Experimental Investigation of Microparticle Sand Sticking Probability from 1000°C to 1100°C

Andrew James Boulanger

Dissertation submitted to the faculty of the Virginia Polytechnic Institute and State University in partial fulfillment of the requirements for the degree of

Doctor of Philosophy In Mechanical Engineering

Wing F. Ng Srinath V. Ekkad Todd Lowe Walter O’Brien Gary R. Pickrell

October 30, 2017 Blacksburg, Virginia, USA

Keywords: Dust Ingestion, Deposition, Arizona Road Dust, Sticking Probability

Experimental Investigation of Microparticle Sand Sticking Probability from 1000°C to 1100°C Andrew James Boulanger

ABSTRACT

Increasing commercial and military aircraft operations in arid environments are increasing the likelihood of sand and dust ingestion. Turbine engines are particularly susceptible to the ingestion of sand and dust, which can erode cold-section components and deposit onto hot-section components. Ultimately, the erosion and deposits will shorten the operational lifespan of these engines and limit their availability thereby increasing maintenance costs and risking safety. Mitigating these risks has become more prevalent in recent years due to increasing combustion temperatures in effort to increase fuel efficiency. Increasing combustion temperatures directly increases deposit formation onto hot-section components. Monitoring deposit formation on existing turbine engine platforms and improving deposit resilience on new designs has been the industry focus for the last two decades. This study focused on statistically modeling the initial onset of microparticle deposits onto an analogous hot-section surface. Generally, as deposits accumulate onto a hot-section surface, the existing deposit formation is more likely to bond with incoming particulate at a faster rate than an exposed bare surface. Predicting the initial deposits onto a bare surface can determine the accelerated deposition rate depending on subsequent particulate impinging onto the surface. To emulate the initial deposits, a HASTELLOY® X test coupon was exposed to 20 µm to 40 µm samples of Arizona Road Test Dust (ARD) at varying loadings and aerosol densities. The Virginia Tech Aerothermal Rig was used for all test scenarios at flow-particle temperatures between 1000°C to 1100°C. Several statistical models were developed as a function of many independent variables, culminating with a final sticking probability (SP) model. Overall, the SP of individual ARD particulate is a primary function of flow-particle temperature and normal impact momentum. Tangential impact momentum of a particle will decrease the SP, while surface temperatures reaching isothermal conditions with the flow will increase SP. However, there are specific cases where lower surface temperatures and high particle temperatures result in a high SP. Particle size was a strong predictor of SP where particles between 10 µm to 19 µm were 5 to 10 times greater than the 19 µm to 40 µm range. Additional studies will be necessary to examine some additional parameters that become more prominent with smaller particle sizes. Ultimately, the intention of the models is to assist turbine engine designers to improve resilience to deposit formation on hot-section components.

Experimental Investigation of Microparticle Sand Sticking Probability from 1000°C to 1100°C Andrew James Boulanger

GENERAL ABSTRACT

Dust ingestion by propulsion turbine engines can have severe negative implications on the operational safety of an aircraft. Recently, increased air traffic, both military and commercial, in desert regions has caused many aircraft engine designers to improve the resilience to dust ingestion effects. One of the detrimental mechanisms is hot particle deposits in the combustion and exhaust sections. This dissertation evaluates deposit formation using carefully developed high temperature experiments. In general, deposit formation can negatively change flow characteristics inside the engine that can limit available power and safety margins. Likewise, deposits can reduce or stop cooling needed for hot-section parts inside a jet engine. Hot-section components need cooling since the main gas path operation temperatures of a jet engine typically exceed the melting points of common high temperature metals. During dust ingestion events, deposits will initially adhere to a hot metallic or ceramic surface inside the engine. Subsequent deposit accumulation will occur at a faster rate since incoming particles will more readily adhere to existing deposits than to a metallic or ceramic surface. The experimental work in this dissertation focused only on quantifying the initial individual particle deposits on a HASTELLOY®-X surface between 1000°C to 1100°C. Arizona Road Dust was the particulate selected for all testing. The dust has sizes ranging between 10 µm to 40 µm. The sticking probability or the likelihood a particle would deposit per an impact was less than 5% for all tests performed. Particles smaller than 19 µm had a sticking probability up to 5% while larger particles were generally less than 3%. Effectively, this implies that the initial deposits onto a hot engine surface are strongly dependent on the smallest particles. Propulsion turbine engine designers can utilize this information to develop mitigation methods against deposit formation of the smallest particles that are ingested. Ultimately, the research presented in this work is intended to improve operational safety of current and future aircraft.

PREFACE

Deposit formation on turbine hardware in propulsion turbine engines can occur in many arid regions globally. Characterizing crystalline deposits on metallic substrates can aid in component resilience and health monitor algorithms during particle ingestion. This dissertation is written in manuscript format and contains three primary papers with an additional chapter intention of the analysis to be submitted to the ASME Journal of Turbomachinery. The author is the first author for all three papers and the final chapter presented in this dissertation. There is an Appendix which contains supplementary analytical information regarding the last chapter that was not deemed necessary for publication purposes. The first paper, presented in Chapter 1, was for the 2016 ASME Turbo Expo in Seoul, South Korea [1]. It was a preliminary investigation of Arizona Road Dust (ARD) deposits onto a HASTELLOY® X (HX) test coupon using the Virginia Tech Aerothermal Rig (VTAR). The investigation was able to establish a multi- linear regression of deposition as a function of flow temperature and test coupon angle. The resulting correlation had an of 0.67, which implied that the remaining 0.33 could be explained by other factors such as coupon temperature and injection rates. The experimental setup impinged combusted air and 20-40 µm ARD particles onto the test coupon for a variety of angles and temperatures. Gas temperatures ranged from 975°C to 1075°C with a constant bulk velocity of 70 m/s. Coupon angles were varied between 30°, 50°, 80°, and 90°. The regression developed indicated that ARD deposition increased linearly from 975°C to 1075°C for all coupon angles. Overall, this study was a preliminary investigation to establish individual ARD particle deposit response at high gas temperatures. Chapter 2 contains the second paper that was published in The Aeronautical Journal 2017 by the Royal Aeronautical Society [2]. It expanded on the first paper’s analysis by performing tests with a consistent ARD injection rate and incorporating surface temperature instead of gas path temperature. Similar to the previous study [1], coupon angles were varied between 20°, 50°, and 80° for flow temperatures of 1000°C, 1050°C, and 1100°C. Averaged deposits were methodically quantified through normalized particle deposit tallies per area and Coverage Ratio (CR) or percent coverage of the surface using microscopic imaging and object recognition scripts. Multi-linear regression models of bulk deposits were developed as a function of coupon angle and flow temperature. The multi linear regression models had values between 0.96 to 0.99 depending on the deposition metric employed. The prediction models can be used to estimate deposit accumulation on similar scenarios in gas turbine applications. Chapter 3 contains the third paper that was presented at the 2017 ASME Turbo Expo in Charlotte, North Carolina, USA [3] in conjunction with a complimentary paper by Barker et. al. [4]. Compared to the second paper [2], this analysis used similar test conditions using ARD and HX but evaluated deposits as a function of local aerothermal conditions on the test coupon and particle impact trajectories. All experiments use 20 µm to 40 µm ARD on a bare HX coupon from 1000°C to 1100°C bulk flow temperature with a constant flow velocity of 70 m/s. As previously indicated, a multi linear regression was fit to CR data as a function of surface temperature and impact trajectories. The resulting value of the model is 0.855. The model is able to predict the initial onset of deposits as a quadratic function of local surface temperature and impact velocity components (normal and tangential). A prominent observation was that tangential impact velocity has a significant nonlinear, independent effect on deposits relative to normal impact velocity and local surface temperatures. The final CR model can be used to predict when rapid deposit accumulation may begin under similar conditions tested. The final chapter is a paper intended to be submitted for journal for the 2018 ASME Journal of Turbo Machinery and the 2018 ASME Turbo Expo in Oslo, Norway. Compared to the previous studies, this study

iv quantifies the Sticking Probability (SP) of individual particles for 10 µm to 40 µm ARD for flow temperatures between 1000°C to 1100°C. Two relatively strong non-linear regressions were developed to the test data, the first uses raw dimension parameters and the second uses non-dimensional terms that are unique to this study. The resulting values range from 0.72 to 0.88 depending on the model and piecewise regression component. In general, SP increased at a quadratic rate as a primary function of particle temperature and normal impact velocity. Increasing tangential velocity and decreasing surface temperature reduced the expected SP. However, there were cases where a low surface temperature combined with a high gas path temperature resulted in a relative high SP. In addition, particle size had a significant effect on SP. Particles between 19 µm to 40 µm had a SP range up to 0.01 while smaller particles between 10 µm to 19 µm would reach approximately 0.05. In particular, the smaller particle sizes should be examined in a future study since the distinctly larger SP could be explained by additional parameters. Ultimately, the SP models developed can be used to predict the SP of 10 µm to 40 µm ARD onset of deposition under similar hot-section temperature conditions. The models are intended to provide valuable experimental data for turbine engine designers to develop innovative solutions to deposition due to dust ingestion. [1] Boulanger, A. J., Patel, H. D., Hutchinson, J., DeShong, W., Xu, W., Ng, W. F., and Ekkad, S. V., 2016, “Preliminary Experimental Investigation of Initial Onset of Sand Deposition in the Turbine Section of Gas Turbines,” GT2016-56059, ASME Turbo Expo 2016, Volume 1: Aircraft Engine; Fans and Blowers; Marine, ASME, Seoul, South Korea, p. V001T01A003. [2] Boulanger, A., Hutchinson, J., Ng, W. F. F., Ekkad, S. V. V, Keefe, M. J. J., Xu, W., Barker, B., and Hsu, K., 2017, “Experimental Investigation of the Onset of Sand Deposits on Hastelloy-X between 1,000°C and 1,100°C,” The Aeronautical Journal, 121(1242), pp. 1187–1199. [3] Boulanger, A. J., Hutchinson, J., Ng, W. F., Ekkad, S. V., Keefe, M. J., Xu, W., Barker, B. J., and Hsu, K., 2017, “Experimental Based Empirical Model Of The Initial Onset Of Sand Deposits On Hastelloy-X From 1000°C To 1100°C Using Particle Tracking,” GT2017-64480, ASME Turbo Expo 2017: Turbomachinery Technical Conference and Exposition, ASME, Charlotte, North Carolina, United States, p. V02DT48A015-V02DT48A015. [4] Barker, B. J., Hsu, K., Varney, B., Boulanger, A., Hutchinson, J., and Ng, W. F., 2017, “An Experiment-Based Sticking Model for Heated Sand,” GT2017-64421, ASME Turbo Expo 2017, ASME, Charlotte, North Carolina, United States, pp. 1–11.

ACKNOWLEDGEMENTS

I am extremely grateful to all members of my committee for their patience, encouragement, and support in overcoming the numerous obstacles encountered during the course of this research. I am especially grateful to Dr. Ng and Dr. Ekkad for their expertise, troubleshooting advice, and mentorship during this investigation. I am deeply grateful to both of these committee members as both mentors and friends. During these past three years, I have learned more than I can imagine and I owe much to Dr. Ng and Dr. Ekkad. A great thanks to Dr. Lowe, Dr. O’Brien, and Dr. Pickrell for having their door open to discuss ideas and possible solutions to challenges encountered during this project. I would also like to thank the support from Brett Barker and Kwen Hsu from Rolls Royce for their technical insight. I would also like to thank the numerous masters and doctoral students working at APPL Laboratory for their continual assistance and ideas for moving this project forward. I would especially like to acknowledge David Gomez, Siddhartha Gadiraju, John Hutchinson, and Suhyeon Park for their assistance building and operating the combustion test cell for the first two years at the lab. Within the last year, Edward Turner, Vy Nguyen, and Renzo La Rosa have been a great source of assistance for test operation and troubleshooting. It has truly been a privilege working with all of you these past few years. To my other colleagues within APPL, the Turbo Lab, and the HEFT lab including Raul Otereo, David Mayo, Ridge Sibold, Luke Luehr, Tamara Guimarães and Kyle Daniels; I appreciate the time you managed to carve out to help despite being busy with your own projects. I would also to recognize several members of the ME department who have supported this project and my degree including Jamie Archual, Ben Poe, Timothy Kessinger, Phillip Long, Bill Songer, Kimberly Clark, Cathy Hill, and Brandy McCoy. I would like to especially thank Diana Israel for those last-minute purchase orders and our drawn out conversations as a distraction from a busy day. Finally, I must express my deepest gratitude to my family and Krista for their unwavering support and encouragement during this entire journey. This great accomplishment would have not been possible without you by my side. I cannot emphasize how special it is to have a family that has always encouraged me to follow my dreams and never give up, even when things felt completely hopeless. I would not be the man I am today without all you have done for me. Thank you for all the late night phone calls and being willing to discuss my research without complaint. To Krista, thank you for all you have done for me these past few months. Without the food, constant supply of coffee, the laughs, and the companionship, this degree would not be possible. Thank you for putting up with all the insanity especially when I was nearing the completion of this degree.

CONTENTS

Abstract ...... ii General Abstract ...... iii Preface ...... iv Acknowledgements ...... vi List of Figures ...... x List of Tables ...... xiii 1 ASME Turbo Expo 2016: Preliminary Experimental Investigation Of Initial Onset Of Sand Deposition In The Turbine Section Of Gas Turbines ...... 1 Abstract ...... 1 Nomenclature ...... 1 1.1 Introduction ...... 2 1.2 Background ...... 2 1.3 Experimental Method ...... 4 1.4 Results ...... 9 1.5 Conclusions ...... 14 1.6 Acknowledgements ...... 14 1.7 References ...... 14 2 ISABE 2017: Experimental Investigation of the Onset of Sand Deposits on Hastelloy-X between 1000°C and 1100°C ...... 17 Abstract ...... 17 Nomenclature ...... 17 2.1 Introduction ...... 18 2.2 Experiment Method...... 19 2.2.1 Virginia Tech Aerothermal Rig ...... 19 2.2.2 Experiment Testing Conditions ...... 21 2.2.3 Statistical Modelling Method ...... 22 2.3 Results and Analysis ...... 23 2.3.1 Data Acquisition and Reduction ...... 23 2.3.2 Raw Data ...... 24 2.3.3 Empirical Models for Prediction ...... 26 2.4 Conclusion ...... 27 2.5 Acknowledgements ...... 27 2.6 References ...... 27

vii

3 ASME Turbo Expo 2017: Experimental Based Empirical Model of the Initial Onset of Sand Deposits on Hastelloy-X from 1000°C to 1100°C Using Particle Tracking ...... 30 Abstract ...... 30 Nomenclature ...... 30 3.1 Introduction ...... 31 3.2 Experimental Method ...... 33 3.2.1 Test Equipment ...... 33 3.2.2 Experiment: Design and Testing Conditions ...... 34 3.2.3 Empirical Deposition Model for Prediction ...... 36 3.3 Results ...... 37 3.3.1 Near-Surface Coupon Temperature ...... 37 3.3.2 Particle Impact Vectors ...... 38 3.3.3 Particle Deposits ...... 41 3.3.4 Empirical Deposit Model for Prediction ...... 43 3.3.5 Validation Testing ...... 45 3.4 Conclusions ...... 46 3.5 References ...... 46 4 Experimental Sticking Probability of Arizona Road Dust on Hastelloy X in Analogous Hot-Section Temperatures...... 50 Abstract ...... 50 Nomenclature ...... 50 4.1 Introduction ...... 51 4.2 Testing and Analysis ...... 52 4.2.1 Test and Analysis Assumptions ...... 52 4.2.2 Test Equipment ...... 52 4.2.3 Updated Analysis Method ...... 54 4.2.4 Additional Test Data ...... 60 4.3 Data Reduction and Analysis ...... 60 4.3.1 Independent Variable Distributions ...... 60 4.3.2 Sticking Probability Prediction Model ...... 63 4.3.3 Non-Dimensional Sticking Probability Model ...... 67 4.4 Conclusions ...... 73 4.5 Appendix: Material Property Calculations...... 73 4.6 References ...... 75 5 Notional Future Research ...... 78

viii

5.1 Test Dust ...... 78 5.2 Testing Process ...... 79 5.3 Alternate Test Geometries ...... 79 5.4 Priority of Analysis/Testing ...... 79 5.5 References ...... 80 6 Appendix ...... 81 6.1 Data Reduction Process ...... 81 6.1.1 Deposition ...... 82 6.1.2 Particle Tracking ...... 86 6.1.3 Sticking Probability Using Particle Tracking and Deposition Data ...... 88 6.2 Maximum Notional Sticking Probability ...... 90 6.3 ARD Softening Temperature and Viscosity Calculations...... 91 6.4 Dynamic Modulus of Elasticity of Hastelloy® X ...... 95 6.5 Dimensional Analysis ...... 98 6.6 Boundary Layer on the Solid Test Coupon ...... 100 6.7 References ...... 101

LIST OF FIGURES

Figure 1-1: Comparison of the grains of Arizona Test Dust (LEFT) and MIL-5007C dust (RIGHT) [22]. 4 Figure 1-2: Current setup of the Virginia Tech Aerothermal Rig (VTAR) for deposition testing up to 1100°C...... 5 Figure 1-3: A CAD representation of the coupon test arrangement...... 5 Figure 1-4: Two views of the test coupon arrangement. (LEFT) Top-down view of coupon and equilibration tube with the test section casing and flanges hidden. (RIGHT) An isometric view of the coupon and equilibration tube indicating thermocouple placement top and bottom of the coupon. The dashed line across the coupon surface indicates the horizontal midline of the coupon and gas flow path...... 6 Figure 1-5: A baseline pre-test image of the coupon surface (LEFT) and a post-test microscopic image at 20x magnification from an 1100°C and 80° coupon angle test condition...... 9 Figure 1-6: Raw data (points), regression lines, and prediction intervals (shaded regions) for conditions between ~975°C and 1100°C gas path temperatures for all tested angles...... 11 Figure 1-7: Contour plot of the MLR model showing the expected number of particles per are based on average injection temperature and coupon angle...... 11 Figure 1-8: (LEFT) A distribution of the average particles per area does not follow a normal distribution and is skewed right. (RIGHT) The distributions of a square root of the average particles follows a normal distribution...... 13 Figure 2-1: The Virginia Tech Aerothermal Rig (VTAR) utilized for deposition testing between 1000°C and 1100°C, highlighting the test section cut-away view and the associated coupon...... 20 Figure 2-2: Two views of the coupon arrangement. (LEFT) Top-down view of coupon and equilibration tube with the test section casing and flanges hidden. The coupon angle, θ, is the acute angle between the bulk flow and the coupon surface. (RIGHT) An isometric view of the coupon and equilibration tube indicating thermocouple placement top and bottom of the coupon. The yellow dashed line across the coupon surface indicates the horizontal midline of the coupon and gas flow path...... 21 Figure 2-3: (LEFT) Pre-test and post-test surface sample image comparison showing the deposits and Hastelloy-X surface. The image size is 698 μm by 522 μm at 20x magnification. (RIGHT) Rendered image of the coupon surface that highlights the locations of the sample image rows that are located at the midline, and the quarter and three quarter locations vertically along the horizontal axis...... 23 Figure 2-4: (LEFT) Average particle deposits per square millimetre on the coupon and (RIGHT) average deposit percent coverage area on the coupon. Both responses are plotted against near-surface coupon temperature. The average deposits per square millimetre is less than anticipated due to significant deposited particles overlapping on the surface. The 95% confidence intervals are shown for each test case...... 25 Figure 2-5: Particle deposits per area for the 50° and 80° coupon angle test cases with an estimated interval using the coverage area data...... 26 Figure 3-1: The Virginia Tech Aerothermal Rig (VTAR) utilized for deposition testing, highlighting the test section cut-away view and the associated coupon...... 33 Figure 3-2: Two views of the coupon arrangement. (a) Top-down view of coupon and equilibration tube with the test section casing and flanges hidden. (b) An isometric view of the coupon and equilibration tube with thermocouple placements...... 34 Figure 3-3: Flowchart for the empirical CR model for prediction across the test conditions in this study. 36 Figure 3-4: Pretest and posttest surface sample image comparison showing the deposits and Hastelloy-X surface with a rendered image of the coupon that highlights the locations of the sample image rows...... 37

Figure 3-5: Linear local coupon near-surface temperature profiles for a 1050°C flow temperature at 20°, 50°, and 80° coupon angle...... 38 Figure 3-6: Normal impact velocity regression models for the 1050°C flow temperature...... 39 Figure 3-7: Tangential impact velocity regression models for 1050°C flow temperatures...... 39 Figure 3-8: A CFD comparison between the coupon angles of 50° (a) and 80° (b) velocity magnitude flow fields at 1050°C...... 40 Figure 3-9: Highlights single and double impact particles for the 1100°C at 80° test case both using a robust regression model with bisquare weighting...... 41 Figure 3-10: Example post-test (1100°C at 50°) coupon surface comparison of deposits at the leading and trailing edges. There is high particle deposit overlap at the leading edge ...... 41 Figure 3-11: Regression models of the CR per image at 1100°C across the coupon for all tested angles. Deposits for the 80° case are lower due to the lower surface temperatures compared to the 50° case...... 42 Figure 3-12: Contour plot of the empirical CR model for prediction at various near-surface temperatures. As near-surface temperature increases, the CR (and effective sticking probability) increases...... 44 Figure 3-13: Empirical CR model for prediction at varying tangential impact velocities. As tangential velocity increases, the CR (and effective sticking probability) decreases...... 44 Figure 3-14: For the 1050°C flow temperature at 60° coupon angle, the predicted versus observed CRs for validation test (a) and observed minus predicted CRs by coupon location (b)...... 45 Figure 4-1: The Virginia Tech Aerothermal Rig (VTAR) utilized for deposition testing, highlighting the test section cut-away view and the associated coupon...... 53 Figure 4-2: Two views of the coupon arrangement. (a) Top-down view of coupon and equilibration tube with the test section casing and flanges hidden. (b) An isometric view of the coupon and EQLB with thermocouple placements...... 54 Figure 4-3: High level data reduction process for each test conducted...... 55 Figure 4-4: Highlighted region of interest (in green) as well as the regions towards the outer edges of the coupon (in red)...... 56 Figure 4-5: Diagram of a simplified impact process depicting the assumption whereby the particle deposit diameter is equal to the particle diameter...... 57 Figure 4-6: Normalized particle size distribution based on number of particles that has been normalized between 10-40 µm for a 20-40 µm ARD sample...... 58 Figure 4-7: Illustration of the particles that will hit or miss a test coupon based on physical orientation to the flow within VTAR...... 59 Figure 4-8: Histogram of the normal (a), tangential (b), and impact velocity magnitude (c) for all tests conducted in this study...... 61 Figure 4-9: Histogram distribution of impact efficiency for particles within the projected normal coupon area after leaving the EQLB tube...... 62 Figure 4-10: A majority of sticking probabilities were less than 0.01...... 63 Figure 4-11: Sticking Probability versus flow-particle temperature for particles between 19 µm to 40 µm. Surface temperature, normal impact velocity, and impact velocity magnitude are median values from the data set...... 65 Figure 4-12: Sticking probability versus flow-particle temperature for particles between 10 µm to 19 µm...... 65 Figure 4-13: Sticking Probability versus temperature ratio (/ ) for particles between 19 µm and 40 µm. Normal impact velocity and the impact velocity magnitude are median values from the data set...... 66

