißverbindungen - Brucharten und geometrische Messgrößen für Widerstandspunkt-, Rollennaht- und Buckelschweißungen.”, (Resistance welding - Destructive tests of welds - Failure types and geometric measurements for resistance spot, seam and projection welds (ISO 14329:2003)), ISO 14329. (7) Mathiszik, C., Reinhardt, T., Zschetzsche, J., and Füssel, U., 2018, “NDT of spot welds by imaging analysis of the residual magnetic flux density – Investigation on the influence of electrode indentation on the measurement results.”, Materials Testing Vol. 60 No. 12 2018: pp. 1179–1183. DOI 10.3139/120.111262. (8) voestalpine Stahl GmbH, 2017, “chemische Analyse und mechanische Kennwerte der gelieferten Versuchswerkstoffe für IGF 19.208.”, (chemical analysis and mechanical characteristics of the supplied experimental materials for IGF 19.208). (9) Metall Jobst, 2019, “Unlegierter Stahl DC04 W.-Nr. 1.0338: Techn. Info unleg. Stahl.”, URL https://www.metall-jobst.de/media/pdf/17/d1/6d/dc04.pdf. (10) Salzgitter Flachstahl GmbH, 2014, “22MnB5 Borlegierte Vergütungsstähle.”, Salzgitter. (11) Kaars, J., Mayr, P., and Koppe, K., 2018, “Determining Material Data for Welding Simulation of Presshardened Steel.”, Metals Vol. 8 No. 10 2018: p. 740. DOI 10.3390/met8100740. (12) Wink, H.-J. and Krätschmer, D., “Charakterisierung und Modellierung des Bruchverhaltens von Punktschweißverbindungen in pressgehärteten Stählen: Teil II - Simulation des Schweißprozesses.”. 11. LS- DYNA Forum.

124

Process for Nondestructive Testing of Pressure Vessels ProcessUsing Electronic for Nondestructive Distance MeasurementsTesting of Pressure to Measure Vessels 3-D Using Coordinates Electronic Distance Measurementsof Cardinal Points to Measure While 3 Pressure-D Coordinates and/or ofForce Cardinal Testing Points While Pressure and/or Force Testing David H. Parker David H. Parker Parker Intellectual Property Enterprises, LLC Parker Intellectual3919 Deepwoods Property Enterprises, Road LLC 3919Earlysville, Deepwoods VA 22936 Road (434) 975-3345;Earlysville, email [email protected] VA 22936 (434) 975-3345; email [email protected]

ABSTRACT This is a companion paper to Opportunities for the use of electronic distance measurement instruments in nondestructive testing and structural health monitoring and implications for ASNT, which was presented at the 2018 ASNT Research Symposium. The earlier paper covered the background and capabilities of electronic distance measurement instruments, and in general how they could be used for nondestructive testing applications, which will not be repeated. This paper covers specific applications limited to pressure vessels, such as; boilers, receivers, nuclear reactor containment buildings, tank trucks, railway tank cars, storage tanks, ships, vacuum chambers, aircraft, spacecraft, and the like. An example experimental architecture and 3-D uncertainty analysis, using manufacturers instrument specifications and commercially available software (MicroSurvey® STAR*NET), is included for nondestructive testing of railway tank cars subjected to measured pressure and coupler forces. Structurally sound tank cars can be quickly, and quantifiably, identified by comparing the measured geometric performance of targeted cardinal points, under the pressures and forces, to finite element models (FEM), historical measurements of the car, or looking for salient characteristics such as; linearity, hysteresis, creep, symmetry, and the like—while defective tank cars will exhibit anomalous geometric performance, which requires further investigation. The net result would be that instead of releasing a tank car into service based on not finding a defect, it would be released into service based on measured structural performance.

Keywords: tank car, nondestructive testing, structural health monitoring, electronic distance measurement, laser tracker

BACKGROUND The 2018 Transportation Research Board (TRB) Annual Meeting, included Workshop 164 Bridging the Gap Between Non-Destructive Evaluation and Structural Health Monitoring. The purpose of the workshop was described in the program as: The nondestructive evaluation (NDE) research community focuses primarily on the identification and characterization of local material-level degradation and damage. Structural health monitoring (SHM) researchers aim to provide global structural assessments associated with load paths, load carrying capacities, and more. Although both approaches clearly are complementary in nature, these research communities are mostly independent and interact little. This workshop brings these groups together to promote and identify opportunities for the integration of NDE and SHM.

Unfortunately, while there was general agreement with the notion that NDE and SHM groups need to be brought together, and a paper expanding on the subject was presented at the 2018 ASNT Research Conference [1], there was no follow up at the 2019 TRB Annual Meeting. This paper will narrow the general notion to a more specific example that incorporates NDE, SHM, and precision dimensional metrology—Electronic Distance Measurement (EDM) in particular—to applications for pressure vessels.

A general background on statutes, regulations, industry standards, and the state-of-the-art of NDT for pressure vessels can be found in US Patent 9,354,043 (‘043) [2,3]. Columns 5-10 in the Background section of ‘043 cover; Nuclear Power Plant Containment Buildings, Boilers and Unfired Pressure Vessels, Transportation of Hazardous Materials (trucks, railroad cars, pipelines), and Aircraft and Spacecraft. Columns 25-31, in the Detailed Description section, cover novel ideas for making a global structural assessment (GSA) of; Containment Buildings, Ground Transportation, Aircraft, and Reservoirs.

This paper will narrow the scope to railroad tank cars, in order to give a specific example of the proposed methods, although it will be recognized that the methods can be adapted to a broad array of other applications. The fleet of railroad tank cars increased from 300,000 in 2010, to 415,000 in 2017, which indicates a large, and growing, need for inspection.

Tank cars present interesting NDT challenges, with potentially catastrophic risks, such as exposure of the population to poison inhalation hazard (PIH), toxic inhalation hazard (TIH), and flammable liquid materials. They can be lined inside to protect against chemicals, covered by an insulation layer on the outside for transport of viscus liquids, and difficult to access on the inside due to toxic, or explosive, material contamination. This can make visual inspections, and inspections that require contact with the metal, impractical.

Unlike fixed pressure vessels; which can be protected from the environment and physical damage, and electronically monitored in real-time; railroad tank cars are subjected to environmental extremes, unknowable forces and impacts, and the only routine monitoring is limited to visual inspection, from the outside, by workers loading and unloading the car.

Fleet owners are in a similar situation with civil structure owners, i.e., they need a GSA to ensure that a car, or bridge, is safe to put in service. Presently, this is based on inspections tailored to finding defects, with visible inspections being the predominant means. If no defects are seen, it is assumed the car, or bridge, is safe.

It will be argued that dimensional metrology has reached a point where precision 3-D measurements can be made in the field that allow a large object to be subjected to loadings, while cardinal points are measured, which can determine the structural performance of the large object. This makes it practical to screen the entire fleet of tank cars at regular intervals, in a go/no go test. If the structural performance, or GSA, meets the specifications, it would be reasonable to put the car in service.

If, on the other hand, the car does not meet the GSA specifications, more detailed inspection and testing would be required to determine the cause of the impaired structural performance. For example, the car may need to be cleaned to allow entry for inspection, a liner may have to be removed, or insulation may have to be removed.

RAILROAD TANK CARS Nondestructive testing of railroad tank cars is regulated by the Association of American Railroads (AAR), the DOT Federal Railroad Administration (FRA), the DOT Pipeline and Hazardous Materials Safety Administration (PHMSA), the DOT Research and Special Programs Administration (RSPA), and the United States Code of Federal Regulations (CFR) [4,5].

In 1987, RSPA opened docket HM-201 for Detection and Repair of Cracks, Pits, Corrosion, Lining Flaws, Thermal Protection Flaws, and Other Defects of Tank Car Tanks [6]. A Notice of Proposed Rulemaking (NPRM) was published in 1993 [7]. Section A of the NPRM concluded with the statement:

126

Based on the ineffectiveness of hydrostatic test in detecting significant fatigue cracking in tank cars resulting from severe loadings, stress risers, and welding defects, RSPA and the FRA no longer consider the hydrostatic test part of the optimum way to qualify fusion welded tank cars for continued service.

Final rule changes were published in 1995 [8]. In support of HM-201, the FRA Office of Research and Development contracted with the Transportation Technology Center, Inc. (TTCI) a subsidiary of the AAR to perform joint government/industry evaluation of possible replacement test/inspections for the previously prescribed hydrostatic test/visual inspection of tank cars. The Office of Research and Development published the research in a 2002 report [9]. Subsection 3.3 covers Developing a Validation Methodology. Sub subsections cover: 3.3.1 Liquid Penetrant Test Method, 3.3.2 Magnetic Particle Test Method, 3.3.3 Radiographic Test Method, 3.3.4 Ultrasonic Test Method, 3.3.5 Visual Test Method, and 3.3.6 Acoustic Emission Test Method, i.e., conventional NDT methods “focused primarily on the identification and characterization of local material-level degradation and damage”, but nothing “to provide a global structural assessment associated with load paths, load carrying capacities, and more.” Other inspection reports are in the literature [10, 11, 12]

In addition to forces exerted by the contents of the vessel, tanks used in transportation are also exposed to forces due to transportation. This is well documented by an experiment in which a tank car was instrumented and put in revenue service, in 2016 [13]. Increased safety margins are required to protect the public in the event of an accident or leak and to reduce liability for property damages.

For example, as explained in US Patent 4,805,540 ('540), in railway tank cars, the cylindrical tank is part of the railcar structure. Modern US tank cars no longer have a center sill running between the two couplers to carry the draft load of the train. Instead, a stub sill and coupler is attached to each end of the tank. The tank is attached to the stub sill by a saddle arrangement as described in US Patents 5,351,625; 5,467,719; and 7,806,058, for example. Each stub sill is pivotally connected to a truck with 4 wheels and springs to support the respective end and roll on the tracks. The coupler forces are transferred from a first coupler on a first end through a first stub sill to the tank, through the tank, to a second stub sill, and to a second coupler on a second end.

As explained in '540, because the stub sill assembly and the coupler are attached to the lower side of the car, there is a significant moment introduced into the tank structure by the coupler forces. For example, for the first coupler and the second coupler in tension, or when the slack is out, the forces on the tank will be in tension at the lower side of the tank and in compression at the upper side of the tank. The forces will be reversed when the slack is in.