Figure 4-14: Sticking probability versus temperature ratio (/ ) for particles between 10 µm and 19µm...... 67 Figure 4-15: Sticking probability increases with for particles between 19 µm and 40 µm...... 70 Figure 4-16: Sticking probability versus for particles between 10 µm to 19 µm particles ...... 70 Figure 4-17: Sticking probability increases pseudo-logarithmically with for particles between 19 µm to 40 µm...... 71 Figure 4-18: Sticking probability increases with for 10 µm to 19 µm particles ...... 72 Figure 4-19: Viscosity of ARD for the range of flow/particle temperatures in this study...... 74 Figure 4-20: HASTELLOY® X dynamic modulus of elasticity for the range of surface temperatures in this study...... 75 Figure 6-1: Flowchart of data reduction process for a single test...... 82 Figure 6-2: Posttest coupon data acquisition for deposits...... 83 Figure 6-3: Process for identifying deposits per each sample image row ...... 84 Figure 6-4: Combining data files from each sample image row and separating particle deposits based on size then output a collated data file to correlate to impact vectors...... 85 Figure 6-5: Process to identify initial position and velocity vectors of particles for each test case using PIV image data...... 87 Figure 6-6: Determine particle trajectories, impact efficiency, relative to particle size using initial trajectories and the CFD velocity flow field. Also establish regression models for particle impacts and rebounds relative to paritcle size...... 88 Figure 6-7: Combining 3 to 4 data files allows for SP to be calculated independent of location on the test coupons...... 89 Figure 6-8: Viscosity of ARD across the range of test temperatures in this study...... 93 Figure 6-9: Viscosity of ARD from the original logarithmic regression compared to a linearized regression...... 94 Figure 6-10: Dynamic modulus of elasticity for HASTELLOY® X for the range of surface temperatures within this study...... 96 Figure 6-11:Dynamic modulus of elasticity for HASTELLOY® X comparison for first and second order regressions...... 97 Figure 6-12: Boundary layer thickness of flow across the solid test coupon for a variety of potential flow velocities...... 100

xii

LIST OF TABLES

Table 1-1: Chemical analysis for typical Arizona Test Dust products from Powder Technologies Inc...... 4 Table 1-2: Constant test conditions for varying gas path temperature and coupon angle testing ...... 7 Table 1-3: Testing matrix for varying gas path temperature bins and coupon angle ...... 8 Table 1-4: Deposits and achieved temperatures during testing for all testing conditions...... 10 Table 1-5: Multiple linear regression (MLR) model parameters using test data input into JMP®...... 13 Table 2-1: Constant test conditions for varying flow temperature and coupon angle testing...... 21 Table 2-2: Aggregated deposit data for each test as well as the average near-surface temperature...... 24 Table 2-3: 2 and 2 values for particle deposits per area (Equation (2-1)) and deposit percentage coverage (Equation (2-2)) ...... 27 Table 3-1: Controlled test conditions for varying flow temperature and coupon angle testing...... 35 Table 3-2: Equation coefficients for the empirical CR model for prediction using near surface temperature (K) and velocity (m/s)...... 43 Table 3-3: Prediction accuracy based on validation testing...... 46 Table 4-1: Additional testing performed to compliment previous test data [4]...... 60 Table 4-2: Data range for each primary variable with associated units...... 61 Table 4-3: Coefficients for Eqn. (4-7) and Eqn. (4-8) ...... 64 Table 4-4: 2 and Root Mean Standard Error (RMSE) for Eqn. (4-7) and (4-8) ...... 64 Table 4-5: Coefficients for Eqn. (4-13) and (4-14) ...... 69 Table 4-6: 2 and Root Mean Standard Error (RMSE) for ...... 69 Table 6-1: Particle metrics and number of particles calculated from testing ...... 90 Table 6-2: Maximum possible number of impacts based on geometry and ARD sample distribution...... 90 Table 6-3: Maximum potential number of particles and mass based on particle size for a monolayer of deposits...... 90 Table 6-4: Mole fraction of ARD sample for testing...... 91 Table 6-5: Coefficients for compositional dependence of ...... 92 Table 6-6: Dynamic modulus of elasticity of Hastelloy® X across a broad temperature range from Haynes International [12] and Varela et. al. [13]...... 95 Table 6-7: Common independent and controlled variables associated with deposition. Variables highlighted in yellow are repeated variables for the Buckingham Pi theorem...... 98 Table 6-8: Variations of non-dimensional terms used for model optimization...... 99

xiii

1 ASME TURBO EXPO 2016: PRELIMINARY EXPERIMENTAL INVESTIGATION OF INITIAL ONSET OF SAND DEPOSITION IN THE TURBINE SECTION OF GAS TURBINES

Proceedings of the 61st International Gas Turbine Institute ASME IGTI 2016 June 13-17, 2016, Seoul, South Korea

GT2016-56059

Andrew Boulanger, Hardik Patel, John Hutchinson, William DeShong, Weibin Xu, Wing Ng, Srinath Ekkad Virginia Polytechnic Institute & State University Blacksburg, VA, USA

ABSTRACT Particle ingestion into modern gas turbine engines is known to reduce performance and may damage many primary gas path components through erosion or deposition mechanisms. Many studies have been conducted that evaluate the effects of particulate ingestion in primary and secondary gas path components. However, modern gas turbines have gas path temperatures that are above most previous studies. As a result, this study performed particle deposition experiments at the Virginia Tech Aerothermal Rig facility at engine representative temperatures. Arizona Test Dust of 20 to 40 μm was chosen to represent the particle ingested into rotorcraft turbine engines in desert and sandy environments. The experimental setup impinged air and sand particles on a flat Hastelloy-X coupon. The gas and sand mixture impacted the coupon at varying angles measured between the gas flow direction and coupon face, hereby referred to as coupon angle. For this study, gas and sand particles maintained a constant flow velocity of about 70 m/s and a temperature of about 1100°C. The coupon angle was varied between 30° to 90° for all experiments. The experimental results indicate sand deposition increased linearly from about 975 °C to 1075 °C for all coupon angles. A multiple linear regression model is used to estimate the amount of deposition that will occur on the test coupon as a function of gas path temperature and coupon angle. The model is adequate in explaining about 67% of the deposition that occurs for the tests. The remaining percentage could be explained with other factors such as particle injection rates and exact surface temperature where the deposits occur.

NOMENCLATURE ATD Arizona Test Dust CAD Computer-aided Design MLR Multiple Linear Regression PIV Particle Image Velocimetry PPMW Parts Per Million by Weight VTAR Virginia Tech Aerothermal Rig Intercept Parameter Temperature Parameter Coupon Angle Parameter

, Estimated Deposits per Area Random error Gas Path Temperature (°C) Coupon Angle (degrees)

1.1 INTRODUCTION Operation of propulsion turbines in dust-laden environments is known to reduce performance and cause damage to various system components. Ingested particles will enter the primary gas paths where they can erode metal surfaces and entrain the coolant bypass system in the compressor section. The particulates will continue into the combustor and turbine sections where they could melt and deposit on the various hot section parts. The deposits may lead to overheating and melting of substrate material from clogged film cooling holes. This study focuses on the initial onset of microparticle deposition at representative gas turbine hot-section temperatures. Specifically, this work follows the high temperature microparticle rebounding experiments and modeling previously performed at Virginia Tech [1–7]. The purpose is to determine a statistical estimate of the initial onset of deposition for 20 to 40 μm Arizona Test Dust (ATD) as a function of gas path temperature and relative microparticle impact angle. The experiments were performed at the near melting point of the ATD on an uncooled flat Hastelloy-X test coupon. Deposition in this study is quantified by counting particles per area on multiple microscopic images on the coupon surface. Quantifying the onset of deposition at engine-representative temperatures is important to estimate subsequent deposits in hot- section gas turbine conditions.

1.2 BACKGROUND Particulate ingestion into propulsion turbines is a common event in many arid regions across the world. The detrimental effects of particulate ingestion can vary depending on a variety of factors, including mass ingested, particulate composition, engine material properties, etc. Increasing service interval frequency is a common way to mitigate particulate ingestion for propulsion turbines. In more severe cases, the engines can be catastrophically damaged. For instance, during the Mount St. Helens eruption, an L-100 aircraft had two engines fail while the other two had partial power loss while flying below the ash cloud after the eruption began [8]. The observed damage to the engines was severe abrasion of the compressor section as well as melted dust in the turbine section of all the engines. Another example was during Operation Desert Shield and Desert Storm in 1991, where several M1 Abrams tank turbines failed due to sand ingestion [9]. A recent example on May 17, 2015, an MV 22 Osprey crashed in dusty conditions during a training exercise in Hawaii. Currently, the exact cause of this crash is unknown, but dust ingestion may have been a factor [10]. Particulate ingestion effects have been previously studied to determine the performance degradation and resilience on common propulsion turbine engines. A catalyst for some of the early studies was the aforementioned eruption of Mount St. Helens and the dust cloud effects on the L-100 aircraft engines. Additionally, the detonation of nuclear weapons can loft dust into the atmosphere where it would be ingested by aircraft engines and subsequently cause engine damage [8]. As a result, a T33 turbofan and J75 turbojet were tested in dust-laden environments to evaluate performance degradation and damage. The dominant damage to the test engines was compressor blade erosion, but there were no glassy deposits found on the hot section components unlike the aforementioned L-100 aircraft [11–13]. Subsequent testing of an F-100 turbofan engine in a simulated nuclear dust environment was intended to extend the operational life of an engine that had been severely damaged due to particle ingestion. The F-100 turbofan experienced

2 damage mechanisms of erosion and deposition in the cold and hot sections of the engine, respectively. The operators of those tests were able to “clean” the engine by rapidly cycling the power levels, inducing thermal transient in the turbine airfoils. The thermal transient caused the deposited material to break free of the internal turbine substrates. However, after a long exposure to various particulates and feed rates, the engine was damaged beyond operational capability [14]. Full scale testing is important in addressing propulsion turbines responses and endurance to particulate ingestion. Improving the resilience of the engines requires an additional understanding of the basic erosion and deposition mechanisms on the individual components. The primary focus of the research presented in this paper is the onset of particle deposits. The erosion mechanisms will not be discussed in extensive detail. Subsequent experiments after the F-100 full engine tests involved testing deposition on a nozzle guide vane (NGV) cascade one quarter section with various cooling schemes and different dust blends. The general conclusion from the testing is that deposition of particulates onto the NGVs is dependent on the particulate type and size as well as the substrate surface temperature [15]. Subsequent testing of cylinders and cascade sections confirmed that substrate surface temperature and particulate type does dictate deposit build up. In addition, large thermal gradients and thermal cycling of the substrate may result in removal of the deposits [16]. Since dust and sand can vary in composition and size. Testing has been performed on various airframes to determine the quantity and sizing of sand occurs in brownout conditions [17]. Typical dust concentrations in extreme conditions (depending on airframe) can be as high as 5000 parts per million by weight (PPMW), which can be ingested into the engines. Dust composition can vary depending on geographic location. For instance, even in a relatively small region in the Dhahran area the composition of sand is primarily SiO2 and Al2O3 but can vary in large proportions [18,19]. The large variation in sand composition will affect the deposition characteristics in the hot sections of the turbine. Typically, the glassy-deposits in the hot sections are referred to as CMAS since they are a composition of calcium, magnesium, aluminum, and silicon oxides. The deposition of particulate is dependent on the motion mechanisms and the particle material phase as a function of temperature. The motion mechanisms of deposition are inertial impaction, turbulent diffusion/eddy impaction, Brownian diffusion, and thermophoresis [20]. Previous testing and modeling has generally demonstrated deposition as a function of particulate size. Above 1 µm, the particulate deposit mechanism is dominated by inertial impacts. Since this study utilizes ATD of 20 to 40 µm, the primary mechanism is inertial impaction for all tests. The particle temperature and phase of the material can determine the probability that a particle will stick to a substrate when coupled with particle motion. The particulate material at elevated temperatures will proceed through four phases as temperature increases. The four phases are characterized as the shrinking temperature, deformation temperature, hemisphere temperature, and flow temperature [21]. Each phase temperature can vary depending on the particulate and all four phases occur during the sintering phase. Additionally, as the temperature increases, the probability of deposition increases since the particulate material is effectively “softened”. For this study, the particulate used for deposition testing is ATD from Powder Technologies Inc. The typical chemical composition, shown in Table 1-1, indicates that silicon and aluminum oxides are the primary constituents. ATD has an onset sintering temperature of 1100°C for sizing less than 63 µm [22]. The sintering temperature indicates the beginning of the softening temperature of the ATD particulate. In the high temperature environments, such as the hot sections of a gas turbine, typical ATD will lose 2-5% of its original sample mass due to the water in the particulate samples evaporating from ambient temperature to 700°C. The shape of ATD is more angular than MIL-5007C test dust, which is primarily composed of

3 quartz. The shape and sintering temperature phases of the particulate will affect deposition on hot turbine components. For this study the particulate shape and impact orientation is not evaluated. The deposition investigated is primarily a function of gas path and environmental temperatures.

Figure 1-1: Comparison of the grains of Arizona Test Dust (LEFT) and MIL-5007C dust (RIGHT) [22].

Table 1-1: Chemical analysis for typical Arizona Test Dust products from Powder Technologies Inc.

Component Percentage by Weight SiO2 68-76% Al2O3 10-15% Fe2O3 2-5% Na2O 2-4% CaO 2-5% MgO 1-2% TiO2 0.5-1.0% K2O 2-5%

1.3 EXPERIMENTAL METHOD Aerothermal Rig. The Virginia Tech Aerothermal Rig (VTAR) was donated by Rolls Royce (Indianapolis, IN) to Virginia Tech in 2010. The previous application of the rig prior to the donation was for heat transfer studies of cascade turbine airfoils [23–26]. The VT Aerothermal Rig was used to conduct all the experiments discussed in this paper. Figure 1-2 is a schematic CAD model of VTAR’s current configuration for sand deposition testing. VTAR was used for erosion based testing by evaluating the coefficient of restitution (COR) [1–7]. Since the previous work performed at Virginia Tech [6], the fuel source was changed from methane to propane and the maximum gas path temperature at the test coupon has been increased to about 1100°C. In addition, the focus of the analysis has shifted from COR to deposition using ATD. The main gas path mass flow rate is about 0.1 to 0.15 kg/s depending on testing temperature. The equilibration tube is 7.62 cm in diameter and 1.82 m long. The flow velocity leaving the equilibration tube is maintained to 70 ± 2 m/s.

Figure 1-2: Current setup of the Virginia Tech Aerothermal Rig (VTAR) for deposition testing up to 1100°C.

The sand particles are entrained in a flow separate from the main flow and are injected into the main flow in the equilibration tube after the burner. The sand is drawn from a hopper using a venturi style entrainment system. The sand laden flow is then injected into the main gas path in the upstream direction to promote sufficient mixing by the end of the equilibration tube. The equilibration tube length allows a sufficient amount of time for the temperature of the sand particles to equalize with the gas path temperature. Figure 1-3 is a CAD representation of the test coupon support system inside the test section (with the outer case wall hidden). The coupon can be rotated in 10° increments. The coupon angle is defined as the acute angle between the gas path and the surface of the coupon, Figure 1-4 (left). Gas path flow temperature measurements are taken with two thermocouples placed above and below the coupon and within the gas path, Figure 1-4 (right). The distance from the end of the equilibration tube to the leading edge of the coupon is approximately 10 cm. The distance will change during operations due to the thermal expansion of the tube into the test section via a slip joint. The coupon material is Hastelloy-X, which is the same from the previous study [6].

Figure 1-3: A CAD representation of the coupon test arrangement.

Figure 1-4: Two views of the test coupon arrangement. (LEFT) Top-down view of coupon and equilibration tube with the test section casing and flanges hidden. (RIGHT) An isometric view of the coupon and equilibration tube indicating thermocouple placement top and bottom of the coupon. The dashed line across the coupon surface indicates the horizontal midline of the coupon and gas flow path.

Testing Conditions. Table 1-2 is a summary of the microparticle deposition testing conditions for this study. The particulate used is 20 to 40 μm ATD and the gas path flow velocity is 70 m/s to maintain consistency from the previous studies using VTAR [1–7]. The ATD used for all testing maintains the same relative ISO distribution from Table 1-1. The authors are aware that ATD is not a practical representation for deposition studies compared to airborne sands and dusts found across the world [11]. Additionally, the particulate sizing is not directly representative of sand sizing in the hot section of turbine engines. Particle sizing is typically below 10 µm after the cold sections of a turbine engine since the particles are “pulverized” as they travel through the compressor. For example, testing of TF33 turbofan and J57 turbojet engines had a mean particulate size after the compressor of 6 µm [13]. Particles larger than a few microns in diameter are dominated by inertia impaction mechanisms for deposition. Alternate mechanisms of deposition such as turbulent diffusion and thermophoresis dominate particles under a few microns in size [20]. Based on particle deposition mechanism sizing from earlier studies, 20 to 40 µm the ATD size is relevant for inertial impact deposition mechanisms. In addition, the ATD used for this study provides continuity with the earlier deposition study at Virginia Tech [6] as well as improves the detection of deposited particles using an image processing computer software. VTAR’s sand injection system is maintained from the previous COR and deposition studies [6,7]. Before each test, the coupon is polished to a mirror finish with a measured root mean square roughness of less than 0.2 μm to ensure roughness effects are minimized. Dust concentrations ingested by turbine engines on airframes can be as high as 5,000 PPMW for extreme low visibility conditions [17]. Microparticle injection rates for VTAR can vary between 1 to 2 grams per second. All the ATD sample is injected between 5 to 10 seconds, which correlates to approximately 10,000 to 20,000 PPMW of ATD equivalent being injected into the equilibration tube where the particles reach the flow temperature and velocity and subsequently impact the test coupon. The total sand injected per test was maintained at a constant 10 grams. Preliminary testing established the relatively low total sand loading to minimize deposits from being inadvertently removed from the surface through subsequent impacts.

Table 1-2: Constant test conditions for varying gas path temperature and coupon angle testing

Particulate Type/Size ATD, 20 – 40 μm Bulk Flow Velocity 70 ± 2 m/s Desired Gas Path Temperature 1000 °C to 1100 °C Range Injection Rate ~1 g/s to 2 g/s Total Sand Loading 10 grams Coupon Angles 30°, 50°, 80°, 90° Coupon Material Hastelloy-X Coupon Surface Roughness ≤ 0.2 μm Coupon Thickness 3.175 mm Particle Stokes Number ~1.6 (20 μm) to 6.8 (40 μm)

Unlike the constant coupon angles, the gas path temperatures can vary during ATD injection. Before injection, the gas path temperatures are maintained to within 5 °C of the target testing main gas path temperature. After all testing parameters are at steady state, the ATD is injected. Since air and sand is being added to the gas path, the average gas path temperature drops, typically, 10 °C. During the injection transient, the temperatures are recorded and averaged. For example, a target temperature would be 1000 °C, but during injection the average gas path temperature drops to 980 °C. Since the temperatures are near and/or above the sintering or melting temperature of ATD [22], slight changes in temperature have a large impact on the overall deposition.

Substrate and surface temperatures of test pieces are directly linked to large scale deposit testing. In the previous study [6], the test coupons were assumed to be under an isothermal condition with a surface temperature approximately 140 °C below the gas path temperature. In this study, the surface temperature assumption is not maintained. Due to test equipment constraints, coupon surface temperature could not be directly measured. However, the surface temperature is a relative function of the gas path temperature and heat transfer to the coupon support structure and the surrounding walls in the test section.

Experimental relations for jet impingement onto a flat plate [27] are used to determine the convective heat transfer coefficient of the system and one-dimensional heat transfer calculations were performed to determine an estimate of the coupon surface temperature from conduction, convection, and radiation effects. From this analysis, it was found that coupon angle has little effect on the surface temperature and that the coupon surface temperature is within ±2 °C of the temperature measured on the back surface of coupon for all test cases. The coupon back temperatures are included in Table 1-4. From testing prior to this study, glassy deposits developed inside a ceramic equilibration tube that began near the injection location and continued approximately half the length of the tube. Typically, ceramic substrates will collect and maintain deposits more readily than a metallic substrate during thermal cycling. The ceramic substrate and the sand particles will maintain their bond due to similar thermal expansion coefficients. For this study, the authors used a stainless steel equilibration tube instead of a ceramic tube. Negligible deposits were found during borescope inspections (every two to three tests). Deposits during sand injection were still anticipated but any significant deposits most likely delaminated from the inside surface of the tube and move downstream during cool down period of each test. Therefore, there is no significant flow change to main gas path during testing between tests. Although there is no significant build up over time inside the equilibration tube, there is uncertainty associated with the deposition occurring inside the equilibration tube during that may reduce the sand

7 reaching the coupon. Due to time constraints and resource limitations, the tube could not be weighed before and after each test. In addition, the relative added weight by any deposited sand would be negligible compared to the weight of the equilibration tube and hardware. Further study is necessary to quantify the deposits on parallel flow surfaces using ATD. However, the statistical method developed in this study helps mitigate the uncertainty of the equilibration tube deposition. Statistical Modeling Method. Compared to the previous study at Virginia Tech [6], this study establishes a method for estimating the onset of deposition at various coupon angles and average gas path temperatures. A method to examine the effect of combinations of gas path temperature and coupon angle on the average sand deposition, i.e. if there is any difference in sand deposition among three target gas path temperatures (1000°C, 1050°C and 1100°C) and among the four coupon angles (30°, 50°, 80° and 90°) is needed. Therefore, a two-way layout experimental design is used to study the effect of gas path temperature and coupon angle on deposition [28]. There are three levels of gas temperatures and four levels of coupon angles, leading to a total of 12 possible experimental conditions. In order to have enough degrees of freedom for the error term [28] and to factor out the differences in sand deposition caused by random variation between experiments, replicate testing is necessary for each experimental condition. Ideally, more replicates reduce the effect of random variation on the response variable, deposition. However, due to the time and budget constraints, only two tests were conducted for each experimental condition. Therefore, the total number of tests performed is 24. A statistical software, JMP®, generated the experimental condition testing order for a total of 24 tests that meets the four rules of a perfect two-way layout experimental design, Table 1-3. JMP® ensures that the experimental design for this study (1) is “randomized”, (2) has “fixed effects”, (3) is “completely balanced” and (4) “completely crossed” [29]. If all four rules are met, the effect of each factor on the response variable can be measured. In addition, the effect of each factor on the response variable and have more degrees of freedom and the power will be maximized for the statistical model. A “randomized” design requires that each ATD sample is randomly assigned to each experimental condition. This is achieved by randomizing the order of the experimental conditions. Randomizing the order reduces bias caused by uncontrollable factors during the testing of each experimental condition, such as the local air pressure and humidity during each test day. The “fixed effects” in this model are the gas path temperature and coupon angle, which means they are the primary variables of interest and not randomly generated. A “completely balanced” design assumes each factor is run the same amount of times in the experiment and equal weights for each treatment in the analysis of variance [29,30]. Lastly, “completely crossed” means that all the levels of gas path temperature and coupon angle are present for all the levels of one another, which results in the aforementioned total of 12 experimental conditions.

Table 1-3: Testing matrix for varying gas path temperature bins and coupon angle

Target Temperature Coupon Angle Target Temperature Coupon Angle Index Index (°C) (degrees) (°C) (degrees) 1 1000 90 13 1050 90 2 1000 80 14 1050 80 3 1100 50 15 1100 80 4 1100 30 16 1000 30 5 1100 30 17 1100 90 6 1000 50 18 1050 50 7 1050 90 19 1050 80 8 1000 50 20 1000 80

Table 1-3: Testing matrix for varying gas path temperature bins and coupon angle

Target Temperature Coupon Angle Target Temperature Coupon Angle Index Index (°C) (degrees) (°C) (degrees) 9 1100 50 21 1000 90 10 1000 30 22 1050 50 11 1100 80 23 1050 30 12 1050 30 24 1100 90

1.4 RESULTS Raw Data. Figure 1-5 is an example of a pre-test and post-test (1100 °C and 80° coupon angle) microscopic sample image at 20x magnification. Images are taken in series across the horizontal midline of the coupon starting from the leading edge and continuing towards the trailing edge, see Figure 1-4 (right). Images are taken in close proximity to each other but cannot be stitched together due to data acquisition equipment and processing limitations. The images are taken with the focus set to the coupon surface and the particles slightly out of focus resulting in a distinct boundary between the uniform particle and the non-uniform background. Each image is processed using the MATLAB® Visual Processing Toolbox to identify and trace the boundaries of each particle with a cross section corresponding to a circular diameter larger than 10 μm, to ensure most of the smaller particle deposits are counted. Due to the irregular shape of ATD particulate [22], the lower limit is assumed to be adequate for a majority of the deposited particles.

Figure 1-5: A baseline pre-test image of the coupon surface (LEFT) and a post-test microscopic image at 20x magnification from an 1100°C and 80° coupon angle test condition.

Each image is then manually checked to ensure all particles have been successfully identified and no false positives have occurred. In addition, based on microscopic images, the particles do not appear to have delaminated from the coupon surface during or after post-test cool down. The local adhesion between Hastelloy-X, a nickel-based substrate, and SiO2, a primary constituent for ATD, creates a strong atomic level bond that is resistant to thermal cycling. Previous research has implied that a crystalline SiO2 layer between a nickel substrate alloy and a thermal barrier coating could provide the bond strength necessary to avoid delamination during thermal cycling [31]. Since the bond between the nickel-based substrate and the SiO2 is stronger than the rest of the crystalline structure of the SiO2, the remaining crystal structure above the bonded region may fracture and liberate itself from the bonded layer. A bonded region with some crystalline structure should remain that can be counted with the image processing method described previously. Overall, the deposition counts confirm the previous study’s conclusion that deposition increased as primary gas path temperature increased from 950 to 1050°C for all test coupon angles [6]. Table 1-4 is the raw data of deposits and achieved gas path and coupon temperatures for all testing conditions. The index for each

9 condition correlates to Table 1-3. Under most conditions, the temperatures dropped below the target temperatures due to the injection of sand and air into the main gas path. The temperature drop is directly caused by the sand injection process. During the injection process, air at ambient temperature (20°C to 30°C) is injected into the main flow. The injection air contributes about 5% to 6% of the total gas mass flow rate.

Table 1-4: Deposits and achieved temperatures during testing for all testing conditions.

Coupon Temperature (°C) Deposits Angle Index Coupon Target Achieved (degrees) Particles/mm2 Back 1 1000 995 854 90 4 2 1000 1000 853 80 3 3 1100 1090 981 50 41 4 1100 1089 1064 30 104 5 1100 1076 1057 30 69 6 1000 988 952 50 12 7 1050 1012 924 90 67 8 1000 972 949 50 8 9 1100 1080 965 50 59 10 1000 982 903 30 0 11 1100 1092 934 80 120 12 1050 1012 937 30 20 13 1050 1008 925 90 73 14 1050 1013 916 80 84 15 1100 1067 948 80 120 16 1000 988 914 30 3 17 1100 1068 902 90 59 18 1050 1032 932 50 21 19 1050 1042 899 80 30 20 1000 989 870 80 10 21 1000 990 859 90 29 22 1050 1042 929 50 22 23 1050 1039 950 30 17 24 1100 1080 927 90 110

Figure 1-6 is the average particles per area for all the test conditions of average injection temperature and coupon angle. JMP® is able to develop regression trend lines for all the coupon angle conditions as well as 90% prediction intervals (shaded regions). For this study, the absolute test data and trend lines are used in the analysis. From the data, the deposition trends at certain coupon angles do not follow anticipated behavior. Some of the discrepancy can be attributed to testing conditions that are discussed later in this section. Figure 1-7 is a contour plot developed from the trend lines in Figure 1-6.

Figure 1-6: Raw data (points), regression lines, and prediction intervals (shaded regions) for conditions between ~975°C and 1100°C gas path temperatures for all tested angles.

Figure 1-7: Contour plot of the MLR model showing the expected number of particles per are based on average injection temperature and coupon angle.

The melting point of ATD corresponds to when deposition is assumed to begin. Previous sintering tests on ATD showed that sintering begins for particles less than 63 µm around 1100°C [22]. The deposition tests in this study use particles between 20 and 40 µm. One can extrapolate that smaller particles begin to soften and/or sinter at a lower temperature. In this case of testing, 20 to 40 µm ATD may start to soften around a gas path temperature of 1050°C. Based on the data in Figure 1-6, deposition generally appears to increase linearly from about 975°C to 1075°C except for the 30° coupon angle case. However, the 30° coupon angle case at higher testing temperatures is a potential outlier.

It is noted that several data points are outliers due to variations in testing conditions. The authors were unable to retest several data points due to resource constraints. For instance, the 30° coupon angle case at higher temperatures (greater than 1075°C) has significantly more deposition than the 50° cases at similar temperatures. The large amounts of deposition are most likely the result of gas path temperatures higher than those measured. The cause was determined by the authors to be a large displacement of the coupon outside of the main gas flow path in the test section for tests 1 through 8 due the thermal expansion of the support rod. For those tests, a secondary thermocouple below the coupon was not present due to prior operating processes and procedures. After test 8, a closed circuit camera was added to the side viewing window of the test section to verify coupon placement in the gas path. During subsequent operation, the authors observed that the coupon was displacing the primary thermocouple outside of the main gas path flow. Since the thermocouple was displaced outside of the gas path flow at high temperatures, the recorded temperature was less than the centerline main gas path temperature. For the test cases greater than 1075 °C and 30° coupon angle, the estimated temperature is closer to 1200 °C since a majority the deposits had circular profiles, which indicates that the sand particles may have completely melted [22] just before impact with the coupon. The feedback control system of VTAR for heat input depends on the temperature at the coupon. If the temperature is lower than indicated, additional heat is added to the main gas path flow until the desired temperature is achieved. For the remaining tests (9 through 24), a secondary thermocouple was placed below the coupon to ensure the coupon was centered in the main gas path flow. In addition, the mounting components for the test coupon were altered to minimize extreme displacement during tests. Unfortunately, repeating tests 1 through 8 was not feasible due to timeline and resource constraints. Utilizing the data from those tests is allowable due to the statistical model developed for this study. The statistical model in the following section still maintains a strong correlation between gas path temperature and coupon angle with deposited sand particles. For the 90° coupon angle tests above 1075°C, the reduced deposition may be attributed to a large stagnation area in front of the coupon. The highest deposition was anticipated at the highest temperature and highest coupon angle. A hypothesis is the stagnation area could reduce the Stokes number of the particles just ahead of the coupon and subsequently divert a large portion of the particles around the coupon. In addition, if the particles are slowed down in the test section which is cooler than the equilibration tube, the particles may solidify and bounce off the surface instead of adhering. Overall, variation in the experimental data are taken into account for the statistical model discussed in the following section. Statistical Model. The typical analysis for a two-way layout is a two-way Analysis of Variance (ANOVA). This analysis would give an estimate for each level of the factors tested assuming they are categorical. For example, the model is only capable of estimating the average deposition at 1000°C and 30° coupon angle or 1050°C and 50° coupon angle. The model would not be capable of estimating the deposition at 1025°C and 40° coupon angle. Therefore, a generic multiple linear regression (MLR) model his used instead. The MLR model response is the average number of particles per area and the predictors are average gas path temperature during ATD injection and coupon angle. The major difference between the MLR model and the categorical model is that the predictors are now considered to be continuous. Therefore, a value for the average number of particles deposited can be calculated for gas path temperatures and coupon angles that were not directly tested. The square root of the average sand particles deposited is the optimal MLR model developed using the JMP® modeling tools for this study. The square root transformation was chosen because it yields a response that follows an approximate normal distribution, Figure 1-8 (right). Alternative models were considered but

12 do not provide the simplicity of a quadratic response to deposition for the two predictor variables tested. The MLR model has three primary requirements. First, the combined response (average deposition) of all sample data is assumed to have an approximately normal distribution. Since the sample average deposition data is not normally distributed, Figure 1-8 (left), a square root transform is conducted. Secondly, the MLR model assumes that there is a linear trend between the square root of average sand deposition and temperature and angle respectively (untransformed would have a quadratic trend). The last assumption is constant variance. All of these assumptions are met when the square root transformation is used.

Figure 1-8: (LEFT) A distribution of the average particles per area does not follow a normal distribution and is skewed right. (RIGHT) The distributions of a square root of the average particles follows a normal distribution.

The following equation is the MLR model equation format developed for estimating deposition.

(1-1) , = + ∗+ ∗+

Where , is the average sand particles deposited per area, is the intercept, is the temperature parameter, is gas path flow temperature in Celsius, is the coupon angle parameter, is the coupon angle in degrees and is random error. Table 1-5 contains the parameters for the deposition model from JMP®. In addition, the third column shows the significance of the variables of gas path temperature and coupon angle in terms of a P-value. Values less than 0.05 are considered significant for the MLR model.

Table 1-5: Multiple linear regression (MLR) model parameters using test data input into JMP®.

Parameter Value Prob > |t| -60.25708 <.0001* 0.0613822 <.0001* 0.046278 0.0105

When making a prediction, the estimated values produced from the MLR model will be used in place of the parameters. For example, to estimate the sand deposited per area at 1025°C and 45° coupon angle, the equation would be evaluated as follows. The random error term is disregarded for the calculation since it cannot be quantified.

(1-2) °,° = + ∗ (1025) + ∗ (45)

Overall, the model’s adjusted value of 0.6493 explains the proportion of information of the response (average deposition) that the model is able to determine. Therefore, the MLR model can explain 64.93% of the average deposited particles per area from coupon angle and average gas path injection temperature. The remaining 35.07% may be the result of random error such as environmental testing conditions sand injection rates, etc. For instance, if more control variables are added to the testing matrix such as sand injection rate, the adjusted value may increase. However, the additional variables may confound with other existing variables that are already being tested and may not improve the adjusted value significantly.

1.5 CONCLUSIONS Deposition increases with increasing gas path temperatures and coupon angle. From prior studies, deposition is dependent on substrate temperature and particulate type [15]. In this study, the coupon surface temperatures is a function of gas path temperature. Based on testing conditions, the substrate surface temperature is lower than the actual gas path temperature due to thermal losses inside the test section. Despite the lower coupon surface temperature, the onset of deposition did occur below the sintering temperature of the ATD [22]. The statistical model developed for this study is intended to quantify deposition with some uncertainty based on gas path temperature and coupon angle. Other factors, such as coupon surface temperature, are not directly considered due to testing limitations. An MLR model is able to estimate average deposition of particles per area as a function of gas path temperature and coupon angle. The strength of this model indicates that about 65% of the deposition can be explained by gas path temperature and coupon angle. Therefore, improvement in the statistical model is necessary to explain the remaining model uncertainty. Surface temperature and ATD injection rates may explain most of the remaining variation. The methodology described in this study provides significant flexibility for estimating deposition with two or more testing variables. The basic model can be expanded to complex functions to account for other variations and provide a basis for estimating all conditions were deposition may occur in hot engine environments.

1.6 ACKNOWLEDGEMENTS The work presented here was supported by Rolls-Royce Plc., especially Brett Barker, Dr. Kwen Hsu, and Paul Davis.

1.7 REFERENCES [1] Reagle, C. J., Delimont, J. M., Ng, W. F., Ekkad, S. V., and Rajendran, V. P., 2013, “Measuring the coefficient of restitution of high speed microparticle impacts using a PTV and CFD hybrid technique,” Meas. Sci. Technol., 24(10), p. 105303. [2] Delimont, J. M., Murdock, M. K., Ng, W. F., and Ekkad, S. V., 2014, “Effect of Near Melting Temperatures on Microparticle Sand Rebound Characteristics at Constant Impact Velocity,” ASME Turbo Expo 2014, Volume 1A: Aircraft Engine; Fans and Blowers, ASME, Dusseldorf, pp. 1–11.

[3] Delimont, J. M., Murdock, M. K., Ng, W. F., and Ekkad, S. V., 2014, “Effect of Temperature on Microparticle Rebound Characteristics at Constant Impact Velocity,” ASME Turbo Expo 2014, pp. 1–11. [4] Reagle, C. J., Delimont, J. M., Ng, W. F., and Ekkad, S. V., 2014, “Study of Microparticle Rebound Characteristics Under High Temperature Conditions,” J. Eng. Gas Turbines Power, 136(1), p. 011501. [5] Singh, S., Tafti, D. K., Reagle, C. J., Delimont, J. M., Ng, W. F., and Ekkad, S. V., 2014, “Sand transport in a two pass internal cooling duct with rib turbulators,” Int. J. Heat Fluid Flow, 46, pp. 158–167. [6] Delimont, J. M., Murdock, M. K., Ng, W. F., and Ekkad, S. V., 2015, “Effect of Temperature on Microparticle Rebound Characteristics at Constant Impact Velocity—Part II,” J. Eng. Gas Turbines Power, 137(11), p. 112604. [7] Delimont, J. M., Murdock, M. K., Ng, W. F., and Ekkad, S. V., 2015, “Effect of Temperature on Microparticle Rebound Characteristics at Constant Impact Velocity—Part I,” J. Eng. Gas Turbines Power, 137(11), p. 112603. [8] Gabbard, C. B., LeLevier, R. E., and Parry, J. F. W., 1982, Dust-Cloud Effects on Aircraft Engines - Emerging Issues and New Damage Mechanisms (U), Marina del Rey. [9] Hinton, H. L. J., Warren, D. R., and Schulz, T. J., 2001, Operation Desert Storm: Bradley Fighting Vehicle, Abrams Tank, Apache Helicopter, Patriot Missle System and Foreign Government and Individual Contributions, DIANE Publishing. [10] Whittle, R., 2015, “Fatal Crash Prompts Marines to Change Osprey Flight Rules,” Break. Def. [11] Dunn, M. G., Padova, C., Moller, J. E., and Adams, R. M., 1987, “Performance Deterioration of a Turbofan and a Turbojet Engine Upon Exposure to a Dust Environment,” J. Eng. Gas Turbines Power, 109(3), p. 336. [12] Batcho, P. F., Moller, J. C., Padova, C., and Dunn, M. G., 1987, “Interpretation of Gas Turbine Response Due to Dust Ingestion,” J. Eng. Gas Turbines Power, 109(3), p. 344. [13] Dunn, M. G., Padova, C., and Adams, M. R., 1987, “Operation of Gas Turbine Engines in Dust- Laden Environments,” p. 16. [14] Dunn, M. G., 1991, Exposure of Air Breathing Engines to Nuclear Dust Environment (U), Buffalo. [15] Kim, J., Dunn, M. G., and Baran, A. J., 1996, The “Most Probable” Dust Blend and Its Response in the F-100 Hot Section Test System (HSTS), Buffalo; Alexandria. [16] Weaver, M., Dunn, M. G., and Heffernan, T., 1996, “Experimental Determination of the Influence of Foreign Particle Ingestion on the Behavior of Hot-Section Components Including Lamilloy,” ASME Turbo Expo 1996, ASME, Birmingham, pp. 911–917. [17] Cowherd, C., 2007, Sandblaster 2 Support of See-Through Technologies for Particulate Brownout Task 5 Final Technical Report. [18] Smialek, J. L., Archer, F. a., and Garlick, R. G., 1991, The Chemistry of Saudi Arabian Sand - A Deposition Problem on Helicopter Turbine Airfoils.

[19] Smialek, J. L., Archer, F. a., and Garlick, R. G., 1994, “Turbine Airfoil Degredation in the Persian Gulf War,” J. Met., 46(12), pp. 39–41. [20] Hamed, A., Tabakoff, W. C., and Wenglarz, R. A., 2006, “Erosion and Deposition in Turbomachinery,” J. Propuls. Power, 22(2), pp. 350–360. [21] Song, W., Hess, K. U., Damby, D. E., Wadsworth, F. B., Lavallée, Y., Cimarelli, C., and Dingwell, D. B., 2014, “Fusion characteristics of volcanic ash relevant to aviation hazards,” Geophys. Res. Lett., 41(7), pp. 2326–2333. [22] Kueppers, U., Cimarelli, C., Hess, K.-U., Taddeucci, J., Wadsworth, F. B., and Dingwell, D. B., 2014, “The thermal stability of Eyjafjallajökull ash versus turbine ingestion test sands,” J. Appl. Volcanol., 3(1), p. 4. [23] Turner, E. R., Wilson, W. D., Hylton, L., D., Kaufman, R. M., 1985, Turbine Vane External Heat Transfer, Volume 1. Analytical and Experimental Evaluation of Surface Heat Transfer Distributions with Leading Edge Showerhead Film Cooling, Indianapolis. [24] Hylton, L. D., Nirmalan, V., Sultanian, B. K., and Kaufman, R. M., 1988, The Effects of Leading Edge and Downstream Turbine Vane Heat Transfer, Indianapolis. [25] Nealy, D. A., Mihelc, M. S., Hylton, L. D., and Gladden, H. J., 1983, “Measurements of heat transfer distribution over the surfaces of highly loaded turbine nozzle guide vanes,” J. Eng. Gas Turbines Power, 106(January 1984), pp. 149–158. [26] Hylton, L. D., Mihelc, M. S., Turner, E. R., Nealy, D. a., and York, R. E., 1983, Analytical and Experimental Evaluation of the Heat Transfer Distribution over the Surfaces of Turbine Vanes, Indianapolis. [27] Livingood, J. N. B., and Hrycak, P., 1973, Impingement heat transfer from turbulent air jets to flat plates: A literature survey, Washington, D.C., USA. [28] Wu, C. F. J., and Hamada, M. S., 2009, Experiments: Planning, Analysis, and Optimization, Wiley. [29] Norton, B. J., and Strube, M. J., 1986, “Guide for the interpretation of two-way analysis of variance.,” Am. Phys. Ther. Assoc., 66(3), pp. 402–12. [30] Oehlert, G. W., 2010, A First Course in Design and Analysis of Experiments. [31] Jarvis, E. A. A., and Carter, E. A., 2003, “Exploiting covalency to enhance metal-oxide and oxide- oxide adhesion at heterogeneous interfaces,” J. Am. Ceram. Soc., 86(3), pp. 373–386.

2 ISABE 2017: EXPERIMENTAL INVESTIGATION OF THE ONSET OF SAND DEPOSITS ON HASTELLOY-X BETWEEN 1000°C AND 1100°C

Andrew Boulanger, John Hutchinson, Wing F. Ng, and Srinath V. Ekkad [email protected], [email protected], [email protected], and [email protected] Virginia Tech College of Engineering, Department of Mechanical Engineering Blacksburg, Virginia United States of America

Matthew J. Keefe and Weibin Xu [email protected] and [email protected] Virginia Tech College of Science, Department of Statistics Blacksburg, Virginia United States of America

Brett Barker and Kwen Hsu Rolls Royce Corporation 450 S. Meridian Street Indianapolis, IN United States of America

ABSTRACT Deposit formation on turbine hardware in propulsion turbine engines can occur in many arid regions globally. Characterizing crystalline deposits on metallic substrates can aid in component resilience and health monitor algorithms during particle ingestion. This study has developed two statistical empirical models for prediction from acquired experimental data for the onset of deposits. The prediction models are for crystalline particulate (Arizona Road Test Dust) deposits on a flat rectangular Hastelloy-X test coupon. Particle impingement angles varied between 20° and 80° in experimental flow temperatures of 1000°C to 1100°C. Averaged deposits are methodically quantified through normalized particle deposit tallies per area and percent coverage of the surface using microscopic imaging and image processing programs. Deposit accumulation is a quadratic function of both near-surface coupon temperature and coupon angle. Keywords: sand; deposition; turbines; Arizona Road Dust

NOMENCLATURE ARD Arizona Road Test Dust CCD Central Composite Design NASA National Aeronautical and Space Administration (United States of America) NGV Nozzle Guide Vane PPMW Parts Per Million by Weight RMS Root Mean Square VTAR Virginia Tech Aerothermal Rig

Symbols

Average particle deposits per area Average percent coverage area Coefficient of determination Adjusted coefficient of determination Near-surface coupon temperature Coupon angle (degrees)

2.1 INTRODUCTION Propulsion turbine engine particulate ingestion is a common issue in many arid regions. Typically, sand and dust ingested can produce glassy deposits on combustor and turbine components, which can cause detrimental effects ranging from reduced performance to complete engine failure. Characterizing deposits depends on a variety of conditions, such as the mass ingested, particulate constituents, turbine component materials, and operating conditions. There are a variety of methods to combat detrimental effects including reducing engine power during an ingestion event or increase the service interval frequency for the exposed turbine engines. However, during extreme dust ingestion conditions, catastrophic engine failure is possible. This study provides results from analysis of data from a conducted experiment involving the initial deposits of crystalline particulate, Arizona Road Test Dust (ARD), on a flat rectangular Hastelloy-X test coupon as a relative analogue to actual turbine components. The data from the experiment is aggregated and analysed through two statistical empirical models for prediction. Prominent examples of partial and complete turbine engine failure in airframes and equipment have been documented for several decades. For instance, during the Mount St. Helens eruption, an L 100 aircraft had two engines fail while the other two had partial power loss while flying below the ash cloud after the eruption began [1]. Recently, on May 17, 2015, a United States Marine MV 22 Osprey had a fatal crash during a training exercise in dusty conditions while performing a hover manoeuvre. The right engine flamed out causing the craft to drop 26 meters to the ground. Two US Marines were killed in the event with the remaining crew sustaining mild to severe injuries [2]. The post-accident report cited glassy deposits on the turbine section components as one of the direct causes of the engine failure. Ingested particulate can be separated into two general categories, crystalline and amorphous. Crystalline particles, henceforth referred to as sand, are commonly found in desert regions whereas amorphous particles are found in volcanic ash. Since sand can vary in composition and size, previous airframe testing determined the quantity and sizing of lofted sand in brownout conditions [3]. Typical lofted sand concentrations in extreme rotor wash conditions (depending on airframe) can be as high as 0.005 kg-sand per kg-air. In addition, sand composition can vary depending on geographic location. For instance, even in a relatively small region in the Dhahran area the composition of sand is primarily SiO2 and Al2O3 but can vary in large proportions [4,5]. The large variation in sand composition will affect the deposit formation and accumulation [6]. Crystalline material phase structure and motion mechanisms relative to temperature can determine the probability that a particle will stick to a substrate. Since sand is composed of a variety of base constituents, the response of each non homogenous material can vary at high temperatures. Generally, particulate material proceeds through four structure phases as temperature increases through the sintering process. They are designated as shrinking, deformation, hemisphere, and flow phases [7]. Each phase temperature can vary depending on the particulate constituents. In the context of turbine engine deposits, as the combustion temperatures increase the particulate effectively softens or liquefies, which increases the

18 probability of deposits onto a turbine component substrate. Particulate motion mechanisms associated with deposition are inertial impaction, turbulent diffusion/eddy impaction, Brownian diffusion, and thermophoresis [8]. Previous testing and modelling has demonstrated deposits as a function of particulate sizing. Stokes number is usually able to characterize the likelihood of a particle trajectory before impacting a substrate using the particle size. For Stokes numbers greater than unity, which are typically larger than one micron, the particulate is more likely to be dominated by inertial effects and resist changing direction due to the flow field. Since this study utilizes 20 to 40 µm ARD, the primary mechanism is inertial impaction for all test conditions.