As explained in '540; Because the center stub sill assembly and the coupler carried thereby were located somewhat below the level of the bottom of the tank (about 8--12 inches below the bottom of the tank), and because the cylindrical tank structure carried the longitudinal train loads axially of the car, an offset moment arm between the tank structure and the centerline of the coupler was present. This offset resulted in a significant overturning moment being induced in the center stub sill and in the end portion of the tank such that the end of the tank and the center sill assembly must withstand these overturning moments. It will be appreciated that the longitudinal train loads that the car is required to withstand, in accordance with the American Association of Railroads (AAR) [5] is a dynamic or impact load of 1,250,000 pounds and a static squeeze of compression load of about 1,000,000 pounds. Because of the vertical offset and the magnitude of the loads, the overturning moments are very significant

Under normal operation, the saddle is supported on a center plate which acts as a bearing between the saddle and truck bolster. The center plate mates to a bowl shaped portion on the bolster, much like a thrust washer. This results in minimal twisting of the tank, even if the track is uneven, e.g., at an industrial siding that is privately owned and maintained. However, as explained in US Patent 6,357,363,

127

... the end portions of the underframe of such a car have to be of relatively heavy construction in order to permit the car when loaded to be supported on jacks located at the corners of the underframe, since there is a relatively long lever arm between the corners of the underframe and the saddle attachment locations, where the weight of the tank and included freight is transferred to the underframe.

As explained in US Patent 5,076,173, AAR requires provisions for vertical lifting of a tank car by a crane, which can also introduce twisting forces into the tank shell. In some cases, such as cars rated for PIH or TIH lading, the structural requirements may be dictated by the ability to withstand a crash, or puncture resistance. US Patent 7,975,622, is an example of such a TIH rated car.

As detailed hereinabove, transportation of hazardous materials is closely regulated by Congress through the DOT. US Patents 6,597,973 ('973); 6,832,183 ('183); and 6,955,100 ('100), outline how General Electric Railcar Services Corporation conducts inspections in compliance with 49 CFR non-destructive testing (NDT) regulations [4]

Table 1 shows representative specification for tank cars built by The Greenbrier Companies. Table 2 shows calculated strain in the length and hoop directions, as well as changes in length and radius as a function of pressure. For example, the DOT 112A340W car, which is designed for liquefied petroleum gas (LPG), will expand in length by 0.034 mm/psig, or 11.5 mm at the rated test pressure. The radius will expand by 0.006 mm/psig, or 2.04 mm at rated test pressure.

Table 1. Specifications for representative tank cars Description Capacity Bursting Test Diameter Length Plate (gallons) Pressure (psig) Pressure (ft) (ft.) thickness (psig) (in) DOT 30,500 500 100 9.5 55 9/16 117A100W1 DOT 29,000 500 100 9 53.25 7/16 111A100W1 DOT 33,700 850 340 10 60.25 9/16 112A340W

Table 2. Calculated strain and changes in length and radius as a function of pressure Description Area Force_z μ strain_l μ strain_h Δ l Δ radius (in^2) (lbf/psig) (/psig) (/psig) (mm/psig) (mm/psig) DOT 201 10,207 1.75 3.49 0.029 0.005 117A100W1 DOT 148 9,161 2.13 4.26 0.035 0.006 111A100W1 DOT 212 11,310 1.84 3.68 0.034 0.006 112A340W

ELECTRONIC DISTANCE MEASUREMENT

Commercially available EDM instruments (laser trackers) are widely used in the aerospace, precision manufacturing, and shipbuilding industries for building large-scale structures. Conventional EDM capabilities have been reviewed extensively in the literature [14, 15, 16, 17, 18]. It has been argued that adoption of EDM capabilities for NDT applications has been largely overlooked, or misapplied, in numerous papers, since 2017, by the author [1, 19, 20, 21, 22], and will not be repeated here. Table 3 is a summary of range and accuracy for laser trackers.

Table 3: Range and accuracy for laser trackers Manufacturer Model Range Accuracy Data Rate API Automated Precision Radian 80 m 10 μm or 0.7 μm/m ? 128 FARO Vantage 80 m 16 μm + 0.8 μm/m 1,000 points/sec Kern (no longer available) ME 5000 Mekometer 4,000 m 200 μm + 0.2 μm/m ? Leica AT403 160 m 10 μm ? Leica AT960-LR 160 m 0.5 μm/m 1,000 points/sec Nikon MV351 HS 50 m 10 μm/m 2 sec/point NRAO (no longer available) PSH97 1,000 m 50 μm/m 1,000 points/sec

For reasons that have been well documented in the literature [14, 22], the accuracy of laser trackers is extremely high in the radial direction, i.e., on the order of one part per million of the distance measured. However, the two angle measurements needed to calculate a 3-D coordinate are much less accurate, i.e., on the order of one arc second—or 5 parts per million of the radial distance. In order to obtain the highest accuracy, for measuring 3-D coordinates of small deflections of stiff structures, multiple instruments are used to measure in a trilateration architecture. For example, the ideal geometry would be to have three instruments measuring from orthogonal directions, which would yield 3-D coordinates with the radial accuracy in all three directions. In most cases, this geometry would be impractical, so tradeoffs are made between number of instruments and required accuracy. Commercially available least squares post-processing software packages, such as MicroSurvey® STAR*NET, are available to adjust the measurements to obtain higher accuracy [23,24].

MEASUREMENT ARCHITECTURE An example will illustrate the methods. Figure 1 shows an end view of a measurement station with 8 laser trackers. For convenience, the conventional surveying coordinate system (North, East, South, West, and elevation) will be used to conform to STAR*NET nomenclature, with the N, E origin being the center of the car, and the elevation being the top of rail. Instruments are identified with 3 on the West side (IW1, IW2 and IW3), 3 on the East side (IE1, IE2, and IE3), 1 on the center top (ICT), and 1 on the center bottom (ICB).

Figure 2 shows the East side of the car with 5 equally spaced measurement point targets numbered from the South end to the North end, i.e., TE1, TE2, TE3, TE4, and TE5. Corresponding targets are located on the West side (TW1, TW2, TW3, TW4, and TW5); top side (TT1, TT2, TT3, TT4, and TT5); and bottom side (TB1, TB2, TB3, TB4, and TB5). Some lines of sight from ICB to the bottom targets will be obstructed by the trucks and other equipment. In some cases, a mirror may be necessary in order to make the measurements.

UNCERTAINTY ANALYSIS

STAR*NET has a Preanalysis feature which calculates the standard deviation for prescribed measurements, based on the geometry, and instrument errors, i.e., before actually making measurements. The analysis was performed under the assumption of a constant error of 0.01 mm + 1 part per million for the range, and one arc second for the angles. Each target was measured by the 3 instruments with a line of sight to the target. As one would expect, due to symmetry, the standard deviations are symmetric about the origin, i.e., there is North-South, and East-West symmetry. The abbreviated results are summarized in Table 4. Page restrictions prohibit including the STAR*NET output listing in an Appendix, but copies are available on request.

Table 4. Standard deviations of measurements

for NDT applications has been largely overlooked, or misapplied, in numerous papers, since 2017, by the author [1, 19, 20, 21, 22], and will not be repeated here. Table 3 is a summary of range and accuracy for laser trackers.

Table 3: Range and accuracy for laser trackers Manufacturer Model Range Accuracy Data Rate API Automated Precision Radian 80 m 10 μm or 0.7 μm/m ? FARO Vantage 80 m 16 μm + 0.8 μm/m 1,000 points/sec Kern (no longer available) ME 5000 Mekometer 4,000 m 200 μm + 0.2 μm/m ? Leica AT403 160 m 10 μm ? Leica AT960-LR 160 m 0.5 μm/m 1,000 points/sec Nikon MV351 HS 50 m 10 μm/m 2 sec/point NRAO (no longer available) PSH97 1,000 m 50 μm/m 1,000 points/sec

UNCERTAINTY ANALYSIS

Table 4. Standard deviations of measurements

129

UNCERTAINTY ANALYSIS

Table 4. Standard deviations of measurements Target N (mm) E (mm) Elev. (mm) Diameter (mm) TE1 (TE5) 0.010 0.017 0.021 0.024 TE2 (TE4) 0.009 0.010 0.012 0.014 TE3 0.004 0.007 0.006 0.009

Note that the tank car will likely shift as it is subjected to pressure and forces, so absolute coordinates are not as useful as differential measurements. For example, measurements between the East and West sides can be combined to determine the diameter, i.e., the measurements act as a virtual 10-foot micrometer to measure the diameter. The combined standard deviation of the diameter will increase by √2, as shown in Table 4.

The standard deviation of a differential measurement of the length of the tank, between TE1 and TE5, will be 0.014 mm. To put this in perspective, the thickness of a standard sheet of printer paper is 0.100 mm. If you assume an uncertainty of three standard deviations, the uncertainty in the length of the car is less than half the thickness of a sheet of paper!

MEASUREMENT ANALYSIS The exceptional accuracy of the measured 3-D coordinates of cardinal points makes new test procedures available. For example, cracks will likely exhibit nonlinear behavior, as well as hysteresis. By applying compression and tension on the couplers, cracks could be opened and closed, which would be highly nonlinear. Alternatively, by pulling a slight vacuum on the car, nonlinear behavior could identify problems.

Asymmetric movements will also be measurable between sections of the car. For example, if the lower section has lost metal due to corrosion, or if there were circumferential cracks in the lower portion, the car will curve like a banana as it is pressurized. Longitudinal cracks would cause a rotational movement in that section, or exhibit asymmetrical bulges. By lifting a tank car at the corners, the tank can be twisted. Nonlinearities and hysteresis in the structural performance would indicate the need for further investigation.

Tank cars are built in programs, with many being built under the same specifications, by the same shop. The structural performance would be expected to be very similar between cars, and should closely match a finite element model (FEM). It would be reasonable to document the dimensions of cars throughout the life of the car. Since the measurements are traceable to NIST, measurements could be made at various locations over the years with the same results. Abnormal changes in the structural performance would be easily recognized, which would take the car out of service for additional inspections. More importantly, cars that pass the structural performance testing would be put back in service with a higher degree of confidence that it is safe to operate.

In other words, if the tank car is subjected to 1,000,000 lbf in compression and tension through the couplers, full pressure, slight vacuum, and twisting in both directions; and the measured 3-D deformations demonstrate a global structural assessment that meets specifications—it is probably safe to put in service.

CONCLUSIONS

ʹͲͳͺͳ͸ͶǢ ǡ ǡ ǡ ǡ ǡ Ǥ ǡǤ

130

an tank of Sideview car 2: Figure

Figure 1: End view of tank car and instruments.

d targets .

Figure 2: Side view of tank car and targets.

131

REFERENCES

[1] David H. Parker. Opportunities for the use of electronic distance measurement instruments in nondestructive testing and structural health monitoring and implications for ASNT. In Proceedings of 27th ASNT Research Symposium, Orlando, FL, pages 159–169. American Society for Nondestructive Testing, March 2018. [2] David H. Parker and John M. Payne. Methods for measuring and modeling the structural health of pressure vessels based on electronic distance measurements, 2016. US Patent 9,354,043. [3] Robert E. Shannon. US 9354043 Method for measuring and modeling the structural health of pressure vessels based on electronic distance measurements. Materials Evaluation, 74(8):1140–1142, August 2016. Published under the What’s New, New Patents Column.

[4] 49 CFR 179–Specifications for Tank Cars. Code of Federal Regulations.

[5] AAR Manual of Standards and Recommended Practices; Section C Part III, Specifications for Tank Cars, M- 1002. Association of American Railroads. [6] Detection and repair of cracks, pits, corrosion, lining flaws, thermal protection flaws, and other defects of tank car tanks. 52 FR 46510, December 1987. [7] Research Department of Transportation and Special Programs Administration. Detection and repair of cracks, pits, corrosion, lining flaws, thermal protection flaws and other defects of tank car tanks. 58 FR 48485, September 1993.

[8] Crashworthiness protection requirements for tank cars; detection and repair of cracks, pits, corrosion, lining flaws, thermal protection flaws and other defects of tank car tanks. 60 FR 49048, September 1995.

[9] Gregory A. Garcia. Railroad tank car nondestructive methods evaluation. Technical Report DOT/FRA/ORD- 01/04, Federal Railroad Administration, January 2002.

[10] Akram Zahoor. Materials and fracture mechanics assessments of railroad tank cars. Technical Report NISTIR 6266, National Institute of Standards and Technology, September 1998.

[11] Mary Ruth Johnsen. Inspecting rail tank cars. Inspection Trends, Summer:15–17, 2007. [12] Gregory A. Garcia, Ward Rummel, and Francisco Gonzalez. Quantitative nondestructive testing of railroad tank cars using the probability of detection evaluation approach. Technical Report DOT/FRA/ORD-09/10, Federal Railroad Administration, May 2009. [13] Narayana Sundaram. Force environment evaluation of stub sills on tank cars using autonomous over-the road testing of the instrumented tank car. Technical Report DOT/FRA/ORD-16/39, Federal Railroad Administration, December 2016. [14] Scott Sandwith and Read Predmore. Real-time 5-micron uncertainty with laser tracking interferometer systems using weighted trilateration. 2001 Boeing Large-Scale Metrology Conference, St. Louis, MO, 2001. [15] W. T. Estler, K.L. Edmundson, G.N. Peggs, and D. H. Parker. Large-scale metrology—an update. Annals of the CIRP, 51(2):587–609, 2002. Keynote Paper. [16] G. N. Peggs, P. G. Maropoulos, E. B. Hughes, A. B. Forbes, S. Robson, M. Ziebart, and B. Muralikrishnan. Recent developments in large-scale dimensional metrology. Proceedings of the Institution of Mechanical Engineers Part B Journal of Engineering Manufacture, June 2009. [17] R. H. Schmitt, M. Peterek, E. Morse, W. Knapp, M. Galetto, F. Ha¨rtig, G. Goch, B. Hughes, A. Forbes, and W. T. Estler. Advances in large-scale metrology–review and future trends. CIRP Annals Manufacturing Technology, 65:643 –665, 2016. [18] Bala Muralikrishnan, Steve Phillips, and Daniel Sawyer. Laser trackers for large-scale dimensional metrology: A review. Precision Engineering, 44:13–28, 2016. [19] David H. Parker. Nondestructive testing and monitoring of stiff large-scale structures by measuring 3-D coordinates of cardinal points using electronic distance measurements in a trilateration architecture. In Conference on Nondestructive characterization and monitoring of advanced materials, aerospace, and civil infrastructure 2017, Portland, OR, volume 10169 of Proceedings of SPIE. SPIE, March 2017. paper 1016918. [20] David H. Parker. Using electronic distance measuremen132 t instruments in NDT and structural health monitoring applications. Quality Digest, August 2017. Paper given at CMSC 2017, Snowbird, UT, original title “Opportunities for the use of electronic distance measurement instruments in nondestructive testing and structural health monitoring applications and how instrument manufacturers can facilitate early adopters in new fields”. [21] David H. Parker. Experimental uncertainty analysis for nondestructive testing of the CSX Wilbur Bridge using electronic distance measurements to measure 3-D coordinates of cardinal points. In CMSC Annual Conference, Reno, NV. Coordinate Metrology Society Conference, July 2018. [22] David H. Parker and John M. Payne. Methods for measuring and modeling the process of prestressing concrete during tensioning/detensioning based on electronic distance measurements, February 2019. US Patent 10,203,268. [23] MicroSurvey Software Inc. STAR*NET 9, 2017. [24] Barry N. Taylor and Chris E. Kuyatt. Guidelines for evaluating and expressing the uncertainty of NIST measurement results. NIST Technical Note 1297, National Institute of Standards and Technology,1994.

A review. Precision Engineering, 44:13–28, 2016. [19] David H. Parker. Nondestructive testing and monitoring of stiff large-scale structures by measuring 3-D coordinates of cardinal points using electronic distance measurements in a trilateration architecture. In Conference on Nondestructive characterization and monitoring of advanced materials, aerospace, and civil infrastructure 2017, Portland, OR, volume 10169 of Proceedings of SPIE. SPIE, March 2017. paper 1016918. [20] David H. Parker. Using electronic distance measurement instruments in NDT and structural health monitoring applications. Quality Digest, August 2017. Paper given at CMSC 2017, Snowbird, UT, original title “Opportunities for the use of electronic distance measurement instruments in nondestructive testing and structural health monitoring applications and how instrument manufacturers can facilitate early adopters in new fields”. [21] David H. Parker. Experimental uncertainty analysis for nondestructive testing of the CSX Wilbur Bridge using electronic distance measurements to measure 3-D coordinates of cardinal points. In CMSC Annual Conference, Reno, NV. Coordinate Metrology Society Conference, July 2018. [22] David H. Parker and John M. Payne. Methods for measuring and modeling the process of prestressing concrete during tensioning/detensioning based on electronic distance measurements, February 2019. US Patent 10,203,268. [23] MicroSurvey Software Inc. STAR*NET 9, 2017. [24] Barry N. Taylor and Chris E. Kuyatt. Guidelines for evaluating and expressing the uncertainty of NIST measurement results. NIST Technical Note 1297, National Institute of Standards and Technology,1994.

133

Resistivity Behavior of Concrete Mixtures Resistivitywith Behavior Included of ConcreteSupplementary Mixtures Cementitious with Included Materials Supplementary Cementitious Materials Cody Shults and Julie Hartell Cody Shults and Julie Hartell Oklahoma State University 207 EngineeringOklahoma South State, Stillwater, University OK 74078 207 Engineering(405) 744-5189 South,; fax Stillwater, (405) 744 OK-7554 74078 email [email protected];(405) 744-5189; fax (405) [email protected] 744-7554 email [email protected]; [email protected]

ABSTRACT The inclusion of supplementary cementitious materials (SCMs) in a concrete mixture will affect the results of surface resistivity tests conducted on that concrete mixture. This study was conducted in order to understand the effect that common SCM replacement ratios will have on surface resistivity results as this is an important relationship to understand if surface resistivity is to be used in the field. In this study, control mixes of set water-to- cementitious material ratios were cast along with the same mixes with the inclusion of three different SCMs at selected cement replacement ratios. The SCM replacements studied included class c fly ash at 5% and 20%; blast furnace slag at 5% and 40%; and silica fume at 2% and 8%. Also studied were the effects of two common admixtures that are typically used in the industry (an air entraining agent and a high range water reducer). The surface resistivity values of the cast cylinders from each mix were taken at 1, 3, 7, 14, 21, 28, and 56 days. An ANNOVA test was performed to see the significant statistical difference between the given parameters of difference in water-to-cementitious material ratio, the percent of SCM replacement, and the inclusion of admixtures.

Keywords: Concrete, Resistivity, Silica Fume, Blast Furnace Slag, Portland Cement

INTRODUCTION Surface resistivity has been utilized by the industry as a way to indicate the apparent resistivity of a concrete in-situ and use this data to indicate areas of potential corrosion. Further research is also being done to correlate resistivity in concrete to permeability. The image below shows the instrument used for this study along with its test principle. The test method is based on the Wenner probe method initially developed for geotechnical purposes. First, the four probes are saturated in water and are placed on the surface of a concrete cylinder along its longitudinal axis. The outer probes produce a small alternating current traveling through the concrete cylinder. The inner probes are connected to a voltmeter and measure the voltage response to current flow. The measuring device will display the apparent resistivity of the concrete cylinder. The resistivity measurement is determined from Eq.1 (shown below). To determine the true resistivity of the concrete, the value recorded can be factorized to compensate for specimen geometry by multiplying the value with a factor based on a ratio of the sample’s cross-sectional area to its length [1].

(Eq. 1) 2𝜋𝜋𝜋𝜋𝜋𝜋 Where, 𝜌𝜌 = 𝐼𝐼 ρ: apparent resistivity (ohm-cm) S: spacing between probes (cm) V: measured voltage (volts) I: amplitude of alternating current (amps)

Instrument and representation of principle

Fresh and hardened concrete are routinely tested to report and verify the quality of the material. There is still a level of uncertainty when it comes to validating the presence of supplementary cementitious materials (SCMs) in early age concrete. SCMs are prescribed to attain a required level of durability depending on concrete exposure class. So far, there is no simple test method which can assess the presence of an SCM within a routine quality assurance and control plan. The combination of both physical and chemical attributes makes concrete a sensitive material to test using electrical resistivity. Concrete level of saturation with water and temperature are two large factors that hinder the development of surface resistivity methods for in-situ testing. In order for this test to be further used in the field, the effects of concrete constituents on surface resistivity must be understood as their effects on readings is critical to gathering accurate data.

Due to its sensitivity to the chemical and physical characteristics of a cementitious material, nondestructive electrical methods such as surface resistivity and bulk resistivity are gaining popularity in the cement and concrete industry. Previous studies demonstrated the existence of a correlation between the conventional method for durability assessment of concrete mixtures, the rapid chloride permeability test (RCPT), and electrical conductivity testing. The latter method was deemed accurate and reliable for determining the corrosion performance of a concrete mixture depending on its performance in resisting ionic flow [2-4]. One can use a simple classification table, derived from the RCPT standard method of testing (ASTM C1202), to estimate the chloride ion penetration level based on the result of a surface resistivity test. [2] These studies led to the development of AASHTO TP 95: Standard Method of Test for Surface Resistivity Indication of Concrete's Ability to Resist Chloride Ion Penetration [5] and AASHTO TP 119: Standard Method of Test for Electrical Resistivity of a Concrete Cylinder Tested in a Uniaxial Resistance Test [6]. Moreover, resistivity testing has been found to be less expensive to perform in comparison to RCPT; therefore, providing motivation for implementation of the method in routine control activities.