2.2 EXPERIMENT METHOD All test were performed using the Virginia Tech Aerothermal Rig (VTAR) at the Advanced Propulsion and Power Laboratory at Virginia Tech. VTAR is capable of an 1100°C maximum flow temperature at the test section for various sand injection concentrations. Tests were conducted at flow temperatures between 1000°C and 1100°C with coupon angles of 20°, 50°, and 80°. The experimental design is a Central Composite Design (CCD) with repeated central tests [9]. The deposition data produced two empirical models for prediction as a function of near surface coupon temperature and coupon angle.

2.2.1 Virginia Tech Aerothermal Rig VTAR was donated by Rolls Royce (Indianapolis, IN) to Virginia Tech in 2010. The previous application prior to the donation was heat transfer studies of cascade turbine aerofoils [10–13]. The test equipment was repurposed for high temperature particle tracking studies by quantifying the coefficient of restitution [14,15]. The most recent study was for ARD deposit accumulation for very high sand loading and concentrations [16]. Figure 2-1 is a rendering of VTAR’s current configuration for sand deposition testing. Propane is the fuel source for the sudden expansion burner. The total air flow rate leaving the burner is about 0.06 kg/s depending on the flow temperature. The equilibration tube is constructed of stainless steel alloy 310 with an inner diameter of 7.62 cm. The distance from the sand injector point to the end of the equilibration tube inside the test section is about 2.22 m. The equilibration tube length allows a sufficient amount of time for the temperature of the sand particles to equalize with the flow temperature. The injector inside the equilibration tube is directed upstream to promote sufficient mixing. A constant flow velocity leaving the equilibration tube is 70 ± 2 m/s for all tests in this study to maintain consistency from the previous studies using VTAR [14–16]. The maximum flow temperature that can be achieved at the coupon is 1100°C.

Figure 2-1: The Virginia Tech Aerothermal Rig (VTAR) utilized for deposition testing between 1000°C and 1100°C, highlighting the test section cut-away view and the associated coupon.

The test sand is dried in an oven at 120°C for a minimum of 12 hours prior to a test. Most excess water trapped in the particulate evaporates, which prevents “clumping” during the injection process. Dry ARD is entrained in a flow separate from the main flow using a conveyor venturi injection system. Sand is placed on a conveyor belt inside a sealed box to prevent contamination from environmental dust and humidity levels. The conveyor belt moves and drops the sand into venturi vacuum pump. A conveyor scraper is used to passively ensure a majority of the sand drops into the pump. From system testing, mass losses are less than 0.01 grams from the conveyor into the vacuum venturi pump. To equalize the pressure inside the box due to the venturi vacuum pump, a particulate filter and an air drying system ensures that the local environmental humidity does not alter the sand consistency during testing. Secondary containment around the conveyor belt reduces potential agitation of the sand due to air movement in the primary containment box. After the sand moves through the venturi vacuum pump, the sand laden flow travels through a hose that is attached to the injector nozzle on the equilibration tube. The coupon angle illustrated in Figure 2-2 (LEFT), is the acute angle between the flow and the surface of the coupon where 0° is parallel and 90° is perpendicular to the flow. The coupon can be rotated in 10° increments along the vertical axis of the test section. Flow temperature is measured with three K type thermocouples placed along the horizontal midline next to the coupon leading edge as well as above and below the coupon while being centred horizontally, illustrated in Figure 2-2 (RIGHT). Due to the size and angle of the coupon relative to the equilibration tube, all the thermocouples are within the gas path. Two K type thermocouples are placed directly behind the coupon to measure near surface coupon temperature. The distance from the end of the equilibration tube to the leading edge of the coupon is approximately 10 cm. The distance decreases by about 2 to 3 cm during operations due to the thermal expansion of the tube into the test section via a slip joint. Hastelloy-X is the coupon material, which is the same from previous studies [14–16].

Figure 2-2: Two views of the coupon arrangement. (LEFT) Top-down view of coupon and equilibration tube with the test section casing and flanges hidden. The coupon angle, θ, is the acute angle between the bulk flow and the coupon surface. (RIGHT) An isometric view of the coupon and equilibration tube indicating thermocouple placement top and bottom of the coupon. The yellow dashed line across the coupon surface indicates the horizontal midline of the coupon and gas flow path.

2.2.2 Experiment Testing Conditions Independent variables (or predictors) for this study are bulk flow temperature and coupon angle. Deposited particles per area and percent coverage on the coupon surface by the deposits are the dependent variables. Table 2-1 provides a summary of the experiment test conditions. Flow temperature during injection is maintained within 5°C of the target temperature, either 1000°C, 1050°C, or 1100°C. Near surface coupon temperatures are measured for the subsequent statistical model, see Figure 2-2. The near surface and surface temperatures of the coupon vary depending on the coupon angle (20°, 50°, or 80°) due to various heat transfer mechanisms during testing inside the test section. Only 10 ± 0.005 grams of ARD are used per test. The coupon is composed of Hastelloy-X and is polished to a mirror-like finish before each test with a measured root mean square roughness of less than 0.2 μm to ensure surface roughness effects are minimized for deposition.

Table 2-1: Constant test conditions for varying flow temperature and coupon angle testing.

Bulk Flow Velocity 70 ± 2 m/s Desired Gas Path Temperature Range 1000°C, 1050°C, and 1100°C Particulate Type/Size ARD, 20 – 40 μm Injection Rate 0.0570 to 0.0646 g/s Total Sand Loading 10 ±0.005 grams Coupon Angles 20°, 50°, and 80° Coupon Material Hastelloy-X Coupon Surface Roughness ≤ 0.2 μm RMS Coupon Thickness 3.175 mm Particle Stokes Number ~1.6 (20 μm) to 6.8 (40 μm)

The 20 to 40 μm ARD batch used in all tests is ISO test dust grade from Powder Technologies Inc. The Stokes number for the experiment is between 1.6 and 6.8 depending on local velocity near the coupon surface. The mean diameter of the ARD is 26.58 μm with a median diameter of 27.3 μm. About 90% of the sand volume is between 20 and 40 μm, with the remaining 10% being less than 20 μm. Silicon and

21 aluminium oxides are the primary constituents of ARD. An earlier study determined that ARD has a sintering temperature (shrinking phase) of 1100°C for sizing less than 63 µm. Since the ARD is in a softened state, the sand particles may adhere to the Hastelloy-X surface [17]. The authors are aware that ARD is not an accurate analogue of airborne sands and dusts found across the world [18]. In addition, the particles typically entering the hot section of a turbine engine have a mean diameter less than 10 μm [19]. However, the ARD sizing used for this study improves the detection of deposited particles for post-test image processing and analysis as well as allows a comparison to previous and potentially future deposition studies at Virginia Tech [14–16].

VTAR’s sand injection system has been updated from the previous microparticle rebound and deposition studies [14–16] to allow for precise injection rate control that would be similar to turbine engine dust ingestion conditions. Dust concentrations ingested by turbine engines can be as high as 0.005 kg-sand per kg-air for extreme low visibility conditions [3]. Injection rates for this study are 0.0570 to 0.0646 g/s, for approximately three minutes, which correlates to about 0.001 kg-sand per kg-air. Sand concentration variation is directly correlated to the variation in mass flow rate per test through the burner. Based on the operational parameters of VTAR, to maintain the velocity at 70 m/s as temperature increases, the mass flow rate must decrease. Regardless of the mass flow rate of the gas path and the associated concentration, the equilibration tube allows the particles to attain thermal and momentum equilibrium with the main flow before impacting or being diverted around the coupon.

For each test, VTAR is brought up to steady-state flow conditions at the appropriate flow temperature and velocity. Flow temperature is set by the maximum temperature associated with each test angle (Figure 2-2). For instance, at angles of 20° and 50°, the maximum temperature is the leading edge temperature. However at 80°, the maximum temperatures are on the top and bottom of the coupon. Since the flow is being deflected around the coupon there are recirculation zones on the trailing and leading edges, which lowers the measured temperature at those locations. Due to the variation in flow temperature measurement locations and heat transfer mechanisms inside the test section, the primary variable compared to deposits is the average near surface coupon temperature (Figure 2-2). Experimental relationships for jet impingement onto a flat plate [20] determined the heat transfer coefficients to estimate coupon surface temperatures for each test angle. Using the heat transfer coefficient and estimating the radiative heat transfer, the variation of temperature between the surface and near-surface is less than 5°C depending on coupon angle for the experimental setup. Therefore, near surface temperature is used as an independent variable for the empirical modelling.

ARD deposit accumulation on the inner walls of the equilibration tube is negligible for all testing. Periodic borescope inspections of the equilibration tube indicate minimal deposit accumulation. Regardless, molten sand particles may adhere to the inside of the stainless steel equilibration tube during testing. However, during the cool down process after each test, any deposited material most likely delaminated from the equilibration tube and carried downstream. Since the equilibration tube has minimal deposits between tests, one test result should not affect a subsequent tests. Future studies will be necessary to accurately quantify any deposits that may occur on parallel flow surfaces using ARD.

2.2.3 Statistical Modelling Method A multiple linear regression model is developed for estimating initial ARD deposits at various coupon angles and near surface temperatures. There are a total of nine possible test combinations for flow temperature and angle. Primary control factors are the flow temperatures set at 1000°C, 1050°C and 1100°C and three coupon angles set at 20°, 50°, and 80°. A Central Composite Design (CCD) is the most popularly used experimental design for fitting a second order response surface [9]. A CCD is composed of factorial

22 points for all possible high and low combinations of the factors, centre points which provide information about curvature of the response surface, and axial points which allow for estimation of quadratic effects. Since there are two factors with three levels each, a cuboidal CCD assists with studying curvature effects of the responses. With a minimum of nine possible combinations, three additional tests are performed at the centre levels for coupon angle (50°) and flow temperature (1050°C) to ensure repeatability and curvature effects. By comparison, the CCD in this study uses 12 total tests, while a full factorial design with test triplication would require a minimum of 27 tests. The results from the CCD can be used for efficient estimation of first and second-order terms of the response surface of interest. The fully randomized test order for the CCD is generated by JMP®. Randomizing the order of the experimental trials reduces bias caused by nuisance factors, such as equipment warm-up effects and daily atmospheric conditions. Each ARD sample is also randomly assigned to each experimental test condition.

2.3 RESULTS AND ANALYSIS Deposits are quantified through microscopic images and processed to identify the normalized particle deposit tallies per square millimetre and percent coverage of the Hastelloy-X coupon surface. Deposits from both metrics are identified automatically and manually validated for each acquired image. The deposit data is averaged per test and combined to create two statistical empirical models. Heavy particle overlapping is observed for flow temperatures above 1050°C and 50° coupon angle.

2.3.1 Data Acquisition and Reduction Figure 2-3 (LEFT) is a comparison example of the coupon surface at 20x magnification before and after each test at 1050°C at 50°. Each microscopic image is the about 698 μm by 522 μm. Deposits metrics are average particles per area (particles per square millimetre) and average percent coverage. Quantifying the total deposits on the coupon surface uses a similar to the earlier deposition study at Virginia Tech [16].

Figure 2-3: (LEFT) Pre-test and post-test surface sample image comparison showing the deposits and Hastelloy-X surface. The image size is 698 μm by 522 μm at 20x magnification. (RIGHT) Rendered image of the coupon surface that highlights the locations of the sample image rows that are located at the midline, and the quarter and three quarter locations vertically along the horizontal axis.

The microscope system available requires manual acquisition of each location on the coupon. Based on the coupon orientation in the test section, three sample image rows are acquired across the horizontal axis of the coupon. Figure 2-3 (RIGHT) shows the location of each image series, which are at one quarter, one half, and three quarters the relative vertical distance of the coupon. Three sample image rows provide triplicate data sets for each test condition and validates the coupon exposure to the main gas path flow conditions. From each sample image row, approximately 50 to 60 images are acquired in relatively equal

23 spacing for each cord across the surface from the leading edge towards the trailing edge. An exact location cannot be determined with manual acquisition so equal spacing is assumed for all images across each series. Each image is acquired with the focus set to the Hastelloy-X substrate, which causes the particle deposits to be slightly out of the focus plane resulting in a distinct boundary between the deposit and the heterogeneous surface pattern. Using the pattern and texture differences from the Hastelloy-X surface and deposits on the coupon surface, each image is processed using several functions from the MATLAB® Visual Processing Toolbox. The image processing script is able to automatically identify most deposit particles and regions larger than 10 μm across for each image. Particle deposits less than 10 μm in diameter are disregarded based on the size distribution of the ARD supplied for this study. After the automatic selection process, each image is manually checked for incorrect identification and corrected if necessary. The combination of automatic and manual identification is capable of low variability and high repeatability between analysts. Delamination of the particles from the coupon after testing (during VTAR cool-down) and during image processing is not observed. The local bond between Hastelloy-X and SiO2, a primary constituent of ARD, is a strong atomic level bond that resists thermal cycles. Prior research has implied that a crystalline SiO2 layer between a nickel based alloy and a crystalline thermal barrier coating would provide the bond strength necessary to prevent delamination during thermal cycling [21].

2.3.2 Raw Data Table 2-2 is the testing order, near surface temperatures, and average deposit data for each test condition for all three sample image rows. As expected, deposits typically increase with increasing surface temperature and coupon angle.

Table 2-2: Aggregated deposit data for each test as well as the average near-surface temperature.

Average Flow Coupon Near-Surface Average Average Temperature Angle Temperature Particles Percent (°C) (°) (°C) per mm2 Coverage 1000 20 877 6 0.342 1000 50 879 13 0.806 1000 80 867 37 1.43 1050 20 911 49 2.20 1050 50 917 210 7.44 1050 50 917 167 6.27 1050 50 917 172 6.61 1050 50 916 155 6.84 1050 80 898 193 8.48 1100 20 951 139 6.63 1100 50 966 364 24.5 1100 80 934 335 21.9

Figure 2-4, correlates the deposit data with near surface temperature shows that the deposit trends increase with coupon angle. In addition, the 95% confidence intervals are plotted at each test result from the resulting deposit model discussed in the following section. The repeated tests at 1050°C and 50° coupon angle show very little variation between the four tests for both deposits per square millimetre and percent coverage area. Comparing the raw deposit data for particle deposits per square millimetre and percent coverage area, the maximum total deposits observed occurs at 1050°C flow temperature at 80° coupon angle. Near surface temperatures at 80° coupon angle are lower than their counterparts at 20° and 50°. Convective recirculation area behind the coupon and radiative heat transfer mechanisms result in lower observed substrate

24 temperatures. Specifically, at 80° coupon angle, there is a large stagnation region that develops in front of the coupon. The stagnation region can cause a large portions of the particles to be diverted around the coupon or decrease in velocity where they will not deposit. Reduced deposits at high angles and high temperatures has been observed before for a previous study at Virginia Tech that evaluated deposits under very high particulate loading at 1100°C and 90° coupon angle [16].

Figure 2-4: (LEFT) Average particle deposits per square millimetre on the coupon and (RIGHT) average deposit percent coverage area on the coupon. Both responses are plotted against near-surface coupon temperature. The average deposits per square millimetre is less than anticipated due to significant deposited particles overlapping on the surface. The 95% confidence intervals are shown for each test case.

Despite the underestimated particle deposits per area, a deposit per area estimated range is calculated using the percent coverage area data, Figure 2-5. Average deposits per square millimetre increase linearly while the percent coverage area increases at a quadratic rate. Deposits greater than 1050°C flow temperatures and at 50° coupon angles show significant particle overlaps and large deposit regions. Overlapping particles are relatively indistinguishable for the image processing scripts and manual validation, which implies particle counts per image are underestimated at higher surface temperatures and coupon angles. Assuming the maximum and minimum size of a potential particle deposit has a diameter between 20 to 40 μm, the lower and upper bounds are calculated. Particle deposits per square millimetre is within the estimate boundaries but the raw data rate of increase does not correlate to the estimated boundary rates. For this study, both average particle deposit and percent coverage area data indicate the necessity to use a combination of empirical models to predict the onset of deposits.

Figure 2-5: Particle deposits per area for the 50° and 80° coupon angle test cases with an estimated interval using the coverage area data.

2.3.3 Empirical Models for Prediction Average deposits per area and percent coverage area are the primary dependent variables for both statistical empirical models for prediction. The near-surface temperature and coupon angle are the predictors (or independent variables) for average deposits per area and percent coverage. Near surface temperature is converted from Celsius to Kelvin, for an absolute temperature scale. JMP® and MATLAB® statistical modelling results are compared and validated for the empirical prediction models. All predictors included in the models presented are statistically significant at the 0.05 level of significance. This implies that predictors in the final models are useful in predicting average deposits per area and percent coverage. The average particle deposits per square millimetre empirical model, Equation (2-1), has a strong correlation with both near surface temperature and coupon angle predictors. The deposits per area empirical model contains an interaction between near-surface temperature and coupon angle. This means that the effect of near-surface temperature on the average particle deposits per square millimetre depends on the coupon angle. The coefficient of determination and adjusted coefficient of determination, and , respectively, are greater than 90% (Table 2-3). The value is a measure of how well the empirical model approximates the deposit data points. A value of 1.0 means that the model fits the data perfectly. By contrast, the value accounts for the number of predictor variables and polynomial order within the model. For example, if there are too many predictors and/or higher order polynomials, random noise can be unintentionally modelled. Therefore, the value can assist in avoiding overfitting an empirical model for prediction.

= −4179.2 + 3.5266 + 3.132 + 0.045799 ( − 1185.4)( − 50) (2-1)

The average deposit percent coverage empirical model, Equation (2-2), has a stronger correlation for the near-surface temperature and coupon angle predictors compared to the Equation (2-1) model. The percent coverage model uses second order predictors for near surface temperature and coupon angle. Second order predictors are necessary to achieve high, and values, shown in Table 2-3. A quadratic response correlates with the average percent coverage trends in Figure 2-4. By comparison, the deposits per area data in Figure 2-4 are relatively linear for the 50° and 80° cases, which will overpower the quadratic response of the 20° tests. As mentioned in previously in Section 2.3.2 there is significant deposit overlapping for the

26 higher coupon angles and temperatures, that results in underestimation. Therefore, the percent coverage area prediction model is more representative of deposits compared to the deposits per area model at higher temperatures.

(2-2) = −269.63 + 0.22378 + 0.20534 + 0.004718 ( − 1185.4)(−50) + 0.0021594 ( − 1185.4) + 0.0014217 (−50)

Table 2-3: and values for particle deposits per area (Equation (2-1)) and deposit percentage coverage (Equation (2-2))

Particle Deposits per Area Deposit Percent Coverage 0.973 0.994 0.963 0.989

2.4 CONCLUSION This study developed two statistical prediction models with experimental data for the onset of ARD sand deposits on a flat Hastelloy-X coupon at high temperatures as quadratic functions of near surface coupon temperature and coupon angle. Flow temperatures are between 1000°C and 1100°C at 70 m/s, ARD is injected between 0.0570 to 0.0646 g/s, and coupon angles are set at 20°, 50°, and 80°. Deposit accumulation between 1050°C and 1100°C flow temperatures for coupon angles greater than 50° has significant particle deposit overlapping. High levels of particle deposit overlap result in an underestimating prediction model, Equation (2-1), about 1050°C. An average percent coverage area prediction model, Equation (2-2) is able to quantify deposit accumulation despite the high levels of particle overlap. Therefore, a combination of Equation (2-1) and Equation (2-2) is necessary to estimate surface deposits. In addition, particle deposit overlap and high substrate percent coverage indicates a shift towards utilizing alternative data processing and prediction models. Using mass or surface roughness analysis of a test specimen is a subsequent data acquisition method that has been used in other studies to establish a way of estimating deposit accumulation. This experiment can be expanded for future test scenarios. For instance, future experiments may involve particle tracking to characterize deposits as a function of near surface temperature and particle impact vectors. In addition, altering coupon temperature as an analogue to turbine cooled hardware can expand the current statistical model.

2.5 ACKNOWLEDGEMENTS The work presented here was supported by Rolls-Royce Plc., especially Brett Barker, Dr. Kwen Hsu, and Paul Davis as well as Virginia Tech’s Laboratory for Interdisciplinary Statistical Analysis (LISA).

2.6 REFERENCES [1] Gabbard, C. B., LeLevier, R. E., and Parry, J. F. W., 1982, Dust-Cloud Effects on Aircraft Engines - Emerging Issues and New Damage Mechanisms (U), , Marina del Rey. [2] Whittle, R., 2015, “Fatal Crash Prompts Marines to Change Osprey Flight Rules,” Break. Def. [Online]. Available: http://breakingdefense.com/2015/07/fatal-crash-prompts-marines-to-change- osprey-flight-rules/. [Accessed: 15-Aug-2016].