This study was conducted in order to understand the effect that common SCM replacement ratios will have on the surface resistivity results, as this is an important relationship to understand if surface resistivity is to be used in the field. The methodology proposed will enable the development of a method based on resistivity criteria to identify whether the mixture contains a certain type of supplementary cementitious material. This will aid in the development of a new quality control and assurance criteria for concrete mixture approval in addition to currently used test methods and specifications. This would allow infrastructure owners and stakeholders to produce high quality and durable concrete. The objectives of the experimental study are to determine the efficacy of resistivity testing in detecting and identifying SCM inclusion.

EXPERIMENTAL METHOD Materials 135 In this study, 75 concrete mixtures were made in the laboratory (ASTM C192). [7] The concrete mixtures varied in water-to-cementitious materials (w/cm) (0.40, 0.45, and 0.50 w/cm). The mixtures also varied in the amount of supplementary cementitious used: 0%, 5%, 20% fly ash; 0%, 2%, 8% silica fume; and 0%, 5%, 40% slag cement. The aggregate proportions were kept constant while the water content was varied to achieve the desired water-to- cementitious materials ratio. For each mixture, one series was prepared with no admixtures; a second series was prepared with the addition of an air-entraining agent to achieve a percent air content of approximately 6%; and a third series was prepared with the addition of both air-entraining and water-reducing agents to evaluate the impact of admixture addition on resistivity testing.

The aggregate that was used in the preparation of the concrete mixtures was a #57 crushed dolomite coarse aggregate and a natural sand fine aggregate (ASTM C33). [8] The cement type used was a type-I (ASTM C150). [9] Three supplementary cementitious materials were tested, a class-C fly ash (ASTM C618), a slag cement (ASTM C989), and silica fume (ASTM C1240). [10-12] Chemical admixtures were also used in select mixtures, an air entrainer (MasterAir AE 90) and a water reducer (ADVA Cast 600).

Concrete mixture batching, mixing, and casting was completed at the Bert Cooper Engineering Laboratory in the mixing room facilities which was temperature controlled. All relevant ASTM standardized procedures were followed in aggregate preparation, mixing, casting, and material quality in order to maximize reproducibility. All cylindrical samples (Ø4” x 8”) per mixture design were cast from a single batch in order to decrease potential error due to mixture design variations. For this study, six replicates were made for each mixture type and the total number of samples cast was 450. Cylindrical samples were prepared in two equal layers and were consolidated by rodding. After 24 hours of curing in their molds, the samples were demolded and placed in a limewater tank stored in a temperature controlled room (ASTM C511). [13] For this study, immersion curing was selected since it is the common method of curing within the state of Oklahoma making study outcomes relevant for this state.

Testing Procedure For this study, surface resistivity testing was performed in accordance with AASHTO T 358, Standard Method of Test for Surface Resistivity Indication of Concrete’s Ability to Resist Chloride Ion Penetration. The values recorded were not factorized; therefore, they correspond to the apparent resistivity of a (Ø4”x8”) cylindrical sample [14]. In accordance with the method, the probe placement location onto the surface of the cylinder was marked immediately following demolding to ensure that each measurement was taken at the same location throughout the testing process.

To ensure the lowest possible variability in the measurements, special care was taken of the surface conditions of the samples. Once removed from the limewater tank for testing, the surface of the cylinder was lightly sprayed with tap water to remove any excess salts that had accumulated on the surface. The surfaces of the cylinders were kept moist while not letting so much water accumulate on the surface that the flow of current passes through it. During testing, the samples were kept in a temperature and humidity controlled laboratory environment. The resistivity probes were also kept in the same temperature and humidity controlled room in order to minimize the effect of temperature fluctuation on the measurements. All measurements taken were taken by the same resistivity probe and same probe spacing of 1.5” in order to minimize the variability of the measurements taken. For each cylinder, 6 resistivity tests

quality and durable concrete. The objectives of the experimental study are to determine the efficacy of resistivity testing in detecting and identifying SCM inclusion.

EXPERIMENTAL METHOD

Materials In this study, 75 concrete mixtures were made in the laboratory (ASTM C192). [7] The concrete mixtures varied in water-to-cementitious materials (w/cm) (0.40, 0.45, and 0.50 w/cm). The mixtures also varied in the amount of supplementary cementitious used: 0%, 5%, 20% fly ash; 0%, 2%, 8% silica fume; and 0%, 5%, 40% slag cement. The aggregate proportions were kept constant while the water content was varied to achieve the desired water-to- cementitious materials ratio. For each mixture, one series was prepared with no admixtures; a second series was prepared with the addition of an air-entraining agent to achieve a percent air content of approximately 6%; and a third series was prepared with the addition of both air-entraining and water-reducing agents to evaluate the impact of admixture addition on resistivity testing.

136

Example of surface resistivity test setup and resistivity probe.

RESULTS

Effect of Class-C Fly Ash Addition

14.0 12.0 cm) -

Ω 10.0 8.0 6.0

Resistivity (K Resistivity 4.0 2.0 0.0 0 7 14 21 28 Days 0% FA 5% FA 20% FA Figure 1: Comparison of variation of percent SCM for mixtures prepared with Type I cement, Class-C fly Ash and 0.4 w/cm

were performed in time: at the time of demolding (day 1), day 3, day 7, day 14, day 21, and day 28. The results shown in the results section are average values for each day of the six sample replicates.

Example of surface resistivity test setup and resistivity probe.

RESULTS

Effect of Class-C Fly Ash Addition

14.0 12.0 cm) -

Ω 10.0 8.0 6.0

Resistivity (K Resistivity 4.0 2.0 0.0 0 7 14 21 28 Days 0% FA 5% FA 20% FA Figure 1: Comparison of variation of percent SCM for mixtures prepared with Type I cement, Class-C fly Ash and 0.4 w/cm

137

14.0

12.0

10.0

8.0 cm) -

Ω 6.0

4.0

2.0 Resistivity (K Resistivity 0.0 0 7 14 21 28 Days 0% FA 5% FA 20% FA Figure 2: Comparison of variation of percent SCM for mixtures prepared with Type I cement, Class-C fly ash and 0.45 w/cm

14.0 12.0 10.0 8.0 cm) -

Ω 6.0 4.0 2.0

Resistivity (K Resistivity 0.0 0 7 14 21 28 Days 0% FA 5% FA 20% FA Figure 3: Comparison of variation of percent SCM for mixtures prepared with Type I cement, Class-C fly ash and 0.50 w/cm

Table 1: Statistical comparison of mean 28-day resistivity values for mixtures of varying SCM percent content (none, low percentage and high percentage) prepared with Type I cement, Class-C fly ash and containing no admixture ANOVA Student T-test 0%/low%/high% 0%/low% 0%/high% Low%/high% 0.40 w/cm p-values 0.049 0.153 0.235 0.021 0.45 w/cm p-values 0.055 0.025 0.324 0.091 0.50 w/cm p-values 0.033 0.174 0.099 0.025

Effect of Silica Fume Addition

138

60.0

50.0

cm) 40.0 - Ω 30.0

20.0

Resistivity (K Resistivity 10.0

0.0 0 7 14 21 28 Days 0% SF 2% SF 8% SF Figure 4: Comparison of variation of percent SCM for mixtures prepared with Type I cement, silica fume and 0.40 w/cm.

60.0

50.0 cm)

- 40.0 Ω 30.0

20.0

Resistivity (K Resistivity 10.0

0.0 0 7 14 21 28 Days 0% SF 2% SF 8% SF Figure 5: Comparison of variation of percent SCM for mixtures prepared with Type I cement, silica fume, and 0.45 w/cm

60.0

50.0

cm) 40.0 - Ω 30.0

20.0

Resistivity (K Resistivity 10.0

0.0 0 7 14 21 28 Days 0% SF 2% SF 8% SF Figure 6: Comparison of variation of percent SCM for mixtures prepared with Type I cement, silica fume, and 0.50 w/cm.

139

Table 2: Statistical comparison of mean 28-day resistivity values for mixtures of varying SCM percent content (none, low percentage and high percentage) prepared with Type I cement, silica fume and containing no admixture ANOVA Student T-test 0%/low%/high% 0%/low% 0%/high% Low%/high% 0.40 w/cm p-values 5.51E-17 2.19E-07 3.49E-12 3.25E-11 0.45 w/cm p-values 6.69E-18 3.82E-06 1.69E-13 5.16E-12 0.50 w/cm p-values 4.03E-18 2.16E-09 9.56E-13 5.58E-12

Effect of Slag Cement Addition 30.0

25.0

cm) 20.0 - Ω 15.0

10.0

5.0 Resistivity (K Resistivity

0.0 0 7 14 21 28 Days 0% SC 5% SC 40% SC Figure 7: Comparison of variation of percent SCM for mixtures prepared with Type I cement, slag cement and 0.40 w/cm.

30.0

25.0 cm)

- 20.0 Ω 15.0

10.0

Resistivity (K Resistivity 5.0

0.0 0 7 14 21 28 Days 0% SC 5% SC 40% SC Figure 8: Comparison of variation of percent SCM for mixtures prepared with Type I cement, slag cement and 0.45 w/cm.

140

30.0

25.0 cm) -

Ω 20.0

15.0

10.0 Resistivity (K Resistivity 5.0

0.0 0 7 14 21 28 Days 0% SC 5% SC 40% SC Figure 9: Comparison of variation of percent SCM for mixtures prepared with Type I cement, slag cement and 0.50 w/cm

Table 3: Statistical comparison of mean 28-day resistivity values for mixtures of varying SCM percent content (none, low percentage and high percentage) prepared with Type I cement, slag cement and containing no admixture ANOVA Student T-test 0%/low%/high% 0%/low% 0%/high% Low%/high% 0.40 w/cm p-values 3.30E-16 4.94E-04 4.72E-12 2.53E-11 0.45 w/cm p-values 5.56E-13 2.23E-04 2.48E-09 5.47E-09 0.50 w/cm p-values 9.67E-14 2.39E-05 1.23E-10 6.91E-09

DISCUSSION The following sections compare each mixture parameter (SCM type and percent replacement) to determine their influence on a resistivity results and whether they may impact the outcome of a test.

Effect of Class-C Fly Ash Addition At an early age, the resistivity value is lowest for samples with a high percent fly ash replacement (20%). The mixture containing no fly ash recorded the highest resistivity. However, it would seem that the curves converge towards day-28 as the mixtures containing a higher percentage of fly ash gain resistivity at a higher rate in comparison to that of the control mixtures containing no SCMs. Thus, after 28 days of continuous curing, it is not possible to distinguish the mixtures based on SCM content (Table 1).

Effect of Silica Fume Addition Looking at the early-age behavior, 1 to 7 days, there are no notable differences between mixtures containing silica fume and no SCM. Thereafter, initiation of pozzolanic reactions increases the rate of resistivity gain in time. The trend demonstrates an increase in resistivity gain with an increase in silica fume percent replacement.

141

Unlike the behavior seen for the fly ash mixture, the “convergence” effect occurs at a much earlier age. On day 1, the mixtures prepared with the SCM did record lower values than that of the mixtures containing no SCM. This was also seen for the fly ash mixtures. However, the gain in resistivity at an early age is more considerable creating a convergence within the first week of continuous curing. This behavior does not permit early age distinction but aids thereafter. As such, silica fume addition can be discernable based on its resistivity test result at 28-days. Moreover, based on the visual comparison of standard deviations between sample means demonstrated in Figures 4-6, there is a potential to differentiate the mixtures as early as 14 days of continuous moist curing.

Here, a minor addition of silica fume (2%) resulted in a significant increase in 28-day resistivity value: 49.6%, 63.7%, 48.9% percent change for the 0.40w/cm, 0.45 w/cm and 0.50 w/cm mixtures respectively. The increase in value was sufficient to statistically discern both mixtures from each other (Table 2). As for a high percentage replacement by weight (8%), the 28-day values substantially increased by 339.3%, 405.5% and 327.3% with respect to the increasing w/cm. This increase in resistivity demonstrates the benefits of silica fume addition through pore refining of the cementitious matrix.

Effect of Slag Cement Addition Mixtures fabricated with the slag cement behaved similarly to that of silica fume mixtures, where a noticeable “convergence” occurred at an early age, within the first few days of curing (Figures 7-9). Thereafter, there is an increase in resistivity in time but, not as prominent as that observed for the silica fume mixtures.

Based on the comparative results of the sample means recorded at 28-day shown in Table 3, a low percentage of slag cement replacement (5%) made a discernable impact on the measurement. Here, the relatively low increase in resistivity (11.1%, 22.0%, and 30.7% for the 0.40 w/cm, 0.45 w/cm, and 0.50 w/cm mixtures respectively) was sufficient to statistically discern between both mixtures. As previously stated for silica fume addition, the impact of low percentage replacement is significant making resistivity testing acceptable for distinguishing mixtures containing slag cement from mixtures containing no SCM.

Although not to the extent as that observed for the silica fume mixtures, 40% replacement with slag cement did contribute to increasing the 28-day resistivity value by 92.3%, 141.8% and 150.0% for the 0.40 w/cm, 0.45 w/cm and 0.50 w/cm mixtures respectively. Again, the benefits of SCM addition are well demonstrated with this increase in resistivity.

As seen in Table 3, the difference in 28-day resistivity value between the low and high percent is statistically different. Looking at the trend in resistivity gain over time, there may be a distinguishable behavior with respect to percentage of slag cement replacement. This behavior is also seen for the silica fume mixtures. Further research into the effects of percent SCM addition is recommended to aid in the development of a potential model for mixture design optimization and resistivity predictions.

CONCLUSION This study was conducted in order to understand the effect that common SCM replacement ratios will have on the surface resistivity results of a concrete sample as this is an important relationship to understand if surface resistivity is to be used in the field. It was found that the method is sensitive to supplementary cementitious material addition. In this study, the resistivity value increases in time for mixtures containing SCMs. However, it has been recommended to determine a more realistic resistivity potential of a mixture at a later age (56 to 91 days); thus, early-age testing may not be adequate for the purpose of classification of mixtures.

142

REFERENCES [1] Morris, W., Moreno, E.I. and Sagues, A. A., “Practical Evaluation of Resistivity of Concrete in Test Cylinders using a Wenner Array Probe”, Cement and Concrete Research, Vol. 26, No. 12, pp. 1779-1787, 1996 [2] Kessler, R. J., Powers, R. G., Vivas, E., Paredes, M. A., and Virmani, Y. P. (2008). “Surface Resistivity as an Indicator of Concrete Chloride Penetration Resistance.” Proc., Concrete Bridge Conference, St. Louis, Missouri, 20. [3] Rupnow, T., and Icenogle, P. (2012). “Surface Resistivity Measurements Evaluated as Alternative to Rapid Chloride Permeability Test for Quality Assurance and Acceptance.” Transportation Research Record: Journal of the Transportation Research Board (2290), 30-37. [4] Spragg, R. P., Castro, J., Nantung, T., Paredes, M., and Weiss, J. (2012). “Variability Analysis of the Bulk Resistivity Measured Using Concrete Cylinders.” Advances in Civil Engineering Materials, 1(1),1-17. [5] AASHTO TP 95. (2014). Standard Test Method for Surface Resistivity of Concrete’s Ability to Resist Chloride Ion Penetration. American Association of State Highway and Transportation Officials, Washington, DC, 10. [6] AASHTO TP 119, Standard Test Method for Electrical Resistivity of Concrete Cylinder Tested in a Uniaxial Resistance Test. American Association of State Highway and Transportation Officials, Washington, DC, 2017, 15. [7] ASTM C192-16a. (2016). Standard Practice for Making and Curing Concrete Test Specimens in the Laboratory. ASTM International, West Conshohocken, PA. [8] ASTM C33-18 (2018). Standard Specification for Concrete Aggregates. ASTM International, West Conshohocken, PA. [9] ASTM C150-18 (2018). Standard Specification for Portland Cement. ASTM International, West Conshohocken, PA. [10] ASTM C618-17a. (2017). Standard Specification for Coal Fly Ash and Raw or Calcined Natural Pozzolan for Use in Concrete, ASTM International, West Conshohocken, PA. [11] ASTM C1240-15. (2015). Standard Specification for Silica Fume Used in Cementitious Mixtures, ASTM International, West Conshohocken, PA. [12] ASTM C989/C989M-18. (2018). Standard Specification for Slag Cement for Use in Concrete and Mortars, ASTM International, West Conshohocken, PA. [13] ASTM C511-13. (2013). “Standard Specification for Mixing Rooms, Moist Cabinets, Moist Rooms, and Water Storage Tanks Used in the Testing of Hydraulic Cements and Concretes,” ASTM International, West Conshohocken, PA. [14] AASHTO T 358, Standard Method of Test for Surface Resistivity Indication of Concrete’s Ability to Resist Chloride Ion Penetration.

143

Laser Shearography and Ultrasound Inspection Laser Shearographyof Composite and UltrasoundLaminates withInspection Overlapping of Composite Fiber Plies Laminates with Overlapping Fiber Plies Sarah L. Stair, David G. Moore, Corinne Hagan and Ciji L. Nelson Sarah L. Stair, David G. Moore, Corinne Hagan and Ciji L. Nelson Sandia National Laboratories P.O. BoxSandia 5800, National Albuquerque, Laboratories NM 87185 (505)P.O. Box 844-4870; 5800, Albuquerque, email [email protected] NM 87185 (505) 844-4870; email [email protected]

ABSTRACT When manufacturing or repairing structures with composite laminate materials, internal features of overlapping fiber plies and ply orientation misplacement can occur. These composite processing steps will influence material characteristics of the component and likely change the inspection results. As part of the quality assurance process, nondestructive evaluation methods are needed for inspecting laminated composites to ensure the materials are undamaged and meet the design parameters required for the structure’s intended application. The present study considers laser shearography and roller probe ultrasound techniques for inspecting laminated composites with overlapping fiber plies. The samples used in this study were designed and manufactured such that two different laminate thicknesses and three designs of overlapping fiber plies were considered. Each of the samples was manufactured using an 8-harness satin carbon fiber weave fabric and a thermoset resin. The benefits and limitations of using these two nondestructive inspection methods for assessing carbon fiber reinforced laminated composites with internal ply overlaps will be presented.

Keywords: carbon fiber laminates, laser shearography, roller probe ultrasound inspection, ply overlaps

INTRODUCTION The use of carbon fiber reinforced laminated composites is increasing as manufacturers seek a high strength, low weight alternative to metals. By varying the ply stacking sequence, the material properties of the bulk laminate are rather tailorable [1]. The heterogeneous, complex nature of composite materials increases the difficulty of inspecting them for damage and monitoring their structural health. Damage, porosity and other anomalies can form within composites when they are manufactured, in-service and during repair processes [2]. For example, porosity within the as-manufactured part is affected by the cure cycle. If pressure is not applied to the laminate at the appropriate time of the cure cycle, increased porosity will be present within the final part [3-4]. Furthermore, the viscoelastic nature of composites enables barely visible impact damage to go visually unidentified as cracks and delaminations may exist internally while the surface of the part appears pristine [5]. Thus, nondestructive inspection methods are needed for the identification of damage and other anomalies within carbon fiber reinforced laminated composites.

Studies have evaluated the use of a variety of non-destructive evaluation (NDE) techniques for inspecting carbon fiber reinforced laminated composites for damage. Gholizadeh [6] provided an excellent review of literature which correlated specific NDE inspection methods to the identification of certain types of composite laminate damage. Additionally, Gholizadeh provided a brief description of each of the NDE methods considered. The present study focuses on the application of two ultrasound inspection methods (immersion and roller probe) and laser shearography for identifying changes in ply orientation and ply overlaps in carbon fiber reinforced laminated composites.

Several studies have applied ultrasound for the identification of damage within composite laminates (see e.g., [7, 8]). Although immersion ultrasound is a common method for inspecting composite laminates or coupons, it is not a feasible method for field use. The roller probe ultrasound inspection system combines a phased array ultrasound probe housed inside of a polymer wheelhouse. The outer surface of the polymer wheel rolls across the surface of the

sample with a couplant, such as water, applied to the sample surface. Jolly, et. al [8] applied this technique to the inspection of a thick walled composite but were unable to identify the backwall of the laminate due to signal attenuation. The present study compares the results obtained from immersion ultrasound and roller probe ultrasound and discusses the benefits and limitations of each technique.

The third inspection technique considered in the present study is laser shearography. For an in-depth description of the fundamentals of laser shearography, the authors suggest the ASNT Industry Handbook: Aerospace NDT [9]. In general, a stress is applied to the sample of interest, and the laser shearography system measures the surface displacement caused by the applied stress. Internal voids and anomalies react to the stress differently than the surrounding material and can be identified in the resulting phase map image. Multiple types of applied stress can be used, such as thermal [10], flexural waves [11], vibration and vacuum [9]. Thermal stress was the option selected for the present study.