[3] Cowherd, C., 2007, Sandblaster 2 Support of See-Through Technologies for Particulate Brownout Task 5 Final Technical Report, Report No. 110565.1.005, U.S. Army Aviation and Missile Command, Arlington, Virginia, United States. [4] Smialek, J. L., 1991, The Chemistry of Saudi Arabian Sand - A Deposition Problem on Helicopter Turbine Airfoils, NASA TM-105234, NASA, Cleveland, OH. [5] Smialek, J. L., Archer, F. A., and Garlick, R. G., 1994, “Turbine airfoil degradation in the persian gulf war,” JOM, 46(12), pp. 39–41. [6] Kim, J., Dunn, M. G., and Baran, A. J., 1992, The “Most Probable” Dust Blend and Its Response in the F-100 Hot Section Test System (HSTS), DNA-TR-91-160, Defense Nuclear Agency, Alexandria, VA. [7] Song, W., Hess, K.-U., Damby, D. E., Wadsworth, F. B., Lavallée, Y., Cimarelli, C., and Dingwell, D. B., 2014, “Fusion characteristics of volcanic ash relevant to aviation hazards,” Geophys. Res. Lett., 41(7), pp. 2326–2333. [8] Hamed, A., Tabakoff, W. C., and Wenglarz, R. A., 2006, “Erosion and Deposition in Turbomachinery,” J. Propuls. Power, 22(2), pp. 350–360. [9] Myers, R. H., Montgomery, D. C., and Anderson-Cook, C. M., 2009, Response Surface Methodology: Process and Product Optimization Using Designed Experiments, , Wiley, Hoboken, NJ. [10] Turner, E. R., Wilson, W. D., Hylton, L. D., and Kaufman, R. M., 1985, Turbine Vane External Heat Transfer, Volume 1. Analytical and Experimental Evaluation of Surface Heat Transfer Distributions with Leading Edge Showerhead Film Cooling, NASA CR-174827, Indianapolis, IN. [11] Hylton, L. D., Nirmalan, V., Sultanian, B. K., and Kaufman, R. M., 1988, The Effects of Leading Edge and Downstream Turbine Vane Heat Transfer, CR-182133, NASA, Washington, D.C., United States. [12] Nealy, D. A., Mihelc, M. S., Hylton, L. D., and Gladden, H. J., 1983, “Measurements of heat transfer distribution over the surfaces of highly loaded turbine nozzle guide vanes,” J. Eng. Gas Turbines Power, 106(January 1984), pp. 149–158. [13] Hylton, L. D., Mihelc, M. S., Turner, E. R., Nealy, D. A., and York, R. E., 1983, Analytical and Experimental Evaluation of the Heat Transfer Distribution over the Surfaces of Turbine Vanes, CR- 168015, NASA, Washington, D.C., United States. [14] Delimont, J. M., Murdock, M. K., Ng, W. F., and Ekkad, S. V., 2015, “Effect of Temperature on Microparticle Rebound Characteristics at Constant Impact Velocity—Part II,” J. Eng. Gas Turbines Power, 137(11), p. 112604. [15] Delimont, J. M., Murdock, M. K., Ng, W. F., and Ekkad, S. V., 2015, “Effect of Temperature on Microparticle Rebound Characteristics at Constant Impact Velocity—Part I,” J. Eng. Gas Turbines Power, 137(11), p. 112603. [16] Boulanger, A. J., Patel, H. D., Hutchinson, J., DeShong, W., Xu, W., Ng, W. F., and Ekkad, S. V., 2016, “Preliminary Experimental Investigation of Initial Onset of Sand Deposition in the Turbine

Section of Gas Turbines,” ASME Turbo Expo 2016, Volume 1: Aircraft Engine; Fans and Blowers; Marine, ASME, Seoul, South Korea, p. V001T01A003. [17] Kueppers, U., Cimarelli, C., Hess, K.-U., Taddeucci, J., Wadsworth, F. B., and Dingwell, D. B., 2014, “The thermal stability of Eyjafjallajökull ash versus turbine ingestion test sands,” J. Appl. Volcanol., 3(1), p. 4. [18] Dunn, M. G., Padova, C., Moller, J. E., and Adams, R. M., 1987, “Performance Deterioration of a Turbofan and a Turbojet Engine Upon Exposure to a Dust Environment,” J. Eng. Gas Turbines Power, 109(3), p. 336. [19] Dunn, M. G., Padova, C., and Adams, R. M., 1987, Operation of Gas Turbine Engines in Dust- Laden Environments, ADP006197, Buffalo. [20] Livingood, J. N. B., and Hrycak, P., 1973, Impingement heat transfer from turbulent air jets to flat plates: A literature survey, TM X-2778, NASA, Washington, D.C., United States.

3 ASME TURBO EXPO 2017: EXPERIMENTAL BASED EMPIRICAL MODEL OF THE INITIAL ONSET OF SAND DEPOSITS ON HASTELLOY-X FROM 1000°C TO 1100°C USING PARTICLE TRACKING

Proceedings of ASME Turbo Expo 2017: Turbomachinery Technical Conference and Exposition ASME IGTI 2017 June 26-30, 2017, Charlotte, NC, USA

GT2017-64480

Andrew Boulanger, John Hutchinson, Wing F. Ng, Srinath V. Ekkad Virginia Tech, Mechanical Engineering Blacksburg, Virginia, United States of America Matthew J. Keefe, Weibin Xu Virginia Tech, Department of Statistics Blacksburg, Virginia, United States of America Brett J. Barker, Kwen Hsu Rolls Royce Corporation Indianapolis, Indiana, United States of America

ABSTRACT Deposition experiments are performed on a Hastelloy-X coupon using Arizona Road Test Dust (ARD). A statistical empirical model of the initial onset of ARD deposits is developed from the experimental data. The initial onset of deposits are a quadratic function of local surface temperature and impact velocity components (normal and tangential). A prominent observation is that tangential impact velocity has a significant non-linear, independent effect on deposits relative to normal impact velocity and local surface temperatures. All experiments use 20 40 µm ARD on a bare metal surface over a 1000-1100°C range. Unlike prior mass-based deposit studies, initial deposits for this study are quantified using a Coverage Ratio (CR), which is the area covered by deposits relative to the total area available. The empirical CR model has a strong correlation to coupon surface temperature and impact velocity vectors and is a foundation for future numerical or experimental model comparisons.

NOMENCLATURE

ARD Arizona Road Test Dust CCD Central Composite Design CFD Computational Fluid Dynamics COR Coefficient of Restitution CR Coverage Ratio ISO International Organization for Standardization PIV Particle Image Velocimetry

PPMW Parts Per Million by Weight SP Sticking Probability VTAR Virginia Tech Aerothermal Rig Area covered by deposits Available area for deposits Particle density Effective particle diameter Bulk flow velocity magnitude Gas dynamic viscosity Coupon face length Stokes Number Near-surface coupon temperature Normal impact velocity Tangential impact velocity

3.1 INTRODUCTION Sand and dust ingestion can result in deposits on turbine engine hot section components, which can degrade performance and ultimately result in premature engine failure. For example, on May 17, 2015, a fatal United States Marine MV 22 Osprey crash occurred in extreme dusty conditions while performing a hover maneuver. The post-accident report cited glassy deposits on the turbine section components as one of the direct causes of the engine failure [1]. Sand and dust ingestion events are not limited to military aircraft. Commercial aircraft deployment in the Middle East has generally increased with economic prosperity in recent years. The increasing risk of sand and dust ingestion forces a demand to improve engine resilience in arid environments. From full engine and hot section component testing, deposits typically depend on two primary factors, aerothermal flow conditions and the ingested particulate. Generally, deposits increase with flow and particulate temperatures due to material softening or melting. Particulate composition and sizing can have a strong impact on deposit initiation and development [2]. In addition, the physical location of deposit accumulation is influenced by local flow regimes [3]. A strong indicator of impact location is related to the Stokes number (), Eqn. (3-1), which indicates how quickly particulate will react to the flow and is given by

(3-1) = 18 where is the density of the particle, is the effective diameter of the particle, is the velocity at the pipe exit, is the gas dynamic viscosity, and is the length of the coupon face. For numbers greater than 1.0, particulate motion will be dominated by inertial effects and resist changing direction due to the flow field. numbers less than 1.0 indicate that the particulates will tend to follow the flow streamlines. Modeling deposition uses a combination of numerical and experimental studies in an effort to encompass the variety of conditions associated with particulate deposition. The critical viscosity [4] and the critical velocity [5] numerical models are commonly used to determine the likelihood of deposition. The critical viscosity model depends on the material composition and temperatures of the particulate and the substrate. By contrast, the critical velocity model depends on comparing the adhesion force to the kinetic energy of the particle to determine the likelihood of deposition. Recent promising studies of elastoviscoplasticity models account for the effect of normal impact velocity and particle size but neglect tangential velocity components on deposition rates [6,7].

Experimental comparisons and resulting modifications to numerical models has been an ongoing effort using test coupons to actual turbine engine hardware. For coupon testing, many recent studies employ bituminous coal ash with accelerated exposure tests to evaluate the accumulation of deposits. Within the context of the coal ash studies, deposit formation initiates around the melting point of coal ash, which is between 900°C and 1100°C [8]. Capture efficiencies are significantly higher for temperatures between 1200°C and 1400°C and can vary depending on coupon surface and flow temperatures independently [9]. Higher capture efficiencies occur with increasing mean coal ash particulate sizes due to the higher Stk numbers [10]. Coupons with thermal barrier coatings tend to develop deposits more readily than uncoated coupons and suffer from spallation due to deposit penetration and subsequent thermal cycling [11,12]. Recent film cooling schemes have shown a decrease of deposits across flat coupons [13]. With the strong experimental studies with coal-ash, there has been an increased interest with deposition associated with polycrystalline particulate, specifically ARD, to mimic sand ingestion events in arid environments. Like coal ash, ARD softening and melting points depend on sizing, heterogeneous composition, constituent crystalline structures, and heating rate [2,14,15]. Therefore, hot section sand deposits will vary depending on particulate variation due to geographic location. Recent ARD deposition studies on test coupons or turbine hardware are more applicable for aircraft ingestion events. Particle-surface interactions using ARD have shown that increasing temperature and impact velocity result in decreasing Coefficient of Restitution (COR) [16–18], Eqn. (3-2). COR is a measure of the particle-surface interaction and can be leveraged for erosion and deposition studies and is given by

(3-2) = where and are the incoming and rebounding velocity vector magnitudes, respectively. Decreasing COR generally correlates with increasing deposits depending on the softening point of the particulate. Test coupon ARD deposition studies have shown general agreement with coal-ash deposition studies, where increasing temperatures and normal impact velocity result in increasing deposits [19]. For actual turbine test hardware, external ARD (0-5 µm) deposits form near the vane stagnation regions and farther downstream towards the trailing edges due to ricocheted particles [20]. Previous studies have experimentally evaluated deposition using accumulated mass analyses and subsequently validated or modified numerical models. Deposition is a gradual process depending on aerothermal and particulate conditions. While previous studies have considered relatively large accumulations of deposits, this study investigates the initial stages of ARD deposits prior to relatively large mass accumulations using designed experiments followed by the development of statistical models. Since particle deposits are variable to various physical phenomenon, a statistical analysis method is employed to quantify the initial onset of deposits. Optical deposit identification is preferred to a mass or surface roughness analysis due to the small deposit quantities. Optical identification can provide metrics of individual particle deposit counts and Coverage Ratio (CR), which is the ratio of area covered by deposits over the total available area, Eqn. (3-3), given by

(3-3) = where is the area covered by deposits and is the relative total available area for deposits. The focus of this study is the metric since it provides information where rapid accumulation develops due to initial deposit agglomeration. Particle counts are not used for this study due to high particle overlap at the higher temperatures and oblique impact angles.

Overall, this study provides an experiment-based statistically strong empirical CR model intended for numerical model development. There is a complimenting numerical study [21], which proposes a Sticking Probability (SP) model dependent on the effects of temperature, normal and tangential velocities, and particle size. The SP model is able to leverage the empirical CR model and experimental data for validation.

3.2 EXPERIMENTAL METHOD

3.2.1 Test Equipment The Virginia Tech Aerothermal Rig (VTAR) was donated by Rolls Royce (Indianapolis, IN) to Virginia Tech in 2010. The previous application of the rig was for heat transfer studies of cascade turbine airfoils [22–25]. In recent years, VTAR was repurposed for microparticle impact and rebound studies at elevated temperatures [16–18, 26]. All tests for this study are conducted using the latest iteration of VTAR, which uses a propane fueled sudden-expansion burner, equilibration tube, and test section. Figure 3-1 depicts a rendering of VTAR’s current configuration for sand deposition testing. The combusted gases from the propane-fueled burner flow from the burner through an equilibration tube where the ARD particulate is mixed and accelerated towards the test section. The main gas path mass flow rate is about 0.06 kg/s to 0.075 kg/s depending on the desired testing temperature. The equilibration tube is constructed of grade 310 stainless steel with an inner diameter of 7.62 cm. The distance from the sand injector point to the end of the equilibration tube inside the test section is about 2.22 m, which allows a sufficient time for the temperature of the sand particles to equalize with the flow conditions.

Figure 3-1: The Virginia Tech Aerothermal Rig (VTAR) utilized for deposition testing, highlighting the test section cut-away view and the associated coupon.

Figure 3-2 provides rendering of the test coupon within the test section that highlights key features, such as the leading and trailing edges of the coupon. The distance from the end of the equilibration tube to the leading edge of the coupon is about 6 cm during hot operations testing. The support structure of the coupon allows for the coupon to be rotated in 10° increments along the vertical axis inside the test section. The coupon angle is the acute angle between the gas path and the surface of the coupon (Figure 3-2 (a)). Gas path flow temperature measurements are taken with three K-type thermocouples placed at the leading edge,

33 above, and below the coupon, which are within the primary gas path (Figure 3-2 (b)). Two K type thermocouples are placed in contact with the back of the coupon along the horizontal midline and equidistant from the coupon leading edge, trailing edge, and each other. The yellow dashed line across the coupon surface indicates the horizontal midline of the coupon and gas flow path.

Figure 3-2: Two views of the coupon arrangement. (a) Top-down view of coupon and equilibration tube with the test section casing and flanges hidden. (b) An isometric view of the coupon and equilibration tube with thermocouple placements.

3.2.2 Experiment: Design and Testing Conditions In order to study the onset of deposits, a designed experiment is used to collect data under different test cases. The designed experiment is a Central Composite Design (CCD) [27] for the gas path temperature and coupon angle combinations. There are three levels of flow temperature and three levels of coupon angle that result in nine testing combinations. Three additional tests at 1050°C at 50° establish test repeatability, which results in 12 total tests. A CCD maintains modeling strength while minimizing resource expenditures and allows for testing flexibility. In addition, a CCD can identify quadratic responses from the experimental factors.

Table 3-1 summarizes the test conditions for this study. The flow velocity leaving the equilibration tube inside the test section is 70±2 m/s for all tests to maintain consistency from the previous studies using VTAR [17–19]. Maximum flow temperature for this study is 1100°C. The test coupon is a flat rectangular metallic piece of Hastelloy-X (63.5 mm by 38.1 mm and 3.175 mm thick). Hastelloy-X is commonly used in hot section components due to excellent resistance to oxidizing and carburizing atmospheres. Direct surface temperature measurements, e.g. thermographic imaging, was not accessible due to equipment limitations. In addition, thermocouple measurements from the coupon face would interfere with the flow characteristics across the coupon during testing. One dimensional heat transfer calculations using jet impingement correlations indicate surface temperatures are about 10°C below the coupon surface depending on coupon angle [28]. For the purposes of analysis, near surface temperatures and surface temperatures are assumed to be synonymous. Ongoing and future studies will experimentally validate surface temperatures and near surface temperature measurements.

Table 3-1: Controlled test conditions for varying flow temperature and coupon angle testing.

Bulk Flow Velocity 70±2 m/s Desired Gas Path Temperature Range 1000°C, 1050°C, and 1100°C Particulate Type/Size ARD, 20 – 40 μm Injection Rate 0.06±0.004 grams/sec Total Sand Loading 10±0.005 grams Coupon Angles 20°, 50°, and 80° Coupon Material Hastelloy-X Coupon Surface Roughness ≤ 0.2 μm RMS Coupon Thickness 3.175 mm Particulate Stokes Number ~1.6 (20 μm) to 6.8 (40 μm)

The ARD batch for all tests was provided by Powder Technologies Inc. ARD is commonly used as a standard test dust for filtration in automotive and industrial equipment test. In addition, ARD is a reasonable analog for deposition studies as mentioned previously. The composition of ARD is assumed to be the same relative ISO (International Organization for Standardization) composition, per ISO 12103-01 A2, where silicon and aluminum oxides are the primary constituents.

Lofted sand concentrations in extreme rotor wash conditions (depending on airframe) can be as high as about 5000 PPMW near the rotor tips. Typically, lofted rotor wash is about 1000 PPMW depending on the airframe and particle sizing [29]. To approximately match 1000 PPMW, ARD is injected at a rate of 0.06±0.004 g/s. The resulting concentrations are between 800 PPMW and 1000 PPMW for all test conditions. The variation is due to the air mass flow rate through VTAR depending on the desired testing temperatures and velocity. In order to maintain constant velocity for all tests, comparably less mass flow is required for the high temperature cases than lower temperatures. For the bulk flow velocity of 70 m/s, the particle Stk number is between 1.6 and 6.8 for 20 µm to 40 µm ARD, respectively. Since the Stokes numbers are greater than 1.0, the ARD particulate will tend to have ballistic trajectories and impact the test coupon at angle similar to the preset coupon angle, but there is some deviation. Similar to prior testing, ARD particles soften around 1000 1100°C, which will increase the probability of deposition with gas path temperature [14,19].

To minimize ARD clumping during injection, 10 g samples are dried in a high temperature incubator at 120°C for a minimum of 12 hours. The dry ARD is entrained in a flow separate from the main flow using a conveyor venturi injection system. Sand is placed on a conveyor belt inside a sealed box to prevent contamination from environmental dust and humidity levels. The conveyor belt drops the sand into venturi vacuum pump. A conveyor scraper is used to passively ensure a majority of the sand drops into the pump. From system testing, mass losses are less than 0.01 g from the conveyor into the vacuum venturi pump. To equalize the pressure inside the box due to the venturi vacuum pump, a particulate filter and an air drying system ensures that the local environmental humidity does not alter the sand consistency during testing. Secondary containment around the conveyor belt reduces potential sand agitation due to air movement in the primary containment box. After the sand moves through the venturi vacuum pump, sand laden flow travels through a hose that is attached to the injector nozzle on the equilibration tube. Sand accumulation inside the various plumbing components prior to injection is insignificant per periodic inspections. In addition, sand agglomeration during the injection process are minimized due to venturi vacuum pump, which disperses the particles from forces induced by the acceleration and shear flow fields [30].

3.2.3 Empirical Deposition Model for Prediction The empirical CR model uses the local near surface coupon temperatures and particle impact vectors. Figure 3-3 shows the process diagram of the empirical CR model. Coupon temperatures and impact velocity vectors are the two independent variables and the dependent variable are deposits.

Figure 3-3: Flowchart for the empirical CR model for prediction across the test conditions in this study.

3.2.3.1 Near-Surface Coupon Temperatures Coupon temperatures are not assumed to be isothermal during testing. However, the coupon temperature profile is assumed to have a linear distribution across the horizontal plane for all test cases. A linear temperature regression function is integrated into the empirical CR model. An ongoing study using additional thermocouples confirms a relatively linear temperature distribution for 50° and 80° coupon angle test cases. Flow and near surface temperatures conditions are repeatable across all tests with less than ±5°C variation during the steady-state ARD injection process.

Particle Impact Vectors. Particle tracking data is compiled from Particle Image Velocimetry (PIV) images during the ARD injection timeframe. A twin head Litron Nd:YAG laser emits about 135 mJ at 532 nm wavelength through a quartz viewing window in the side of the test section. The laser sheet illuminates the particles across the horizontal coupon center line, yellow dashed line in Figure 3-2 (b). The PIV system is limited to a 7.4 Hz image pair acquisition rate, which can only acquire initial particle locations and trajectories at the bulk testing velocity in this study. The initial trajectories are a starting point for a Lagrangian particle tracking algorithm, which uses a Computational Fluid Dynamics (CFD) flow field from ANSYS Fluent® for the Eulerian phase of the calculations. The hybrid PIV CFD technique has been used in prior COR studies [16–18]. The impact vectors are collated into a regression model with bisquare weighting, which is described further in the Results section.

3.2.3.2 Particle Deposits Empirical CR data is quantified based on relative location across the coupon unlike the previous deposition study at Virginia Tech [19], which examined the average deposits across the coupon assuming an isothermal condition. CRs per test image are acquired via a semi-autonomous image processing method. Microscopic sample image rows are manually acquired using a Zeiss Axio Vert.A1 with an AxioCam MRc5. Figure 4 displays an example microscopic image comparison of the coupon surface at 20x magnification before and after a test with an image size equivalent to 698 μm by 522 μm. Sample image rows produce a deposit

36 profile for impact velocity vector comparison. Equal horizontal image spacing is assumed since the microscope requires manual manipulation. A repeatable physical method with consistent controls was used to ensure relatively equal spacing. The method produces between 50 and 60 images per sample row. Three sample image rows validate equal vertical deposit exposure. Figure 3-4 highlights the relative location of each image sample row (dashed yellow lines).

Figure 3-4: Pretest and posttest surface sample image comparison showing the deposits and Hastelloy-X surface with a rendered image of the coupon that highlights the locations of the sample image rows.

Each image is acquired with the focus set to the Hastelloy-X substrate, which causes the particle deposits to be slightly out of the focal plane of the microscope. As a result, a relative boundary between the deposit and the heterogeneous surface pattern is formed. Using the pattern, texture, and color gradient differences between the oxidized/carburized Hastelloy-X surface and deposits, each image is processed using functions from the MATLAB® Image Processing Toolbox. Most particle deposits larger than 10 μm in diameter are automatically selected for each image. Particle deposits less than 10 μm in equivalent diameter (based on deposit area) are disregarded based on the size distribution of the 20 μm to 40 μm ARD supplied for this study. In addition, deposits less than 10 μm in equivalent diameter were infrequent for all tests. All images are converted from color (RGB) to grayscale. The gradient of the grayscale images in combination with a Gaussian blurring effect can identify areas with low variation. Those areas are typically “blurred” in the original image due to the deposit being out of the focal plane. Blurred gradient images are then converted to black and white and then an automatic selection function identifies the potential particles. Each image is manually validated and manipulated to add or remove deposits. Variation between analysts due to the manual validation is less than 1% across a large randomly selected set of image samples. The combination of automatic and manual identification is capable of low variability and high repeatability between analysts.

3.3 RESULTS The empirical CR model is a function of three independent variables, which are near-surface coupon temperatures and impact vectors (normal and tangential). The resulting CR deposition model increases as a quadratic function of local coupon surface temperature and normal impact vectors with an adjusted value is 0.855. Tangential impact vectors have a significant effect on the resulting empirical CR model response.

3.3.1 Near-Surface Coupon Temperature The assumed linear temperature gradient across the coupon depends on the coupon angle (Figure 3-5). Surface temperature gradients decrease with increasing angle. Thermocouple locations are placed relative to the coupon structure, which causes the thermocouples to change location depending on the desired coupon angle. As a result, temperatures vary with location depending on the coupon angle and local flow conditions. Non-linear temperature profiles near the coupon edges may exist depending on coupon angle.

However, the linear temperature profiles are still able to explain most of the empirical CR model response with strong statistical significance. Future inclusion of non-linear temperature profiles will only improve the significant of the model response. Average surface temperatures are lower for the 80° cases most likely due to recirculation behind the coupon. A large recirculation zone behind the coupon for the 80° case may induce cooling effects along with the flow expansion inside the test section. In addition, the coupon back plate will be radiating heat towards the test section exhaust, which is significantly cooler than the side walls. By contrast, the 20° and 50° coupon cases have less recirculation and are radiating towards the test section side walls.

Figure 3-5: Linear local coupon near-surface temperature profiles for a 1050°C flow temperature at 20°, 50°, and 80° coupon angle.