EXPERIMENTAL OVERVIEW Three different laminate stacking sequences were evaluated in this study. Figure 1 presents the ply stacking sequences for samples one and two. The laminates were manufactured with tooling on one side and a vacuum bag on the other side. In reference to the ply stacking sequences shown, the tooling surface is located at the bottom of each ply stack. The solid red rectangles indicate an up ply orientation while the rectangles filled with blue diagonal lines indicate a down ply orientation, where “down” and “up” refer to physically flipping the ply down or up.

The sample one stacking sequence includes multiple butt joints throughout the sample thickness. At each joint in sample one, there is a change in the ply orientation meaning that a down ply is located next to an up ply in the same lamina of the bulk laminate. Sample two includes a ply overlap located at the center of the laminate stacking sequence. Multiple laminates were manufactured using the design of sample two with the overlap region, referred to

(a) (b) (c) Figure 1: Diagrams of the ply stacking sequences for (a) sample one, (b) sample two with 6 plies, and (c) sample two with 10 plies. (Solid red is up and blue diagonal lines is down ply orientation). as ‘x’ in Figures 1(b) and 1(c), having values of 6.35 mm, 12.7 mm, 19.1 mm, and 25.4 mm. Furthermore, there were two different thicknesses of sample two that were evaluated in this study, a 6 ply laminate and a 10 ply laminate. The ply stacking sequence for sample three is depicted in Figure 2(a) with a diagram of the top view and side view of sample three presented in Figures 2(b) and 2(c), respectively. Five different overlap lengths, referred to as ‘x’ in Figure 2(a), were used while manufacturing the sample three laminates: 0 mm (butt joint), 6.35 mm, 12.7 mm, 19.1 mm, and 25.4 mm. As seen in Figure 2(c), the type of ply overlap used in sample three is similar to that of a roof shingle design.

Each of these laminates was inspected using three nondestructive inspection methods: immersion ultrasound, roller probe ultrasound and laser shearography. A summary of the results from each inspection method is provided in the following sections. 145

(a) (b) (c) Figure 2: (a) Diagram of the ply stacking sequence for sample three. (b) A top view and (c) a side view of the sample three design, where the overlap, ‘x’, is greater than zero.

IMMERSION ULTASOUND INSPECTION Each of the three sample types was inspected using an immersion ultrasound inspection technique and were scanned from the tool side. For the inspection, the sample was fully submerged in a tank of water and scanned with a 0.5 MHz spherically focused ultrasound probe. The inspection data was collected and post-processed in Mistras UTwinTM software. Using the inspection data from sample one, the plies containing butt joints and changes in ply orientations can be identified by using a small gate width (approximately 0.20 µs). Figures 3(a) through 3(d) present the results for lamina 1, 3, 5, and 11, respectively in the sample one ply stacking sequence. Figure 3 provides a summary of the information that can be gained while inspecting sample one with immersion ultrasound but does not include all of the butt joints that are capable of being identified using this technique.

Figure 4 presents the immersion ultrasound inspection results for the 12.7 mm (0.5 in) and 25.4 mm (1 in) overlap present in the ten ply sample two laminate design. The red rectangular region in the lower left-hand corner of each C-scan amplitude plot is the sample label. From these inspection results, the width of the ply overlap located at the center of the laminate, with respect to both the planar dimension and along the laminate thickness, can be accurately measured using immersion ultrasound.

The immersion ultrasound inspection results for the sample three 12.7 mm (0.5 in) and 25.4 mm (1.0 in) overlap laminates are presented in Figure 5(a) and 5(b), respectively. As with the sample two results, the overlap region can be accurately identified and measured using the immersion ultrasound inspection technique. Although immersion ultrasound can successfully identify and measure these overlap regions, each sample must be fully submerged. Such an inspection is not always possible due to customer preferences or environmental exposure requirements for laminated composite parts. Thus, alternative nondestructive inspection methods should also be considered, and the results from roller probe ultrasound and laser shearography are presented in the following sections.

mm, 19.1 mm, and 25.4 mm. As seen in Figure 2(c), the type of ply overlap used in sample three is similar to that of a roof shingle design.

(a) (b) (c) Figure 2: (a) Diagram of the ply stacking sequence for sample three. (b) A top view and (c) a side view of the sample three design, where the overlap, ‘x’, is greater than zero.

146

(a) (b)

(c) (d) Figure 3: Identification of the internal ply variations for sample one. The resin-rich region between different laminate orientations can be identified for (a) lamina 1, (b) lamina 3, (c) lamina 5, and (d) lamina 11 as measured from the tooling side of the bulk laminate.

(a) (b) Figure 4: Immersion ultrasound inspection results for ten ply sample two laminate with (a) 12.7 mm (0.5 in) internal overlap, and (b) 25.4 mm (1.0 in) internal overlap. Axes are in units of inches.

ROLLER PROBE ULTASOUND INSPECTION The Olympus RollerFORMTM phased array wheel probe was the second nondestructive method used for inspecting these composite laminate samples. A 5 MHz, 64 element phased array wheel probe was used for the inspection. A spray of water on the sample surface coupled the RollerFORMTM probe to the composite laminate. The probe was then rolled across the surface of the laminate while it collected ultrasound data. Using the Olympus OmniPC software, the inspection data was post-processed using a narrow gate for the sample one inspection data. As seen in Figure 6, the location of butt joints and changes in ply orientation can be identified for lamina 1, 3, 5, and 11, just as they were located using the immersion ultrasound technique. The data presented in Figure 6 represents one pass of the roller probe across the surface of the sample one laminate. To cover the surface of the laminate, at least three passes of the roller probe should be used. Since the goal of this inspection is to identify the presence of the butt

147

(a)

(b)

Figure 5: Immersion ultrasound results for sample three with (a) a 12.7 mm (0.5 in) overlap and (b) a 25.4 mm (1 in) overlap. The samples were scanned from the tool side which meant the overlap was farthest away from the ultrasound probe, and the ultrasound signals were gated on the backwall of the 6 ply laminate. joints and changes in ply orientation, review of the data from one pass provides the desired information. Although the image data in Figure 6 appears to have more noise as compared to the immersion ultrasound technique, the RollerFORMTM probe has the benefits of portability and not requiring the part to be fully immersed.

The roller probe inspection results for the 12.7 mm (0.5 in) and 25.4 m (1.0 in) overlap sample two, ten ply laminates are presented in Figure 7(a) and 7(b), respectively. As seen in Figure 7, the presence of the overlap was difficult to identify near the edges of the laminate (left and right in the C-scan images). However, toward the middle of the inspection region, the presence of the ply overlap can be identified and measured. The edges of the overlap region are not as well-defined in the C-scan images in Figure 7 as they were when using the immersion ultrasound technique (see e.g., Figure 4), but the roller probe inspection was still capable of identifying the ply overlap.

The roller probe inspection results for the 12.7 mm (0.5 in) and 25.4 mm (1.0 in) overlap sample three laminates are presented in Figure 8(a) and 8(b), respectively. The overlap region is indicated by the light to dark blue regions in the center of the C-scan data. The edges of the overlap region appear to be feathered as compared to the edges of the overlap region as determined from the immersion ultrasound inspection. Although this may be viewed as a limitation of the roller probe inspection technique, the portability of the roller probe expands its applications as compared to that of the immersion ultrasound technique, and the roller probe inspection data demonstrated its ability to identify the overlap as well as provide a reasonable estimate of the overlap length.

148

(a)

(b)

(c)

(d)

Figure 6: Identification of the internal ply variations for sample one via roller probe ultrasound inspection. The changes in ply orientation are highlighted by the black dashed line and arrow for (a) lamina 1, (b) lamina 3, (c) lamina 5, and (d) lamina 11 as measured from the tooling side of the bulk laminate.

(a)

(b)

Figure 7: Roller probe inspection results for 10 ply sample two laminate with (a) 12.7 mm (0.5 in) overlap and (b) 25.4 mm (1 in) overlap.

149

(a)

(b)

Figure 8: Roller probe inspection results for sample three laminate with (a) 12.7 mm (0.5 in) overlap and (b) 25.4 mm (1 in) overlap.

LASER SHEAROGRAPHY INSPECTION The third nondestructive inspection method used for evaluating the three laminates was laser shearography. A LTI 2100 laser shearography system was used in conjunction with an applied thermal stress. The thermal stress was induced via two heat lamps with one lamp placed on either side of the shearography system and aimed at the center of the laminate surface. Each laminate was inspected from both the bag surface and the tool surface.

Sample one was inspected using this technique, but it was difficult to discern the butt joints and changes in ply orientation within the laminate since sheargraphy data cannot step through the thickness of the sample. Thus, results from the inspection of the sample one laminate are not included in this section.

The results from inspection of the bag surface of the 10 ply sample two laminate with a 6.35 mm (0.25 in), 12.7 mm (0.50 in), 19.1 mm (0.75 in) and 25.4 mm (1.0 in) overlaps are presented in Figure 9(a) through 9(d), respectively. Similar results were also obtained for the sample two 6 ply laminate design. A red horizontal bar and an arrow are used in each of the figures to highlight the overlap region. Using the shearography inspection software, the overlap region on each laminate was measured a couple of ways. The first method was using the linear measurement tool, which provided measurements within ± 1.3 mm (0.05 in) of the designed overlap length. The second measurement method was using the signal analysis tool within the LTI 2100 shearography software. Unfortunately, identification

(a) (b) (c) (d) Figure 9: Shearography inspection results of the bag side of sample two with (a) 6.35 mm (0.25 in) overlap, (b) 12.7 mm (0.5 in) overlap (c) 19.1 mm (0.75 in) overlap and (d) 25.4 mm (1 in) overlap.

150

of the overlap via the signal alone was difficult due to a low signal-to-noise ratio. Since the presence of the overlap could still visually be identified in the greyscale shearography image, the linear measurement tool was adequate for this particular application.

When the sample two laminates were inspected from the tool side, there was an increase in the surface reflection as compared to the bag side of the laminate, which increased the difficulty of inspecting the samples via shearography. As seen in Figure 10, identification of the overlap region for any of the four lengths (6.35, 12.7, 19.1, and 25.4 mm or 0.25, 0.50, 0.75 and 1.0 inch) was difficult from the tool side. While trying to optimize this inspection, multiple heat times and data collection times were evaluated. The images in Figure 10 represent the best images that were obtained from inspecting the sample two laminates from the tool side. Of these four laminates, only the 12.7 mm (0.50 in) overlap was identified. Decorrelation in the shearography data was observed near the top of the scan region, which was located farthest away from the shearography system since the sample was at an angle of approximately 18o relative to the camera. However, when the angle of the sample was decreased, a streaked region appeared in the results due to the laser’s reflection off of the smooth tool surface. Thus, as evidenced by Figures 9 and 10, the laminate surface finish affects the ability to identify internal ply overlaps. An approach to consider in the future is whether a thin layer of spray paint would assist in reducing the amount of surface reflection on the tool side of the laminate.