3.3.2 Particle Impact Vectors The hybrid PIV CFD technique is used to correlate impact vectors relative to the location on the coupon using a robust regression model with bisquare weighting [31]. All impact vector regression models were developed using MATLAB® Statistical Modeling Toolbox with a combination of Cook’s Distance and a robust regression with bisquare weighting. Outliers are identified using Cook’s distance, which is the influence of each data point on a fitted response value. Generally, Cook’s distances greater than three times the mean Cook’s distance are regarded as outliers and subsequently down weighted or removed from the impact regression models. Some of the outliers can be attributed to the assumptions within the PIV CFD technique, such as spherical particles and uniform equivalent diameter. A robust regression model using a tuning factor of 2.0 closely corresponds to the impact data without removing outliers identified by the Cook’s Distance method for all the test cases. By comparison, the default tuning constant of 4.685 provide estimates that are 95% as statistically efficient as an ordinary least-squares estimate. Reducing the tuning factor increases the down weight assigned to large residuals that occur with particle impact vectors. The resulting impact vector components (normal and tangential) for all coupon angles have linear or quadratic responses across the coupon. Figure 3-6 and Figure 3-7 are examples of the resulting impact velocity vector regression models for a flow temperature of 1050°C. For all flow temperatures, normal impact velocity increases with coupon angle. Normal impact velocity trends are approximately linear,

38 except for the 80° cases. Tangential impact velocity trends decrease with increasing coupon angle. Both impact vector components have a significant effect on the empirical CR model.

Figure 3-6: Normal impact velocity regression models for the 1050°C flow temperature.

Figure 3-7: Tangential impact velocity regression models for 1050°C flow temperatures.

Due to the large stagnation region for the 80° test cases (Figure 3-8 (b)), peak normal velocity is offset slightly from the coupon leading edge. The stagnation region location of the 80° case is a caused by the orientation of the coupon relative to the exit of the equilibration tube. The projected coupon area parallel to the equilibration tube flow is larger for the 80° case than the 20° and 50° cases (Figure 3-8 (a)), which have smaller stagnation regions at their leading edges. The corresponding SP modeling study independently

39 validated the relative stagnation locations shown in Figure 3-8 [21]. A sensitivity analyses indicate that the impact vector variation for the 20° cases do not significantly impact the empirical CR model. However, a reduction in the variation (e.g. with more test data) should increase the model strength.

Figure 3-8: A CFD comparison between the coupon angles of 50° (a) and 80° (b) velocity magnitude flow fields at 1050°C.

Particulates that double impact can cause additional deposits for actual turbine vanes. There are studies suggesting that ricochets result in downstream deposits on the pressure side of a vane after impacting near the stagnation point [20]. Although particles that impact the coupon twice are not part of the primary analysis in this study, they have occurred with increased frequency for the 80° cases compared to 20° and 50° cases. The hybrid PIV CFD technique is able to identify the double impacts when particles are extrapolated forward and backward to the coupon face at different locations. As mentioned previously, the resulting impact regression models, for single and double impacts, use a combination of a Cook’s Distance method and a robust regression with bisquare weighting. Double impacts on the test coupon account for less than 1% of total impacts at 50°, and less than 10% at 80°. Figure 3-9 plots the impact velocity magnitude versus location on the coupon for the 1100°C at 80° case with the regressions and 95% confidence intervals (dashed lines). Single initial impacts are at higher velocities, about 60 m/s, and second impacts are at lower velocities, about 10 m/s. Deposits may occur upon secondary impacts, however, a future study will be necessary to examine the secondary effects of double impact deposits.

Figure 3-9: Highlights single and double impact particles for the 1100°C at 80° test case both using a robust regression model with bisquare weighting.

3.3.3 Particle Deposits In general, post-test microscopic imaging of each coupon indicates increasing CR with near surface temperature and coupon angle. High particle deposit overlap is observed above 1050°C flow temperature and 50° coupon angle. CRs of 0.2 or greater indicate heavy particle deposit overlaps. Figure 3-10 visually compares an example of heavy deposits between the leading and trailing edges from the 1100°C at 50° test. The probability that particulates will adhere to existing deposits increases with CR compared to the likelihood of deposit formation on a bare metallic surface. Estimating particle deposits per area is possible using CR but the variation is about ±50% using 20-40 μm ARD. Ongoing studies are evaluating lower total ARD loadings and the resulting CR with particle counts [21]. Some initial 5 g loading test data is available with the corresponding numerical SP model study [21]. However, this study focuses on the 10 g total ARD loading per test and the CR response.

Figure 3-10: Example post-test (1100°C at 50°) coupon surface comparison of deposits at the leading and trailing edges. There is high particle deposit overlap at the leading edge

The CR trend across the coupon surface depends on the coupon angle and the CR magnitude depends on flow and surface temperatures. For visualization purposes, Figure 3-11 is an example of CR regression models at 1100°C flow temperature for all tested coupon angles. The deposit regression model in Figure 3-11 is not applied to the empirical CR model, only the raw CR data is integrated. For the 20° and 50° cases, there is a high CR starting at the leading edge that decreases linearly across the coupon. Similar to the impact vector regressions, the 80° cases produce a parabolic profile with the highest deposits located near the stagnation region. The CR for the 80° cases are lower than the 50° cases for the first half of the coupon, which is most likely a direct result of the lower average coupon surface temperatures (Figure 3-5) and flow field stagnation region (Figure 3-8 (b)). Due to the stagnation region for the 80° cases, larger particles deposited with greater frequency due to a higher number. Smaller particle deposits in the stagnation region could be attributed to double impacts or diffusion. Testing of smaller particulate is intended for a future study.

Figure 3-11: Regression models of the CR per image at 1100°C across the coupon for all tested angles. Deposits for the 80° case are lower due to the lower surface temperatures compared to the 50° case.

There is minimal delamination of deposits from the coupon after testing (during VTAR cool-down) and during image processing. Prior research has implied that a crystalline SiO2 layer between a nickel based alloy and a crystalline thermal barrier coating would provide the bond strength necessary to prevent delamination during thermal cycling [32]. Since ARD is composed primarily of SiO2, the adhesion between both materials could be assumed to be strong enough to withstand thermal cycling depending on the local bond strength. A prior deposition study, using bare metal coupons and different particulates, has shown a rapid increase in deposits from surface roughness and deposit thickness [12]. Total deposits and accumulation rate change depending on surface exposure time. Under initial exposure, accumulation is slow, followed by a rapid exponential or quadratic increase, and finally stabilizes with a continuous non-linear increase in deposits. Comparatively, this experimental study has tested the initial particle-surface stage of deposition prior to rapid accumulation stage. The CR data can be leveraged in future studies to determine the point of initial rapid non-linear accumulation before the surface is saturated with deposits.

3.3.4 Empirical Deposit Model for Prediction The empirical CR model for prediction can be approximated using near surface coupon temperature and impact velocity vectors. The CR model, Eqn. (3-4), uses the estimated coefficients from Table 3-2. The near surface temperature uses Kelvin for an absolute temperature scale instead of Celsius. The near-surface temperature and tangential velocity interaction term is insignificant in the model. The and adjusted values are 0.856 and 0.855, respectively. All variables in the model hold a P-value less than 0.001. Model assumptions are met for the resulting empirical CR model. The model is checked for multicollinearity and all variance inflation factors are below 10. This ensures that the variance of the estimated regression coefficients is not inflated.

Table 3-2: Equation coefficients for the empirical CR model for prediction using near surface temperature (K) and velocity (m/s).

= 11.6 − 1.89(10 ) T −7.03(10 ) V −5.41(10 ) V (3-4) +6.54(10 ) TV +7.74(10 ) T −8.00(10 ) V +6.31(10 ) V

Contour plots visually represent the resulting empirical CR model at several near-surface temperatures and velocity conditions. The CR can range from 0.0 (no deposits) to 1.0 (complete surface saturation). Figure 3-12 and Figure 3-13 depict the model response for varying tangential velocity and near surface coupon temperature, respectively. The model estimates a maximum CR of about 0.4 for the range of conditions in this study. Deposit acceleration appears to initiate when CR values are greater than 0.2, most likely due to particulate deposit overlap. As anticipated from previous studies, CR increases with normal impact velocity and temperatures. Likewise, for low impact velocity (normal and tangential) but high near-surface temperature, deposits are unlikely. An interesting response is when tangential velocity increases, the CR begins to decrease throughout the model range. The effects of tangential velocity are significant on the resulting CR model and subsequent or SP model [21].

Figure 3-12: Contour plot of the empirical CR model for prediction at various near-surface temperatures. As near-surface temperature increases, the CR (and effective sticking probability) increases.

Figure 3-13: Empirical CR model for prediction at varying tangential impact velocities. As tangential velocity increases, the CR (and effective sticking probability) decreases.

The variables included in the empirical CR model explain about 85.5% (adjusted R2 of 0.855) of the variability in the response. The remaining 14.5% can be explained from other factors, such as non-linear coupon temperature effects. In addition, prior studies have shown that flow temperature and surface temperature directly impact deposit accumulation [9]. An ongoing study is decoupling the ARD temperature in the flow from the coupon surface temperature using coupon cooling methods. Acquiring

44 direct coupon surface temperatures is an ongoing effort using the current configuration of VTAR. The additional factors can increase the experimental model’s fidelity that may reduce the model variability and improve the confidence intervals. The deposition probability depends on the surface conditions. For example, a particle will be more likely to adhere to existing deposits than to a clean metallic surface. Experimentally or numerically determining the onset of deposits may require a combination of empirical modeling methods depending on the flow and surface conditions. For example, a clean surface (e.g. new turbine hardware) exposed to low particle loadings may be evaluated based on particle deposits per area using optical methods. After a low amount of deposits exist on the substrate, the analysis method could change to a CR metric. Finally, if the surface is nearing saturation or large counts of particle deposit overlaps, a mass based evaluation would be appropriate. The empirical CR model can assist in an overall deposit accumulation models on a hot-section turbine component for designers and analysts.

3.3.5 Validation Testing In order to assess accuracy of the empirical CR model, a test was conducted at 1050°C flow temperature and at a 60° coupon angle. The resulting data from this test is compared to the predicted results from empirical CR model. Figure 3-14 (a) shows the relationship between the observed and predicted CR with a 45° reference line to indicate perfect matches between observed and predicted values. These results show that the prediction is better for low CRs and that higher CRs tend to be underestimated for the relative test conditions. Figure 3-14 (b) provides information about the predictions relative to the location from the leading edge of the coupon. CR predictions are more accurate towards the center and trailing edge of the coupon (blue) but are underestimated near the leading edge for the first third of the coupon (red). The CR variation is still relatively low despite predicting almost double the observed values near the leading edge. A 0.05 to 0.1 variation in CR across a zero to unity scale can be considered reasonable in the context of this study. For example, some factors may be physical surface and particulate conditions as well as non-linear coupon temperature profiles. Table 3-3 provides several measures of prediction accuracy, including the average difference between observed and predicted values, mean squared error (0.15%), and mean absolute error (2.74%). All measures of accuracy indicate that overall predictions for the empirical CR model are close to the observed CR.

Figure 3-14: For the 1050°C flow temperature at 60° coupon angle, the predicted versus observed CRs for validation test (a) and observed minus predicted CRs by coupon location (b).

Table 3-3: Prediction accuracy based on validation testing.

3.4 CONCLUSIONS The experimental empirical CR model is intended to assist designers and analysts to determine deposit accumulation on hot-section turbine components. CR can determine the probability whether impacting particulate will adhere to existing deposits or to a clean substrate. The empirical CR model is a strong quadratic function of particle impact vectors and near surface temperature resulting with an adjusted R2 value of 0.855. Tangential impact vectors have a significant independent effect on deposits relative to normal impact velocity and local surface temperatures. For the range of test conditions, the maximum CR predicted by the model is 0.4 at high normal impact velocity and near surface temperature but low tangential velocity. Deposition appears to accelerate when CRs are greater than about 0.2 due to the increased probability that particles will readily adhere to existing deposits instead of the Hastelloy-X metallic substrate. Validation testing indicates that the overall predictions of the empirical CR model are relatively close to observed values. A corresponding study is using some of the initial test data to develop an SP model as a potential alternative to critical viscosity or velocity models [21]. Current studies at Virginia Tech are evaluating reduced sand loadings to extrapolate particle deposits per area in combination with CR metrics. Future studies intend to expand the current experimental testing process. Additional variables such as using a cooled coupon, lower sand loading, and alternate materials, will improve the fidelity of the empirical deposition models. Incorporating cooling schemes can provide a realistic comparison to modern turbine engines [13]. In addition, alternate particulates and test materials can establish categorical models depending on turbine engine design and operation localities. In general, the statistical methodology in this study can use a variety of variables and establish the most important factors, depending on physical conditions, that will ultimately improve turbine engine operations in austere environments.

3.5 REFERENCES [1] Whittle, R., 2015, “Fatal Crash Prompts Marines to Change Osprey Flight Rules,” Breaking Defense [Online]. Available: http://breakingdefense.com/2015/07/fatal-crash-prompts-marines-to-change- osprey-flight-rules/. [Accessed: 15-Aug-2016]. [2] Kim, J., Dunn, M. G., and Baran, A. J., 1992, “The ‘Most Probable’ Dust Blend and Its Response in the F-100 Hot Section Test System (HSTS),” DNA-TR-91-160, Defense Nuclear Agency, Alexandria, VA. [3] Kim, J., Dunn, M. G., Baran, A. J., Wade, D. P., and Tremba, E. L., 1993, “Deposition of Volcanic Materials in the Hot Sections of Two Gas Turbine Engines,” Journal of Engineering for Gas Turbines and Power, 115(3), p. 641. [4] Sreedharan, S. S., and Tafti, D. K., 2010, “Composition Dependent Model for the Prediction of Syngas Ash Deposition With Application to a Leading Edge Turbine Vane,” GT2010-23655, Volume 4: Heat Transfer, Parts A and B, ASME, Vol. 32, pp. 615–626.

[5] Brach, R. M., and Dunn, P. F., 1992, “A Mathematical Model of the Impact and Adhesion of Microspheres,” Aerosol Science and Technology, 16(1), pp. 51–64. [6] Adams, M. J., Lawrence, C. J., Urso, M. E. D., and Rance, J., 2004, “Modelling Collisions of Soft Agglomerates at the Continuum Length Scale,” Powder Technology, 140(3), pp. 268–279. [7] Fu, J., Adams, M. J., Reynolds, G. K., Salman, A. D., and Hounslow, M. J., 2004, “Impact Deformation and Rebound of Wet Granules,” Powder Technology, 140(3), pp. 248–257. [8] Jensen, J. W., Squire, S. W., Bons, J. P., and Fletcher, T. H., 2005, “Simulated Land-Based Turbine Deposits Generated in an Accelerated Deposition Facility,” Journal of Turbomachinery, 127(3), p. 462. [9] Laycock, R., and Fletcher, T. H., 2015, “Independent Effects of Surface and Gas Temperature on Coal Fly Ash Deposition in Gas Turbines at Temperatures up to 1400 °C,” Journal of Engineering for Gas Turbines and Power, 138(2), p. 21402. [10] Crosby, J. M., Lewis, S., Bons, J. P., Ai, W., and Fletcher, T. H., 2008, “Effects of Temperature and Particle Size on Deposition in Land Based Turbines,” Journal of Engineering for Gas Turbines and Power, 130(5), p. 51503. [11] Bons, J. P., Crosby, J. M., Wammack, J. E., Bentley, B. I., and Fletcher, T. H., 2007, “High-Pressure Turbine Deposition in Land-Based Gas Turbines From Various Synfuels,” Journal of Engineering for Gas Turbines and Power, 129(1), p. 135. [12] Wammack, J. E., Crosby, J. M., Fletcher, D., Bons, J. P., and Fletcher, T. H., 2008, “Evolution of Surface Deposits on a High-Pressure Turbine Blade—Part I: Physical Characteristics,” Journal of Turbomachinery, 130(2), p. 21020. [13] Ai, W., Murray, N., Fletcher, T. H., Harding, S., Lewis, S., and Bons, J. P., 2012, “Deposition Near Film Cooling Holes on a High Pressure Turbine Vane,” Journal of Turbomachinery, 134(4), p. 41013. [14] Kueppers, U., Cimarelli, C., Hess, K.-U., Taddeucci, J., Wadsworth, F. B., and Dingwell, D. B., 2014, “The Thermal Stability of Eyjafjallajökull Ash versus Turbine Ingestion Test Sands,” Journal of Applied Volcanology, 3(1), p. 4. [15] Whitaker, S. M., Peterson, B., Miller, A. F., and Bons, J. P., 2016, “The Effect of Particle Loading, Size, and Temperature on Deposition in a Vane Leading Edge Impingement Cooling Geometry,” GT2016-57413, ASME Turbo Expo 2016: Turbomachinery Technical Conference and Exposition, ASME, Seoul, South Korea, Vol. 5B, p. V05BT16A013. [16] Reagle, C. J., Delimont, J. M., Ng, W. F., and Ekkad, S. V., 2014, “Study of Microparticle Rebound Characteristics Under High Temperature Conditions,” Journal of Engineering for Gas Turbines and Power, 136(1), p. 11501. [17] Delimont, J. M., Murdock, M. K., Ng, W. F., and Ekkad, S. V., 2015, “Effect of Temperature on Microparticle Rebound Characteristics at Constant Impact Velocity—Part II,” Journal of Engineering for Gas Turbines and Power, 137(11), p. 112604. [18] Delimont, J. M., Murdock, M. K., Ng, W. F., and Ekkad, S. V., 2015, “Effect of Temperature on Microparticle Rebound Characteristics at Constant Impact Velocity—Part I,” Journal of

Engineering for Gas Turbines and Power, 137(11), p. 112603. [19] Boulanger, A. J., Patel, H. D., Hutchinson, J., DeShong, W., Xu, W., Ng, W. F., and Ekkad, S. V., 2016, “Preliminary Experimental Investigation of Initial Onset of Sand Deposition in the Turbine Section of Gas Turbines,” GT2016-56059, ASME Turbo Expo 2016, Volume 1: Aircraft Engine; Fans and Blowers; Marine, ASME, Seoul, South Korea, p. V001T01A003. [20] Lundgreen, R., Sacco, C., Prenter, R., and Bons, J. P., 2016, “Temperature Effects on Nozzle Guide Vane Deposition in a New Turbine Cascade Rig,” GT2016-57560, ASME Turbo Expo 2016, Turbomachinery Technical Conference and Exposition, ASME, Seoul, South Korea, Vol. 5A, p. V05AT13A021. [21] Barker, B. J., Hsu, K., Varney, B., Boulanger, A., Hutchinson, J., and Ng, W. F., 2017, “An Experiment-Based Sticking Model for Heated Sand,” GT2017-64421, Submitted for ASME Turbo Expo 2017, ASME, Charlotte, North Carolina, United States, pp. 1–11. [22] Turner, E. R., Wilson, W. D., Hylton, L. D., and Kaufman, R. M., 1985, “Turbine Vane External Heat Transfer, Volume 1. Analytical and Experimental Evaluation of Surface Heat Transfer Distributions with Leading Edge Showerhead Film Cooling,” NASA CR-174827, Indianapolis, IN. [23] Hylton, L. D., Nirmalan, V., Sultanian, B. K., and Kaufman, R. M., 1988, “The Effects of Leading Edge and Downstream Turbine Vane Heat Transfer,” CR-182133, NASA, Washington, D.C., United States. [24] Nealy, D. A., Mihelc, M. S., Hylton, L. D., and Gladden, H. J., 1983, “Measurements of Heat Transfer Distribution over the Surfaces of Highly Loaded Turbine Nozzle Guide Vanes,” Journal of Engineering for Gas Turbines and Power, 106(January 1984), pp. 149–158. [25] Hylton, L. D., Mihelc, M. S., Turner, E. R., Nealy, D. A., and York, R. E., 1983, “Analytical and Experimental Evaluation of the Heat Transfer Distribution over the Surfaces of Turbine Vanes,” CR-168015, NASA, Washington, D.C., United States. [26] Reagle, C. J., Delimont, J. M., Ng, W. F., Ekkad, S. V., and Rajendran, V. P., 2013, “Measuring the Coefficient of Restitution of High Speed Microparticle Impacts Using a PTV and CFD Hybrid Technique,” Measurement Science and Technology, 24(10), p. 105303. [27] Myers, R. H., Montgomery, D. C., and Anderson-Cook, C. M., 2009, “Response Surface Methodology: Process and Product Optimization Using Designed Experiments,” Wiley, Hoboken, NJ. [28] Livingood, J. N. B., and Hrycak, P., 1973, “Impingement Heat Transfer from Turbulent Air Jets to Flat Plates: A Literature Survey,” TM X-2778, NASA, Washington, D.C., United States. [29] Cowherd, C., 2007, “Sandblaster 2 Support of See-Through Technologies for Particulate Brownout Task 5 Final Technical Report,” Report No. 110565.1.005, U.S. Army Aviation and Missile Command, Arlington, Virginia, United States. [30] Gotoh, K., Masuda, H., and Higashitani, K., 2006, “Dispersion of Particles,” Powder Technology Handbook, Third Edition, CRC Press, pp. 449–464. [31] Huber, P. J., 1981, “Robust Statistics,” John Wiley & Sons, Inc., Hoboken, New Jersey, United States.

[32] Jarvis, E. A. A., and Carter, E. A., 2003, “Exploiting Covalency to Enhance Metal-Oxide and Oxide- Oxide Adhesion at Heterogeneous Interfaces,” Journal of the American Ceramic Society, 86(3), pp. 373–386.

4 EXPERIMENTAL STICKING PROBABILITY OF ARIZONA ROAD DUST ON HASTELLOY X IN ANALOGOUS HOT-SECTION TEMPERATURES

ABSTRACT Dust ingestion in gas turbine engines in austere environments is a serious concern for safe operation in commercial and military sectors. Identification and mitigation of deposits is necessary to prevent or minimize the risk to aircraft operation. This study statistically quantifies sticking probability of Arizona Road Dust particles in a combustion gas flow at 1000°C to 1100°C using direct independent variables and non-dimensional terms. Deposit data showed that individual particle sticking probabilities increased at a quadratic rate as a primary function of particle temperature, normal impact velocity, and particle size. However, secondary effects of substrate surface temperature and tangential impact velocity created non- linear sticking probability responses. Sticking probabilities reached a maximum of approximately 0.05 with median impact velocity magnitudes approximately 50 to 70 m/s. Particles between 19 µm to 40 µm had sticking probabilities range up to 0.01 while smaller particles between 10 µm to 19 µm would reach approximately 0.05. There are distinct non-linear responses depending on particle size that will need further investigation beyond this study. The empirical models developed are potentially a powerful tool for estimating particle deposit likelihood for existing and future engine designs.