(a) (b) (c) (d) Figure 10: Shearography inspection results of the tool side of sample two with (a) 6.35 mm (0.25 in) overlap, (b) 12.7 mm (0.5 in) overlap (c) 19.1 mm (0.75 in) overlap and (d) 25.4 mm (1 in) overlap.

The results from inspecting the sample three laminates with 6.35 mm (0.25 in), 12.7 mm (0.50 in), 19.1 mm (0.75 in) and 25.4 mm (1.0 in) overlaps are presented in Figures 11 and 12. The results shown in Figure 11 correspond to the inspection performed on the bag surface of the laminate while the results in Figure 12 represent the results obtained from inspecting the tool side of the laminate. In each of these images, the overlap region can be clearly identified and measured. There is some signal decorrelation near the edges of the inspection region as evidenced by the speckled appearance in the image, but the area of interest containing the laminate overlap is easily identified since it is located near the center of the inspection region. The closer the feature of interest is to the center of the scan region, the more accurate the measurement of the feature will be because of its proximity to the shearography camera. Conversely, near the edges, the feature would be located at a greater angle relative to the shearography camera, which could prevent the measurement from being as accurate [9].

The results from these inspections indicate that the ability to identify the presence of the overlap in the sample two design is different than the ability to identify the overlap in the sample three design. The authors hypothesize that due to the sample three laminates essentially being two laminates overlapping each other, the effects of the displacements caused by the applied thermal stress are greater as compared to having a single lamina overlapping another lamina in the center of the bulk laminate (e.g., the sample two design).

151

(a) (b) (c) (d) Figure 11: Shearography inspection results of the bag side of sample three with (a) 6.35 mm (0.25 in) overlap, (b) 12.7 mm (0.5 in) overlap (c) 19.1 mm (0.75 in) overlap and (d) 25.4 mm (1 in) overlap.

(a) (b) (c) (d) Figure 12: Shearography inspection results of the tool side of sample three with (a) 6.35 mm (0.25 in) overlap, (b) 12.7 mm (0.5 in) overlap (c) 19.1 mm (0.75 in) overlap and (d) 25.4 mm (1 in) overlap.

CONCLUSIONS AND FUTURE WORK Three non-destructive inspection techniques were used for evaluating three sample designs that contained butt joints, changes in ply orientation and ply overlaps. Both the immersion ultrasound and roller probe ultrasound techniques identified the butt joints and ply orientation changes in the sample one design and the ply overlap regions of the sample two and sample three designs. The scan quality of the immersion ultrasound C-scans was better than that of the roller probe ultrasound C-scan images. However, there are two key benefits for using the roller probe as compared to the immersion ultrasound approach. First, the part must be fully submerged in water for the immersion ultrasound technique, which is not always a feasible option depending on part size and material handling restrictions. The roller probe inspection method requires a couping medium, such as water, be applied to the part surface but does not require the part to be submerged. The second benefit of using the roller probe inspection technique is the portability of the roller probe as compared to the immersion ultrasound system. The immersion system is tied to the presence of an immersion tank, but the roller probe equipment can be hand carried to field inspections.

The third nondestructive inspection technique evaluated in this study was laser shearography. Unfortunately, this technique was unable to identify the butt joints and changes in the ply orientation in the sample one laminate, but this application is not necessarily the intended use for a shearography system. Rather, the goal of shearography inspections is to identify internal damage and anomalies through measurement of surface displacements caused by

152

an applied stress. The results obtained for the sample two and sample three inspections indicated that shearography could identify the ply overlaps in these two sample designs. However, when inspecting the sample two laminates form the tool side, as seen in Figure 10, identification of the ply overlap was difficult with only the 12.7 mm (0.50 in) overlap successfully identified. The authors hypothesize that the dramatic overlap in the sample three design, where one laminate is overlapped on top of another laminate, contributed to the ability to successfully identify the overlap from both sides of the sample three laminates as seen in Figures 11 and 12.

Future work in this area will consider diverse types of internal overlap joints in carbon fiber reinforced and fiber glass reinforced laminated composites. Additionally, curved samples rather than flat laminate plates will also be considered for future studies in this area.

ACKNOWLEDGEMENTS Sandia National Laboratories is a multimission laboratory managed and operated by National Technology & Engineering Solutions of Sandia, LLC, a wholly owned subsidiary of Honeywell International, Inc., for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-NA0003525.

This paper describes objective technical results and analysis. Any subjective views or opinions that might be expressed in the paper do not necessarily represent the views of the U.S. Department of Energy or the United States Government.

REFERENCES [1] Barbero, E.J., 2010, Introduction to Composite Materials Design, second edition, Taylor & Francis, Boca Raton, FL. [2] Adams, R.D. and P. Cawley, 1989, “Defect Types and Non-Destructive Testing Techniques for Composites and Bonded Joints,” Construction & Building Materials, 3(4), pp. 170-183. [3] Guo, Z., L. Liu, B. Zhang and S. Du, 2009, “Critical Void Content for Thermoset Composite Laminates,” Journal of Composite Materials, 43(17), pp. 1775-1790. [4] Liu, L., B. Zhang, D. Wang. and A. Wu, 2006, “Effects of Cure Cycles on Void Content and Mechanical Properties of Composite Laminates,” Composite Structures, 73, pp. 303-309. [5] Burkov, M., P. Lyubutin, A. Byakov and S. Panin, 2017, “Detecting Barely Visible Impact Damages of Honeycomb and Laminate CFRP Using Digital Shearography,” Proc. Of the Intl. Conf. on Adv. Materials with Hierarchical Structure for New Technologies and Reliable Structures, 1909, pp. 02022-1 – 02022-4. [6] Gholizadeh, S., 2016, “A Review of Non-Destructive Testing Methods of Composite Materials,” XV Portuguese Conference on Fracture, 1, pp. 050-057. [7] Amaro, A.M., P.N.B. Reis, M.F.S.F. de Moura and J.B. Santos, 2012, “Damage Detection on Laminated Composite Materials Using Several NDT Techniques,” Insight: NDT and Cond. Monitoring, 54(1), pp. 14-20. [8] Jolly, M.R., A. Prabhakar, B. Sturzu, K. Hollstein, R. Singh, S. Thomas, P. Foote and A. Shaw, 2015, “Review of Non-Destructive Testing (NDT) Techniques and Their Applicability to Thick Walled Composites,” Procedia CIRP, 38, pp. 129-136. [9] Bossi, R.H., 2014, ASNT Industry Handbook: Aerospace Nondestructive Testing, ASNT, Columbus, OH. [10] Burkov, M., P. Lyubutin, A. Byakov and S. Panin, 2015, “Development of High Resolution Shearography Device for Non-Destructive Testing of Composite Materials,” Proc. Of the Advanced Materials with Hierarchical Structure for New Technologies and Reliable Structures, 1683, pp. 02029-1 – 02029-5. [11] Lamboul, B., O. Giraudo, and D. Osment, 2015, “Detection of Disbonds in Foam Composite Assemblies using Flexural Waves and Shearography,” 41st Annual Review of Progress in Quantitative Nondestructive Evaluation, 1650, pp. 1155-1161.

153

ProcessProcess Monitoring monitoring at at Capacitor capacitor Dischargedischarge weldingWelding

Jörg Zschetzsche11,, MaxMax-Martin-Martin Ketzel1, Uwe Füssel1, Hans-JürgenHans-Jürgen Rusch2, Nicolas Stocks2

1 1TechnischeTechnische UniversitätUniversität DresdenDresden Chair of Joining Technology and Assembly DD-01062-01062 Dresden +49 351 463 3530735307;; email [email protected]@tu-dresden.de-dresden.de

2 2KKAPKAPKΩΩNN GmbHGmbH Industriestrasse 64a DD-28876-28876 Oyten +49+49 4207 4207 987 987 8588; 8588 email; [email protected] [email protected]

ABSTRACT The quality of welds can be proven by non-destructive or destructive testing. According to the state of the art, these quality checks are expensive and lower the economy and productivity. Hence, only on a random sample with few components, compared to the amount of all manufactured components, are tested in intervals. However, process monitoring can be used with high economy and low cost for a continuous quality assessment of all components. The example of resistive projection welding with capacitor discharge (CD-welding) shows how the process sequence is designed on the basis of a physically justified process understanding and how the monitoring parameters are selected. CD-welding belongs to the resistance welding methods in which the connection is made by pressing the components and simultaneously melting a projection. According to recent researches, the joint is formed differently at cd welding. Due to very high power density within very short period of time metal vaporizes and activates the surfaces. The joint occurs in less than a few milliseconds, when the activated surfaces are pressed together while the projection is plastically deformed. At the same time, molten material is pushed out of the joint plane. On this new basis of process understanding, a new machine has been developed in order to control the process purposefully and to monitor the stages of the process.

Keywords: capacitor discharge welding, projection welding, process monitoring,

INTRODUCTION Resistance welding is very widespread in industrial production and has got numerous variants. There are many examples for applications: welding of sheet metals, joining elements like welding nuts, rifles and bolts for the automotive industry or joining of contact elements, stranded wires and single wires for the electrical industry.

Capacitor discharge welding (CD welding) is a special variant of resistance welding. During CD welding, the energy of a charged capacitor, coupled via transformer, is discharged to the components. There are the following special properties [1]:  The result of the discharge is a high current pulse within a very short welding time (a few milliseconds) which leads to very low heat effect input in the components.  The required power rating of the electrical grid is significantly lower than the welding power.  The discharge process is very repeatable, because it is determined by the parameters of the electrical circuit.

Process monitoring at capacitor discharge welding

Jörg Zschetzsche1, Max-Martin Ketzel1, Uwe Füssel1, Hans-Jürgen Rusch2, Nicolas Stocks2

1 Technische Universität Dresden Chair of Joining Technology and Assembly D-01062 Dresden +49 351 463 35307; email [email protected]

2KAPKΩN GmbH Industriestrasse 64a D-28876 Oyten +49 4207 987 8588; [email protected]

Keywords: capacitor discharge welding, projection welding, process monitoring,

Therefore, CD welding is used in areas of application which are not or only partially covered by other welding processes. Examples of this are welding of:  Nuts and bolts on high-strength, coated steel sheet metals,  steels with high carbon equivalent,  sintered components,  ring projections up to 200 mm diameter,  different combination of materials,  high conductive materials and  finished components without welding distortion [1].