NOMENCLATURE ARD Arizona Road Test Dust CFD Computational Fluid Dynamics EQLB Equilibration Tube HX Haynes International HASTELLOY® X Alloy PIV Particle Image Velocimetry PTCL Particle RMSE Root Mean Square Error VTAR Virginia Tech Aerothermal Rig

Equation Terms

Coefficient terms for sticking probability empirical model Ratio of the number of particles between 10 µm to 40 µm per ARD sample Ratio of the number of particles lost to deposition in the equilibration tube Ratio of the normal projected area from the test coupon , Ratio of the number of particles that hit the coupon for bin Coverage Ratio Equilibration tube inner diameter Particle diameter ® Dynamic modulus of elasticity for HASTELLOY X Impact efficiency for a region of interest Critical particle viscosity Particle velocity depending on flow temperature Number of particle injected into the equilibration tube Number of particles that impact a substrate at a specific location Number of particles that stick to a substrate at a specific location

Sticking Probability Stokes number Particle temperature (or flow temperature) Local coupon surface temperature Coupon angle Normal impact velocity Tangential impact velocity Impact velocity magnitude Coupon horizontal width Horizontal region between the two outermost coupon thermocouples

4.1 INTRODUCTION Ingestion of dust particulate for turbine engine applications has been studied intensely for several decades focusing on volcanic ash and arid region particulates. A variety of full scale engine tests in the last several decades have been performed to estimate operational endurance in dusty and austere environments [1,2]. Recently, higher combustor temperatures in turbine engines (for improved overall efficiencies) has created an increased likelihood of deposit formation on hot-section components. Various mitigation methods have been employed for military aircraft such as oiled particle filters or inlet particle separators. Unfortunately, these mitigation methods are not adaptable to all ingestion scenarios, such as the fatal MV-22 Osprey crash in 2015 during a training exercise [3]. Particulates can bypass filters and cause significant damage to primary and secondary flow passages. Many studies have evaluated these deposit processes by matching aerothermal conditions to test coupons or turbine vane sections. Typically, capture efficiencies are the primary metric for measuring deposits, which is a ratio of the particle mass ingested to the total deposited onto the test piece. Unfortunately, this type of method is specific to the available test equipment. This study is a reanalysis of the test data from Boulanger et. al. [4], which examined Arizona Road Test Dust (ARD) deposit formation on Haynes International HASTELLOY® X (HX). Deposits were originally quantified as Coverage Ratio (CR), which was a function of surface temperature and impact velocity vectors. However, CR is not directly capable to predict the likelihood an individual particle will deposit onto the surface. Since ARD particles are crystalline and have a relatively random fractal physical structure, deposits cannot be deterministically predicted when compared to homogeneous particulates of uniform shape. Therefore, probabilistic functions and statistical regressions for specific particulate material must be used to estimate the initial deposits onto a substrate. This study uses Sticking Probability (SP) instead of CR as a function of particle (or flow) temperature, surface temperature, impact vectors, and ARD particle size. SP, Eqn.(4-1), is defined as the ratio of particles that stick to a surface () to the number of particles that impact in the same location () and was conceptually introduced by Barker et. al. [5].

= (4-1)

The likelihood a particle will deposit to a surface is dependent on both impact kinematics and physical properties on the surface of the particle. Hertzian impact mechanics can be used to describe elastic collisions [6,7]. However, for many particle impact studies, inelastic collisions are more applicable. A variety of studies over the last century have examined the inelastic impact regime by quantifying elastoviscoplastic deformation and adhesion work forces [8,9]. For example, the Johnson-Kandall-Roberts (JKR) model included the effect of adhesion forces on an elastic sphere against a rigid surface [10]. Another impact model example is the Derjaguin-Muller-Toporov (DMT) model assumes Hertzian elastic

51 deformation with the included adhesion force effects [11]. In general, adhesion will occur between a particle and a surface as a function of liquid bridging, electrostatic, and Van der Waals forces. This study does not directly evaluate each of these factors but specifically focuses on SP for a specific set of controllable test conditions comparable to hot-section aerothermal conditions.

Many SP studies typically rely on computational modeling using accumulated mass experimental studies for validation. By comparison, this study semi-empirically determines the SP using a unique data reduction method that employs a hybrid PIV-CFD technique [12] and microscopic object identification algorithms [4] for individual particle deposits. Statistically quantifying individual ARD SP at similar conditions can provide invaluable information for engine designers to mitigate the risk of deposit formation on hot section components.

4.2 TESTING AND ANALYSIS This study focused on SP as a function of three primary independent variables, temperature (flow-particle and surface), impact velocity, and particle size. Data was reanalyzed from a previous study by Boulanger et. al. [4] and the test data portion from Barker et. al. [5]. Testing was performed using a propane fueled burner while entraining a sample of 20-40 µm ARD particles into the flow. The ARD used in all testing is classified as silt due to size the high quartz content and particle size from ISO 14688-1:2002. For comparison, sand is typically 62 µm to 2 mm, while silt is generally less than 62 µm [13]. Flow temperatures and surface temperature gradients affected both the particle and surface mechanical properties, specifically viscosity and dynamic elasticity. Particle impact vectors were established using a hybrid PIV- CFD technique similar to that used by Reagle et. al. [14]. The analysis methodology was redeveloped to quantify individual particle deposits using updated object recognition algorithms and determine impact efficiencies on a region of interest associated with the test coupon. Finally, some additional test data was collected at higher temperatures and higher impact velocities to help establish a wider range of data for the SP regression curves.

4.2.1 Test and Analysis Assumptions Practically examining ARD deposits onto the HX coupons required several assumptions that were leveraged throughout testing and analysis. The following list highlights the primary assumptions. 1) This study intends to examine only the onset of deposits focusing on the interaction between the coupon metallic surface and the particle deposit. 2) The Lagrangian particle tracking algorithms assumed spherical particles. 3) Heterogeneous ARD constituents were assumed to be evenly distributed. 4) Flow and particle temperatures are assumed to be the same just prior to impact due to the use of an Equilibration Tube (EQLB), discussed in the following section. 5) Dry particles are equally distributed in the flow via an air powered vacuum venturi pump, which would have enough fluid shear forces to separate particles mechanically held together.

4.2.2 Test Equipment Testing was performed using the Virginia Tech Aerothermal Rig (VTAR), which uses a propane fueled sudden-expansion burner, EQLB, and test section. Figure 4-1 illustrates the configuration for deposition testing. The combusted gases from the propane-fueled burner flow from the burner through an EQLB tube where the ARD particulate is mixed and accelerated towards the test section. The main gas path mass flow rate is between 0.06 kg/s to 0.075 kg/s depending on the desired testing temperature. The EQLB tube is constructed of grade 310 stainless steel with an inner diameter of 7.62 cm. ARD was injected into the upstream direction to allow for sufficient mixing and time for the particles to reach an equilibrium state

52 with the flow at the exit of the equilibration tube. The distance from the injector point to the end of the EQLB inside the test section is 2.22 m, which allows a sufficient time for the temperature of the particles to equalize with the flow conditions.

Figure 4-1: The Virginia Tech Aerothermal Rig (VTAR) utilized for deposition testing, highlighting the test section cut-away view and the associated coupon.

Figure 4-2 is a rendering of the test coupon within the test section from Figure 4-1. The test coupon is located approximately 6 cm from the exit of the EQLB during steady state hot operations testing. The coupon can be rotated in 10° increments along the vertical axis inside the test section. For reference, the coupon angle is the acute angle between the gas path and the surface of the coupon, as seen in Figure 4-2 (a). Measurements of gas, path flow, and temperature are taken with three K-type thermocouples. These thermocouples are placed at the leading edge, above, and below the coupon, which are within the primary gas path, as seen in Figure 4-2 (b). Three additional K-type thermocouples are placed in contact with the back of the coupon along the horizontal midline, shown as the dashed yellow line in Figure 4-2 (b). The thermocouples are placed equidistant apart with the middle thermocouple in the center of the coupon.

Figure 4-2: Two views of the coupon arrangement. (a) Top-down view of coupon and equilibration tube with the test section casing and flanges hidden. (b) An isometric view of the coupon and EQLB with thermocouple placements.

Optical access ports on the side and top of the test section to allow for the use of a PIV laser system for particle tracking prior to impact with the coupon. The PIV system consists of a twin head Litron Nd:YAG laser emits 135 mJ at 532 nm wavelength through a quartz viewing window in the side of the test section. The laser sheet illuminates the particles across the horizontal coupon centerline, as shown by the yellow dashed line in Figure 4-2 (b).

4.2.3 Updated Analysis Method The previous study by Boulanger et. al. [4] used CR as a primary metric for deposition as a function of surface temperature and impact vectors. CR provided a conceptual metric of accelerated deposit formation. When CR reached approximately 0.2, deposit formation accelerated due to the increased probability that particulate would be more likely to adhere to existing deposits than to a bare metallic surface. Accelerated deposit formation has been observed in numerous experimental studies [14,15] with a variety of substrates and existing deposits. Transitioning from CR to SP data reduction required reprocessing a significant portion of the raw data to identify individual particles from heavy overlapping particle deposits. All existing and new test data were reduced to derive the appropriate multi-linear regressions with reasonable statistical strength. Figure 4-3 explains the general flowchart of the data processing methodology for each individual deposition test. Each test produces a data set that correlates SP with impact vectors, particle size, surface temperature, and particle temperature prior to impact. Finally, all the test data are compiled into a single data set for a multi-linear regression.

Single Test

Data Acquisition

Flow and Particle Particle Surface Deposits Tracking Temperatures

Correlate Deposits with Impact Locations

Sticking Probability

Figure 4-3: High level data reduction process for each test conducted.

The data reduction process was optimized and modified from the previous study [4] to create the appropriate SP models. Data towards the leading and trailing edges of the coupon are ignored due to potentially non- linear surface temperature gradients. The regions outside of the first and last coupon thermocouples, shown as pink stars in Figure 4-4, are ignored due to potentially non-linear surface temperature gradients. The thermocouples are located at 17.5 mm and 46.0 mm from the leading edge respectively. As a result of ignoring edge data, the 20° test cases were discarded due to sparse impact data and subsequently deemed statistically insignificant. Temperature data between the thermocouples on the coupon surface is linear based on additional test cases.

Figure 4-4: Highlighted region of interest (in green) as well as the regions towards the outer edges of the coupon (in red).

For deposit data, an improved object recognition algorithm was developed to identify individual particles based on color differentiation. Individual particles can be identified with improved certainty than the previous study which relied upon CR as the primary dependent variable [4]. The color separated regions are assumed to be individual particles that conglomerate. Similar to the previous study [4], each microscopic image was manually validated. The author recognizes the imperfections of utilizing this method, but the large amount of image data collected across a variety of test conditions reduces the overall model variation. The crucial portion of the reanalysis is the correlation between the particle deposits and impact data from particle tracking. For this study, all particles are assumed to be a spheroid with relatively circular cross sections for both particle tracking methods and deposit quantification. For deposition, each individual particle deposit identified has an associated coverage area on the coupon surface, which can be correlated to an equivalent circular diameter. For this analysis, the deposited area is assumed to be similar to the cross section area of an ARD particle in the flow prior to impact. Effectively, the planar deformation of each particle upon impact is assumed to be negligible. Figure 4-5 shows the assumed shapes prior and after impact. There was minimal individual deposit spread unlike a splatter profile cold or warm spray processes [17] or other ash and silt deposition testing at higher testing temperatures [18].

Figure 4-5: Diagram of a simplified impact process depicting the assumption whereby the particle deposit diameter is equal to the particle diameter.

From all tests, the smallest equivalent particle size identified was approximately 10 µm. For reference, the 20-40 µm ARD sample from Powder Technologies Inc. is labeled based on particle size relative to volume distribution. However, a 20-40 µm sample has an actual particle size distribution of 0-40 µm based on number of particles. Since there were negligible deposits of particles less than 10 µm, this analysis only focuses on the particles that range in equivalent diameters of 10-40 µm. Figure 4-6 shows the five bins of particle sizes used for both the deposition and the particle tracking portions of the analysis. Each bin has an equal number of particles for a given sample of ARD. Equal size bins based on number of particles were used to directly compare with a nominal particle size. It is assumed that each particle size will have an equal chance after leaving the exit of the equilibration tube to hit, miss, and/or stick to the coupon.

Figure 4-6: Normalized particle size distribution based on number of particles that has been normalized between 10-40 µm for a 20-40 µm ARD sample.

Each deposit size must be correlated to an associated particle of a similar size in the flow prior to impact. However, using the hybrid PIV-CFD technique developed at Virginia Tech [18–20], the available test equipment is unable to distinguish particle size based on the refracted light from each particle. Therefore, semi-empirical method is necessary to estimate the effects of particle sizes on notional trajectories. The available PIV system (a twin head Litron Nd:YAG laser emitting 135 mJ at 532 nm with a FlowSense 4M camera) is able to establish the initial positions and velocities of the particles in the flow field. The subsequent Lagrangian particle tracking method depends on an assumed particle size to estimate a spherical drag coefficient for each tracked particle. Iterating through a nominal particle size (and associated mass) for each particle bin, will produce different trajectories and different impacts vectors on the coupon face. Figure 4-7 highlights the placement of the test coupon relative to the equilibration tube in the test section that will result in different particle impact trajectories depending on particle size. Each nominal particle size per bin from the Lagrangian particle tracking will correlate to an associated deposit size. For example, particle deposits with an equivalent diameter between 19.0 µm and 21.8 µm will correspond to particle impact trajectories that assume a size of 20.4 µm. As a result, individual particle deposits can be directly correlated to impact vectors along the coupon surface based on size. Similar to the prior study [4], impact vectors are reduced to regression models for each vector component and size. Therefore, each bin will have a normal and tangential impact regression model, totaling 10 impact regression models per test. Only the normal and tangential impact velocity regressions will be used in this study. At this point, there is a known number of deposits on the surface that correlate with particle impacts. In order to calculate SP, the number of particles that impact the coupon surface will need to be estimated based on the number of particles injected. Wall deposits on the EQLB tube were assumed negligible for this study due to physical restrictions to weigh the EQLB tube after each test. Figure 4-7 shows potential particle trajectories after exiting the EQLB tube. For most tests, over half the particulates would miss the coupon

58 due to the physical orientation of the coupon to the flow. Any particles flowing outside of the projected area normal to the coupon face was assumed to miss the coupon face entirely. For example, using Eqn. (4-5), a 50° coupon angle should cause 36% of the particles to miss the coupon entirely from the projected area normal to the bulk flow. Prior literature and numerical models have simulated impact efficiency (or hitting rate) depending on coupon orientation [14,21]. The particle tracking technique is able to identify impact efficiency for VTAR depending on particle size and orientation. For example, for a 50° coupon angle the impact efficiency for particles in Bin 5 will be higher than Bin 1. Since Bin 5 particles are larger and heavier, they will have a higher Stokes number (Stk) number and resist the change in direction due to the flow field compared to the smaller particles in Bin 1. Eqn. (4-2) is the governing equation for calculating impact efficiency for VTAR depending on the amount of ARD injected into the system. Each ratio term, , effectively reduces the total number of potential particles that may impact the coupon.

Figure 4-7: Illustration of the particles that will hit or miss a test coupon based on physical orientation to the flow within VTAR.

(4-2) , = , (4-3) =0.574 (4-4) =1

(4-5) = sin() (4-6) , =,

Where is the ratio of the number of particles from 10 to 40 µm from the ARD sample injected, is the assumed number of particles that do not deposit onto the equilibration tube walls, which is assumed to be negligible for this study. For geometric considerations, , Eqn. (4-5) is the ratio of particles that do not impact the region of interest where W is the coupon width, is the equilibration tube inner diameter, and is the coupon angle. Finally, Eqn. (4-6) is the impact efficiency (,) in the region of interest onto the coupon surface for each bin size in the region labeled in Figure 4-7.

Applying Eqn. (4-2) into Eqn. (4-1), can be calculated relative to the horizontal location on the coupon surface for each test. The additional variables of surface temperature, impact vectors, and particle deposits are all relative to locations on the HX coupon. Collating all the primary deposition factors as a function of location allows for the location term to be removed from the data set for each test case. After combining all the test case data sets, a multi-linear regression can be used to model SP under the range of test conditions for this study.

4.2.4 Additional Test Data In addition to the updated analysis method, additional test data was acquired after the study by Boulanger et. al. [4]. Both data sets were combined in this study and analyzed based on the methodology described in the previous section. During reprocessing, it was discovered that the 20° tests cases were deemed statistically insignificant. Fortunately, removing the 20° test cases proved negligible to the data reduction since deposition is a function of particle conditions upon impact. Several particles for the higher angle cases would experience some particle impact cases with angles near 20°. Particles with a lower Stk number will tend to impact the coupon at a shallow angle towards the trailing edge, which is similar to a high Stk number impact for a 20° case. Regardless, to help statistical significance, the previous validation test of 1050°C at 60° was incorporated into the data set. Table 4-1 lists the two additional tests performed after the previous study. To aid with particle identification on the coupon surface, a lower 5 g ARD loading was selected to ease particle identification on the coupon surface. As anticipated the lower loading produced proportionally less deposits, in this case, a decrease of 50%. Fortunately, the new data reduction method takes incorporates total particles injected and therefore, was determined to be independent of particle loading. Therefore, one can extrapolate that as long as the initial loading is carefully measured, SP should be consistent for all loadings as long as there is limited particle overlap to allow for the identification of individual particles.

Table 4-1: Additional testing performed to compliment previous test data [4].

Flow Temperature (°C) Coupon Angle (°) ARD Loading (g) 1050 60 10 1100 50 5 *1100 80 5 * This test was conducted at a bulk flow of 89 m/s

4.3 DATA REDUCTION AND ANALYSIS All the test data are collapsed into two primary statistical models, one based on the raw independent variables, and the second using non-dimensional terms developed for this study. Due to the context of the testing performed, all independent variables are not necessarily control variables. The distribution of independent variable values is a factor for statistical strength. Secondary effects such as “double bounce” particles or thermophoresis are not included as part of this analysis. It is assumed that deposits undergo a single impact and either deposit or bounce off the surface. The statistical regression models have a fairly strong correlation with values between 0.7 to 0.8 and low root-mean-square errors (RMSE). The raw independent variable statistical regression model is a piecewise equation and a function of flow-particle temperature, surface temperature, normal impact velocity vector, impact velocity magnitude, and particle diameter. Comparatively, the non-dimensional model is a function of three non-dimensional terms which are subsequently functions of material properties, impact vectors, and particle diameter.

4.3.1 Independent Variable Distributions Table 4-2 is the list of data ranges for each independent variable associated with the raw data from this study. Flow temperature () is close to the desired range between 1000°C (1273K) and 1100°C (1373K) while the range of surface temperature () is between 863°C (1136K) and 983°C (1256K). The range of surface temperature varies due to the coupon angle effect on heat transfer during testing and the assumed linear distribution of temperatures between the coupon thermocouples discussed previously. Normal impact velocity () and the magnitude of impact velocity () have a similar range, but their distributions vary depending on the tangential component. Particle diameter has a range between 10 µm and 40 µm but the binned particle sizes range between 14.5 µm to 34.3 µm to accommodate particle tracking, as shown in

Figure 4-6. Overall, SP is between 0 to 0.0559, with a varied range depending on the piecewise models developed.

Table 4-2: Data range for each primary variable with associated units.

Variable Minimum Maximum Units 1274.7 (1001.5) 1376.5 (1103.3) (°) 1136.4 (863.25) 1255.8 (982.65) (°) 21.589 88.657 / 22.406 89.065 / * 10.0 40.0 μ ** 0.0 0.0559 − * Nominal particle sizes used for particle tracking are 14.5 µm to 34.3 µm ** range encompasses all models

The distribution of impact velocity varied significantly despite maintaining and relatively constant bulk flow of 70 ± 2 m/s for most tests. Due to coupon angle, the impact trajectories and velocities deviated along the flow path near the coupon surface. As a result, the impact velocities were typically less than 70 m/s and would have larger tangential components farther from the leading edge. Velocities exceeding 70 m/s are the result of an additional test at 1100°C at 80° coupon angle shown in Table 4-1. Figure 4-8 are histograms showing a majority of impact velocity magnitudes are centered around 60 m/s with the normal component centered around 40 m/s. The tangential impact velocity has two peaks, the first centered around 10 m/s and the second near 45 m/s. The first tangential velocity peak corresponds to the 80° test cases, where the impact velocity decreases dramatically towards the leading and trailing edges due to the large stagnation and high pressure region on the front face of the coupon. The second peak corresponds to the 50° and 60° tests cases, where the particles will turn with the flow along the coupon surface but still have enough normal momentum to impact towards the trailing edge.

(a) (b) (c) 2200

2000

1800

1600

1400

1200

1000

800

600

400

200

0 0 255075100 02550600 255075100 Normal Impact Velocity (m/s) Tangenal Impact Velocity (m/s) Impact Velocity Magnitude (m/s)

Figure 4-8: Histogram of the normal (a), tangential (b), and impact velocity magnitude (c) for all tests conducted in this study.

Determining impact efficiency from particle impacts will vary depending on coupon angle and particle size as a function of number. The distribution of impact efficiencies was between 0 and 1 and can vary depending on the test application. Impact efficiencies in Figure 4-9 are the number of particles that impact the coupon within the projected area normal to the coupon face highlighted in Figure 4-7. As described previously in Eqn. (4-2), the impact efficiency is multiplied by the ratio of the projected normal area of the coupon relative to the size of the EQLB tube. Overall, a very small proportion of the particles will impact the surface compared to the original injected amount.

1200

1000

800

600

400

200

0 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Impact Eﬃciency within Projected Normal Coupon Area Figure 4-9: Histogram distribution of impact efficiency for particles within the projected normal coupon area after leaving the EQLB tube.

A majority of SP are less than 0.01 as shown by the distribution in Figure 4-10. SP does exceed 0.01 for specific cases, which will be discussed in the following sections. Although there is no direct experimental comparison of SP to available literature, a similar metric is capture efficiency. Capture efficiency is specific to the test equipment and conditions, and as such, cannot be directly corrected to the SP described in this study. Regardless, Fletcher et. al. [18] showed capture efficiencies for 5 µm coal ash at temperatures up to 1400°C and up to 200 m/s bulk velocity had a maximum capture efficiency of approximately 0.15. Crosby et. al. [15] using 3 µm coal ash at temperatures up to 1183°C and 170 m/s bulk velocity had capture efficiencies less than 0.10 for all testing. Thermal spray technologies (plasma, cold, warm) generally maximize the SP and capture efficiencies. Typically, higher temperatures and velocities are employed compared to this study to facilitate a higher SP. Qualitative comparison indicates that the relatively low SP for large ARD particles at 70 m/s bulk velocity is realistic. The crystalline and fractal shape of ARD centered around a critical temperature will result in a fairly low SP. In addition, the ARD is not fully molten [23] nor possesses adequate kinetic energy plastically deposit onto the HX coupon surface.

700

600

500

400

300

200

100

0 0 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.008 0.009 0.01 Scking Probability Figure 4-10: A majority of sticking probabilities were less than 0.01.