Process monitoring, based on a comprehensive knowledge of the process, is the basis for a reliable application of CD welding.

CHARACTERISATION OF THE PROCESS The process runs in four characteristic phases, which merge smoothly into each other [2], Figure 1:

Contacting Contacting is characterized by  movement and contact of electrodes to the components,  upslope of the electrode force to the welding force (force build-up),  plastic deformation in the contact area,  shaping of the contact surface (apparent and actual).

Due to the local varying surface pressure and the local varying relative movement, the contact resistance varies locally, too, which leads to varying degrees of heating later.

Activation Activation is characterized by  current flow with very high current gradients, until shortly before the maximum current and  very high power density in the contact area.

Metal vaporizes in the contact area, foreign and oxide layers are entrained with the expanding metal vapour (micro expulsions), due to this the surface is activated. The current density is not distributed constantly over the contact surface. Due to the geometric of the projection, the current density is higher at the edge of the welding zone. As a result the power density at the edge is bigger, too. The metal vaporisation starts at the point of the maximum power density. At the edge of the projection metal vapour leaks under atmosphere pressure whereby the electrical conductive contact area decreases, as the metal vapour has very low electric conductivity. The current flows through the remaining contact surface, which is activated successively. The volume of the projection is heated marginally. The electrodes do not follow-up during this phase.

Material closure Material closure is characterized by  starting follow-up movement of the electrodes,  pressing the activated surfaces together, thereby occurring material closure and eliminating the contact resistance,  decreasing current and current density,  decreasing metal vaporisation,  conductive heating of the projection volume,  decreasing projection strength and starting plastic deformation of the projection.

155

Repressing Repressing is characterized by  follow-up of the electrodes,  decreasing current,  plastic deformation of the projection and resulting material closure in non-activated areas, liquid material is pressed out of the welding zone  enlargement of the joint,  ending follow-up movement of electrodes,  heat dissipation and cooling down of the welding zone.

Figure 1: Process course

CD welded joints differ from conventional projection welded joints. Very high power densities (over 106 W/cm²) occur extremely fast in the welding zone. This matches to the power density which causes the formation of a keyhole (caused by metal vaporisation) during laser beam welding. Therefore, the joint is not formed by dilution of molten phases, but by pressing activated surfaces together and plastic deformation of the projection. The result is a joint without a welding nugget. The quality of the joint cannot be reliably evaluated in cross sections (Figure 2).

156

Figure 2: Cross section of a CD welded joint

CD welding with one current pulse is common. The course of welding current is essentially determined by design of the machine (capacity of capacitors, transmission ratio of transformer and inductance of electric welding circuit). The course of the welding current could only be easily modified by the charging voltage of the capacitors. The Multi-Capacitor-Source system (MCS) can extend these limits. With four capacitors in parallel connection it is possible to fit the course of the welding current to the joining task and the process phases [3] [4]. For example, the peak current can be increased as well as reduced (Fehler! Verweisquelle konnte nicht gefunden werden.).

Figure 3: Welding current of MCS welding process

157

QUALITY ASSURANCE There are three methods for quality assurance: 1. quality inspection after the process has been completed, 2. process monitoring and 3. process control.

Quality inspection after the completed process Non-destructive testing (NDT) of projection welds has not been established yet, but is used almost exclusively on spot welds [5]. Destructive-testing is used for projection welds, although evaluations of cross sections are difficult. DT causes high cost as the components cannot be used after testing. Therefore spot checks are only possible.

Process control Due to the process characteristics, process control is not possible.

Process monitoring The process parameters are recorded, analysed and archived for monitoring. They imply information of the quality of the welded joint. The mechanical process parameters electrode movement and (less common) electrode force as well as the electrical process parameters welding current and transition voltage between the electrodes are monitored and compared with reference courses. In the case of deviations the welded components are declared as rejects. Furthermore, it is possible to define safety latencies for the capacitor charging and starting welding process. If these times differ from point of reference, the process is interrupted and an error is issued. The time-independent analysis of parameters in phase spaces (e.g. current/voltage or mechanical work/electrical power) is not common yet. Compared to usual time-based interpretations of the process parameters, the state of the process course in the phase space can be better characterized and limit values can be defined.

Process monitoring using the example of weld nuts Weld nuts are a typical application for CD welding. The quality of the joint is tested destructively by the normal tensile test. Additionally, damage of the thread because of expulsions is rated. As an example Figure 4 shows the results of the tensile test of a test series with 100 weld specimens of galvanized M6 weld nuts with ring projection to galvanized steel HDT700 [6]. Nuts with a normal tensile force less than approx. 6 kN (double standard deviation).Furthermore, a transition range was defined, which starts at approx. 8 kN (single standard deviation).

Figure 4: CD welded steel sheet material–weld nut M6, results of the tensile test

There are various causes of faults, e.g. different component geometries, different component surface quality, wrong inserted nuts or worn electrodes. Machine faults are rare. Single methods are unable to detect all defect. The combination of monitoring of several parameters is productive.

158

Static monitoring of the electrode motion is essential. After the electrodes have been closed, the position of the electrode is checked to compare the height of inserted components with guideline (component check). After welding, the electrode path is checked. A nominal-actual comparison of the charging voltage of the capacitors before ignition of the welding current is common. The process parameters peak current, current-time integral, root mean square of current, current rise time, welding time and current flow time are used individually or in combination to evaluate the process. An error is issued when the monitored value falls below a lower limit or exceeds an upper limit. Similarly, it is possible to calculate mean courses of time-based or time-independent courses. Deviations from the mean courses indicates welding faults.

In the tests mentioned above (see Figure 4), not all faulty welds were detected by monitoring the electrode path. With one exception (specimen 51, Fig. 4) all bad welds with a resulting normal tensile force, which is lower than the mean normal tensile force minus twice as much of the standard deviation, is detected by comparison of the actual electrode path course with the mean electrode path course of all welds. With the additional evaluation of welding current and voltage, the remaining faulty weld was detected. The thread of this was damaged by expulsions between the upper electrode and the nut, too. There was another sample with destroyed thread (specimen 72, Fig. 4), which was detected with this method, too. The time course of the electrode path and the current-voltage phase space are shown in Figure 5. The plots show the mean value of 100 welds, a good weld (specimen 95) and a weld with expulsions in the thread of the weld nut. The occurrence of expulsions is not detectable by the electrode path, but by the increased voltage. The significant different course in the U-I phase space indicates the expulsions during the weld.

Figure 5: Time-based electrode distance course and U-I-phase space

SUMMARY

CD welding is almost always used for projection welding. It is common to assure the quality by destructive testing of spot checks. Effective process monitoring based on fundamental process knowledge can reduce the number of samples, which has to be destroyed, significantly. It is necessary to evaluate several time-based parameters or parameters in phase spaces.

159

REFERENCES

(1) Merkblatt DVS 2911. Juli 2014. Kondensatorentladungsschweißen – Grundlagen, Verfahren und Technik (2) Füssel, U.; Ketzel, Max-Martin; Zschetzsche, Jörg: Erwärmungsverhalten der Kontaktzone beim Kondensatorentladungsschweißen unter Berücksichtigung der dynamischen Stromänderung und des Nachsetzverhaltens der Elektroden. Schlussbericht IGF-Nr. 18.987 BR/DVS-Nr. 04.069 : Technische Universität Dresden, Professur Fügetechnik und Montage, 2018 (3) Rusch, Hans-Jürgen: Neue Möglichkeiten der Prozessgestaltung beim KE-Schweißen. In: DVS Media GmbH 2016 – Widerstandsschweißen Sondertagung in Duisburg. Düsseldorf : DVS Media, 2016 (DVS-Berichte, 326), S. 201–210 (4) Rusch, Hans-Jürgen: Prozesstechnische Anforderungen an und elektrotechnisches Design von Kondensatorentladungsmaschinen. TU Dresden, Fakultät Maschinenwesen. Dissertation. 2018-11-04 (5) DVS Media GmbH 2016 – Widerstandsschweißen Sondertagung in Duisburg. Düsseldorf : DVS Media, 2016 (DVS-Berichte 326) (6) Khosravi, Farhad: Qualitätsbewertung von KE-Buckelschweißungen durch Prozessüberwachung. Dresden, Technische Universität Dresden, Professur für Fügetechnik und Montage. Diplomarbeit. 2017-05-15

160 AUTHOR INDEX

Aguilar-Ortega, Miguel ...... 69 Hagan, Corinne...... 144

Ali, S...... 78 Hartell, Dr. Julie Ann ...... 94, 134

Bedekar, Vikram ...... 2 Haszler, Alfred...... 52

Belanger, Yohan ...... 9 Hernandez, F...... 78

Bunn, Jeff ...... 2 Hyde, R. Scott...... 2

Burra, Sandeep G...... 20 Jiménez-Garrido, Jesús Antonio...... 69

Chaplan, S...... 86 Karunaratne, Evan ...... 94

Chitti, Abhishek...... 20 Ketzel, Max-Martin...... 154

Chu, Tsuchin P...... 20 Kolay, Prabir...... 20

Clares-Crespo, César...... 69 Kumar, Sanjeev...... 20

Diaz, K...... 86 Kwan, Chi-Hang ...... 104

Delatte, Dr. Norbert ...... 94 Lareau, J. P...... 78

Dunford, T...... 86 Lepage, Benoit ...... 104

Füssel, Uwe...... 113, 154 Lopez, B...... 78

Farhangdoust, Saman...... 30 Maldague, Xavier P. V...... 9

García-Benavides, Víctor...... 69 Masoumi, Masoud...... 59

García-Gómez, Joaquín...... 69 Mathiszik, Christian ...... 113

Garg, Vaibhav...... 42 Mehrabi, Armin...... 30

Ghaziary, Hormoz...... 52 Moore, David G...... 144

Gil-Pita, Roberto...... 69 Nelson, Ciji L...... 144

Giles, Ryan Kent...... 59 Painchaud-April, Guillaume...... 104

Glass, S. W...... 78 Parker, David H...... 125

Goldfine, N...... 86 Perron, Luc...... 9

161 Romero-Camacho, Antonio...... 69 Stocks, Nicolas ...... 154

Ross, K. S...... 78 Voothaluru, Rohit ...... 2

Rusch, Hans-Jürgen...... 154 Washabaugh, A...... 86

Shults, Cody ...... 134 Zschetzsche, Jörg ...... 113, 154

Stair, Sarah L...... 144

162