4.3.2 Sticking Probability Prediction Model

The sticking probability models developed in this study use a similar format to Tafti et. al. [22], which estimates sticking probability as a function of particle temperature and impact kinetic energy. The first model discussed in this section, is a regression of sticking probability as a function of flow temperature, local coupon surface temperature, normal impact velocity, impact velocity magnitude and particle size. Eqn. (4-7) and Eqn. (4-8) are the regression models for two particle size ranges. Particles between 10 µm and 19 µm have a significantly higher compared to larger and heavier particles between 19 µm and 40 µm. A single regression for particles ranging from 10 µm to 40 µm would have a low statistical significance. By separating the data into two regressions, the result is a stronger correlation for both size bins. Coefficients and statistical metrics for each regression are located in Table 4-3 and Table 4-4 respectively.

(4-7) 19.0 μ < ≤40.0 μ = + + + + + + + + + + + + +

(4-8) 10.0 μ < ≤19.0 μ = + + + + + + + + + + +

Table 4-3: Coefficients for Eqn. (4-7) and Eqn. (4-8)

3.0886 2.4290E+01 −1.5971(10 ) -7.3247E-03 −4.9061 -4.6438E+01 −1.6246(10 ) -1.0285E-03 0.43049 2.8132E+00 1.0406(10 ) 6.0768E-03 1.0771(10 ) -4.1028E-04 −5.4385(10 ) -2.7337E+00 −0.44274 1.0216E-03 −8.3556(10 ) 9.3452E-07 5.0452 2.2752E+01 2.6253(10 ) 9.3928E-02 2.1771 2.5699(10 )

Table 4-4: and Root Mean Standard Error (RMSE) for Eqn. (4-7) and (4-8)

Eqn. RMSE Particle Size Range (4-7) 0.7219 0.000966 19.0 μ < ≤40.0 μ (4-8) 0.8867 0.00389 10.0 μ < ≤19.0 μ

SP is highly sensitive to both flow temperature and particle size. Figure 4-11 and Figure 4-12 illustrate the predicted responses as a function of flow-particle temperature overlaid with sample data points and correlated to a nominal particle size. As mentioned previously, SP regressions were separated based on particle size. Figure 4-11 is the predicted response for particles between 19.0 µm and 40 µm while Figure 4-12 is the response between 10 µm and 19.0 µm. For the larger particle bin, SP increases with flow temperature and size with nominal constant values of surface temperature and impact velocity components. The smaller particle bin size has a predicted SP approximately seven times greater than the large particle bin sizes. Higher SP for smaller particles have been notionally observed under full turbine engine operation [24], which typically experience particles sizes between 0.001 µm to 10 µm. The results here show that the SP increases by almost an order of magnitude as particle sizes fall below 19 µm with increasing flow temperature.

Scking Probability

Figure 4-11: Sticking Probability versus flow-particle temperature for particles between 19 µm to 40 µm. Surface temperature, normal impact velocity, and impact velocity magnitude are median values from the data set. Scking Probability

Figure 4-12: Sticking probability versus flow-particle temperature for particles between 10 µm to 19 µm.

Since most modern turbines utilize some form of blade cooling, the difference between the particle temperature and the surface temperature can affect the particle adhesion strength [25]. For example,

Crosby et. al. [15] evaluated cooled test coupons with coal ash and determined that capture efficiency decreased with increased cooling but the study also observed reduced deposit spallation with increased coolant. The study implies there is a non-linearity where a reduced coupon surface temperature may increase and maintain deposits adhesion to the surface. Comparatively, Figure 4-13 and Figure 4-14 shows the predicted increases with temperature ratio (/). As temperature ratio approaches isothermal conditions, the SP increases at a quadratic rate. By contrast, decreasing temperature ratio (mainly through surface temperature), results in fewer deposits even with higher flow-particle temperatures. For both models based on particle size, increases quadratically with temperature ratio. However, there is a cluster of high data values with low temperature ratios that does not visibly match with the predicted response. Part of the difference is attributed to data being collapsed onto a 2D plot that does not take into account all impact velocities. In addition, each data point represents two temperatures, flow and surface, whereby the same temperature ratios can exist at 1050°C and 1100°C, where a higher flow temperature will produce a higher . Furthermore, the test data may also indicate that can increase with a decreased surface temperature similar to observations by Crosby et. al. [15]. Scking Probability

Figure 4-13: Sticking Probability versus temperature ratio (/ ) for particles between 19 µm and 40 µm. Normal impact velocity and the impact velocity magnitude are median values from the data set.

Diameter (µm) V = 30.56 m/s 14.5 n 0.05 V = 48.92 m/s mag

0.04

0.03

0.02

0.01

0 0.87 0.875 0.88 0.885 0.89 0.895 0.9 0.905 0.91 0.915 0.92 Temperature Rao

Figure 4-14: Sticking probability versus temperature ratio (/ ) for particles between 10 µm and 19µm.

The raw variable SP regression/prediction models provide direct comparable predictions for depending on impact velocity, temperatures, and particle size. However, SP can be analyzed in a non-dimensional format using the primary variables, which will be discussed in the following section.

4.3.3 Non-Dimensional Sticking Probability Model The non-dimensional models developed in this section are a function of three primary parameters. The parameters are based on the five primary variables discussed from the previous section. The parameters are treated as compounding variables to determine the resulting SP for a given set of impact and thermal conditions. Tafti et. al. [22] used a similar method to calculate SP as a function of particle viscosity ratio (based on temperature) and an impact energy ratio based on Coefficient of Restitution data from Delimont et. al. [20,21]. Several model iterations have determined that directly multiplying or combining probabilities to estimate a final SP was statistically inadequate. Therefore, two complex higher order regression equations were developed to empirically model the SP. The non-dimensional SP model, Eqn. (4-9), is a function of temperature, velocity, and size parameters derived from five primary variables. Eqn. (4-10) through Eqn. (4-12) are the non-dimensional parameters.

(4-9) = (Π, Π, Π)

(4-10) Π = (4-11) Π =

(4-12) Π =

Where Π, Π, and Π are the non-dimensional temperature, velocity, and size parameters, respectively. ARD viscosity () and critical viscosity () are the primary temperature dependent variables in Π. The viscosity terms are modified by logarithmic functions to create a linear probability with temperature, which ultimately improved the correlations with test data. Terms and are the normal and magnitude impact velocities, respectively. Finally, and are the particle diameters and HX dynamic modulus of elasticity, respectively.

The velocity parameter, Eqn. (4-10) is the ratio of normal impact velocity () to the total impact velocity magnitude (), which is a direct relationship that provides a relative probability due to kinematic effects. As tangential impact velocity increases the total velocity magnitude increases comparatively with normal velocity decreasing accordingly. If there is no tangential velocity component Π will increase towards unity, effectively increasing the . Various analytical and numerical models have demonstrated that higher normal impact velocity with low tangential components increase the particle kinetic energy transfer into adhesion work. Eqn. (4-11) is the temperature parameter, which is a ratio of a particles critical viscosity () to viscosity () developed by Tafti et. al. [22]. Viscosity was selected as a primary variable describing the physical state of the ARD due to the existence of the critical viscosity model in several recent publications by Tafti et. al. The authors recognize that using viscosity to describe the physical response of ARD crystalline particles is not necessarily physically accurate due to the sudden melting phase change of crystalline materials. Instead, the viscosity term is intended to provide a connection between computational models and experimental empirical results. In general, the temperature parameter decreases as a logarithmic function with particle temperature since critical viscosity () remains constant. The various ARD viscosities are calculated in 4.5 Appendix: Material Property Calculations. Eqn. (4-12) is the final non- dimensional term that accounts for particle size and the dynamic modulus of elasticity of HX. Similar to viscosity, the dynamic modulus of elasticity is a function of coupon surface temperature (). The critical viscosity and impact velocity magnitude terms were selected to non-dimensionalize the final term of Eqn. (4-12) since it provided the best fit associated with the final models.

Ideally, the non-dimensional terms would be independent probabilities that would combine for a predicted . Instead a multi-linear regression was optimized to match the test data available for this study. Eqn. (4-13) and Eqn. (4-14) are the models for same two particle bin sizes as discussed from the previous section. As discussed previously, a piecewise model was necessary to optimize the regression. Table 4-5 contains the coefficients for the equations and Table 4-6 are the associated statistical metrics for each regression. Both regressions have overall and coefficient P-values less than 0.005.

(4-13) = +Π + Π 19.0 μ < ≤40.0 μ + ΠΠ + ΠΠ + Π + Π +Π (4-14) = +Π + Π + Π + ΠΠ 10.0 μ < ≤19.0 μ + ΠΠ + ΠΠ + ΠΠΠ

Table 4-5: Coefficients for Eqn. (4-13) and (4-14)

0.0 −7.0170(10 ) 7.5999(10 ) 8.2609(10 ) −7.7045(10 ) 0.13955 3.0350(10 ) 1.2201 1.3581(10 ) -0.13753 −4.0864(10 ) -1.2884 −1.2263(10 ) -1.8615 0.46486 2.0156

Table 4-6: and Root Mean Standard Error (RMSE) for

Eqn. RMSE Particle Size Range (4-13) 0.7817 0.00065 19.0 μ < ≤40.0 μ (4-14) 0.7442 0.00588 10.0 μ < ≤19.0 μ

Compared to the direct variable prediction models described previously, similar quadratic trends emerge for SP depending on the non-dimensional variables. Figure 4-15 and Figure 4-16 are the predicted SP compared to Π. Since Π is a function of Π, see Eqn. (4-10) and (4-12), median values for normal impact velocity (), dynamic modulus of elasticity (), and critical viscosity () are included to provide relative context along with the median value of Π. As observed previously with the direct variable models, SP increases with normal impact velocity at a quadratic rate for particle sizes between 19 µm and 40 µm, shown in Figure 4-15. Particles less than 19 µm, Figure 4-16, have a nonlinear response depending on the Π value selected. For lower Π values, the SP increases with normal impact velocity. However, with increasing flow temperature, the SP is very high for lower impact velocities and then decreases with increasing normal impact velocity. One hypothesis is that the tangential velocity component may translate into rotational momentum upon impact. The rolling motion combined with a particle in a semi-molten state may reduce the local impact force but increase the time of impact thereby maintaining the same impact momentum. This hypothesis will require further studies that examine the microparticle impact motions. Notionally, a tangential velocity impact will impart a rotational motion to the particle, but that motion will depend strongly on the particle shape and the respective adhesion forces.

0.01 Diameter (µm) 0.009 20.4 23.3 0.008 26.7 34.3 0.007

0.006 V = 40 m/s n E = 145.1 GPa 0.005 s = 0.6985 MPa-s c 0.004 = 0.318 T 0.003

0.002

0.001

0 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 V

Figure 4-15: Sticking probability increases with for particles between 19 µm and 40 µm.

0.05

D = 14.5 µm 0.045 p V = 40 m/s = 1.0 n 0.04 T E = 144.9 GPa s = 0.6985 MPa-s 0.035 c = 0.6985 0.03 T Increasing Flow Temperature

0.025

0.02 = 0.318 0.015 T

0.01

0.005

0 0.40.50.60.70.80.91 V

Figure 4-16: Sticking probability versus for particles between 10 µm to 19 µm particles

For particles between 19 µm to 40 µm, Figure 4-17, SP versus Π shows a similar response to the direct variable model discussed previously for flow temperatures. SP generally increases with Π and with particle size beyond the critical point. As discussed previously, the SP response is highly dependent on Π and Π.

The example response shown in Figure 4-17 indicates that SP begins to decrease above a certain limit. The reduced SP rate can be due to secondary effects by the velocity (Π) and size (Π) parameters. As observed before, particles between 10 µm and 19 µm have a SP significantly greater than the larger counterparts. With the higher SP, Figure 4-18 shows a fairly linear response for median values of Π and Π as a function of Π.

0.01 Diameter (µm) V = 40 m/s 0.009 20.4 n V = 55 m/s 23.3 mag Crical Viscosity 0.008 26.7 = 0.7273 34.3 V E = 145.1 GPa s 0.007 = 0.6985 MPa-s c 0.006

0.005

0.004

0.003

0.002

0.001

0 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 T

Figure 4-17: Sticking probability increases pseudo-logarithmically with for particles between 19 µm to 40 µm.

0.05 Diameter (µm) 0.045 14.5 V = 40 m/s 0.04 n V = 55 m/s mag Crical Viscosity = 0.7273 0.035 V E = 144.9 GPa s 0.03 = 0.6985 MPa-s c = 0.05468 0.025 D

0.02

0.015

0.01

0.005

0 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 T

Figure 4-18: Sticking probability increases with for 10 µm to 19 µm particles Both non-dimensional models are able to predict SP based on three primary non-dimensional parameters within the testing conditions of this study. In general, SP increases with the temperature parameter (Π) and velocity parameter (Π), with the latter having an attenuated effect. The most notable response is the distinctly higher SP for smaller particles compared to the larger counterparts. This response is counterintuitive based on the Π parameter whereby smaller particles should have a lower SP. Although the authors were able to empirically model SP for smaller particles using Π, there is still a potential need to examine additional variable effects such as the effects of the ARD crystalline constituents. Generally, increasing temperature for an amorphous particulate would usually indicate decreasing viscosity or increased softening of individual particles. However, as discussed previously, ARD is composed of primarily crystalline constituents where particle viscosity and melting occurs suddenly relative to temperature. For example, quartz, a primary constituent in ARD, has a melting temperature around 1700°C with notable crystalline phase changes such as -quartz to -quartz around 500°C but with fairly minimal changes in material softening. The other ARD crystalline constituents would most likely undergo a similar process with temperature. Nonetheless, ARD deposits were observed at temperatures far below individual constituent melting temperatures. As a result, for practical purposes, the viscosity terms in Π can be assessed empirically as a logarithmic function of particle temperature until an alternative model can be developed. Experimental evidence by Kueppers et. al. [23], while studying volcanic ash from Eyjafjallajökull, demonstrated that ARD less than 63 µm melted at approximately 1200°C. For comparison, the lowest burner outlet temperature for all testing near the sand injector was approximately 1200°C. The lower ARD melting temperature was hypothesized by Kueppers et. al. [23] to be due to the heating process where the other constituents may break down around above 700°C and release volatiles such as water vapor (from clay) and carbon dioxide from calcite. For practical purposes to understand ARD changes, a follow-up study should experimentally examine the transient heating process of ARD relative to particle size in a

72 combustion gas environment. Sampling the ARD at various points during the heating process similar to VTAR may provide useful data to the changes in particulates crystalline structures in the presence of other constituents. Likewise, examining ARD under varying flame exposures could provide useful insight with regards to the state of the material prior to entering a turbine section. In addition, unlike the study by Keuppers et. al. [23], the ARD sample is in an aerosol form that will respond differently than heating material in a crucible. Therefore, the composition and crystalline structure changes will be dependent on the thermal and chemical combustion gas interactions.

4.4 CONCLUSIONS This study has developed two significant non-linear piecewise statistical regressions, using dimensional and non-dimensional terms, of ARD SP on a HX test coupon between 1000°C to 1100°C. Both models are able to account for 70% to 80% of the predicted response based on the primary independent variables and non-dimensional terms. The maximum predicted SP was approximately 0.01 for particles between 19 µm and 40 µm. SP for particle sizes between 19 µm to 40 µm increased at a quadratic rate primarily with particle temperature and size. Coupon surface temperature and tangential impact velocity effects are secondary, but do have a notable impact on overall SP. Smaller particles between 10 µm to 19 µm had a maximum predicted SP of approximately 0.05. Like the larger particles, SP had a quadratic response across the various parameters. SP for smaller particles is notably higher than the larger counterparts. In some cases, SP can be 10 times greater than larger particles. The higher for the smaller particle size bin is not unusual since deposit formation occurs for particles less than 10 µm based on observations from actual turbine engine ingestion events. The non-dimensional parameters developed for the second set of SP regression models were originally intended to be a combination of probabilities. For particles between 19 µm to 40 µm, each parameter can be compounded to reasonably predict SP. However, a piecewise model was necessary to account for the higher SP of particles between 10 µm to 19 µm. A single empirical SP model with non-dimensional parameters was unable to account for the full range of particle sizes selected for testing. The higher values of SP for smaller particles indicates that there are additional parameters that should be addressed especially for smaller particles. For example, the large variation in SP data could be caused by high turbulent flow prior to impact compounded by the lower particle Stk numbers. Prior studies have noted that turbulent eddy diffusion can cause turbine vane deposits due to various flow features. Adding turbulence parameters for smaller particles should assist in quantifying the large variation in SP data. Another additional parameter could quantify the mechanical and chemical processes for crystalline particle impacts. The physical state of small crystalline particles mixtures like ARD has a distinctly different thermos-mechanical response than the respective individual constituents. The addition of turbulence and material parameters could provide an improved empirical SP model for a large variety of crystalline particulates. Overall, this study was intended to provide a method for compounding several known variables affecting deposition and reducing them down to a new set of non-dimensional parameters. Empirical models of ARD SP between 10 µm to 40 µm were statistically significant, but particles between 10 µm and 19 µm will need further testing and analysis to encapsulate additional parameters with decreasing particle size. Future work should focus on crystalline particles less than 10 µm and the subsequent deposit formation.

4.5 APPENDIX: MATERIAL PROPERTY CALCULATIONS The viscous and elastic properties of ARD and HX are direct functions of temperature. Each property can be calculated as either a function of flow temperature or coupon surface temperature. The viscosity of ARD is calculated using the same method as Tafti et. al. [22]. The viscosity of ARD depends on a critical

73 temperature and the quantity of oxide constituents. The critical temperature is determined by Eqn. (4-15) from Yin et. al. [26].

= 100 (10.75 +13.03 −5.28 −5.88 −10.28 ) +3.75 (4-15) + 453

Where each constituent is the molar fraction within the sample of ARD. Based on the ISO composition of ARD, the critical temperature is 1089.6°C. The critical temperature provides an inflection point for calculating the viscosity as a function of temperature. Eqn. (4-16) is the regression relationship based on test data for viscosity as a function of temperature and composition [27].

(4-16) log = + 10 10

Where is viscosity in Pa-s and is temperature in Kelvin. Terms and are determined from composition of oxides and temperatures. The critical viscosity from the critical temperature determines the points between a low and high temperature regime for viscosity. The critical velocity calculated is 6.958(10) -. Figure 4-19 depicts the logarithmic relationship between viscosity and temperature for the range of flow temperatures.

Noonal ARD Viscosity 108

107

106

105 1280 1290 1300 1310 1320 1330 1340 1350 1360 1370 Temperature (Kelvin)

Figure 4-19: Viscosity of ARD for the range of flow/particle temperatures in this study.

The dynamic modulus of elasticity for HX used a regression of available data from Haynes International® [28]. The data was curve-fit, producing Eqn. (4-17) and subsequently plotted for the appropriate testing temperature range in Figure 4-20. The dynamic modulus of elasticity is a direct function of surface temperature that was acquired during testing. Surface temperatures ranged between 1150K and 1250K (~875°C and 975°C) which results in a dynamic modulus of elasticity between 139 GPa to 154 GPa.

= −37,437.6 −11,330,632.63 + 211,480,830,511.59 (4-17)

HASTELLOY® X Young's Modulus Regression 154

152

150

148

146

144

142

140

138 1100 1150 1200 1250 Temperature (K)

Figure 4-20: HASTELLOY® X dynamic modulus of elasticity for the range of surface temperatures in this study.

4.6 REFERENCES [1] Dunn, M. G., Padova, C., and Adams, R. M., 1987, “Operation of Gas Turbine Engines in Dust- Laden Environments,” ADP006197, Buffalo. [2] Barnstorff, K., 2015, “NASA Studying Volcanic Ash Engine Test Results,” pp. 1–20 [Online]. Available: http://www.nasa.gov/feature/langley/nasa-studying-volcanic-ash-engine-test-results. [Accessed: 01-Jan-2016]. [3] Whittle, R., 2015, “Fatal Crash Prompts Marines to Change Osprey Flight Rules,” Breaking Defense [Online]. Available: http://breakingdefense.com/2015/07/fatal-crash-prompts-marines-to- change-osprey-flight-rules/. [Accessed: 15-Aug-2016]. [4] Boulanger, A. J., Hutchinson, J., Ng, W. F., Ekkad, S. V., Keefe, M. J., Xu, W., Barker, B. J., and Hsu, K., 2017, “Experimental Based Empirical Model Of The Initial Onset Of Sand Deposits On Hastelloy-X From 1000°C To 1100°C Using Particle Tracking,” GT2017-64480, ASME Turbo Expo 2017: Turbomachinery Technical Conference and Exposition, ASME, Charlotte, North Carolina, United States, p. V02DT48A015-V02DT48A015. [5] Barker, B. J., Hsu, K., Varney, B., Boulanger, A., Hutchinson, J., and Ng, W. F., 2017, “An Experiment-Based Sticking Model for Heated Sand,” ASME Turbo Expo 2017, pp. 1–11. [6] Hertz, H., 1881, “On The Contact of Elastic Solids,” Journal fur die reine und angewandte Mathematik, 92, pp. 156–171. [7] Hertz, H., 1882, “On The Contact of Rigid Elastic Solids and on Hardness,” Verhandlungen des Vereins zur Beförderung des Gewerbefleisses.

[8] Brach, R. M., 1991, “Restitution in Point Collisions,” pp. 168–181. [9] Brach, R. M., and Dunn, P. F., 1992, “A Mathematical Model of the Impact and Adhesion of Microspheres,” Aerosol Science and Technology, 16(1), pp. 51–64. [10] Johnson, K. L., Kendall, K., and Roberts, A. D., 1971, “Surface Energy and the Contact of Elastic Solids,” Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 324(1558), pp. 301–313. [11] Derjaguin, B. ., Muller, V. ., and Toporov, Y. ., 1975, “Effect of Contact Deformations on the Adhesion of Particles,” Journal of Colloid and Interface Science, 53(2), pp. 314–326. [12] Reagle, C. J., Delimont, J. M., Ng, W. F., Ekkad, S. V., and Rajendran, V. P., 2013, “Measuring the Coefficient of Restitution of High Speed Microparticle Impacts Using a PTV and CFD Hybrid Technique,” Measurement Science and Technology, 24(10), p. 105303. [13] Assallay, A., 1998, “Silt: 2–62 Μm, 9–4φ,” Earth-Science Reviews, 45(1–2), pp. 61–88. [14] Reagle, C. J., Delimont, J. M., Ng, W. F., Ekkad, S. V., and Rajendran, V. P., 2012, “A Novel Optical Technique for Measuring the Coefficient of Restitution of Microparticle Impacts in a Forced Flowfield,” ASME Turbo Expo 2012, Volume 7: Structures and Dynamics, Parts A and B, ASME, p. 1. [15] Crosby, J. M., Lewis, S., Bons, J. P., Ai, W., and Fletcher, T. H., 2007, “Effects of Particle Size, Gas Temperature and Metal Temperature on High Pressure Turbine Deposition in Land Based Gas Turbines From Various Synfuels,” Volume 4: Turbo Expo 2007, Parts A and B, ASME, Montreal, Canada, pp. 1365–1376. [16] Lewis, S., Barker, B. J., Bons, J. P., Ai, W., and Fletcher, T. H., 2011, “Film Cooling Effectiveness and Heat Transfer Near Deposit-Laden Film Holes,” Journal of Turbomachinery, 133(3), p. 31003. [17] Vardelle, M., Vardelle, A., Leger, A. C., Fauchais, P., and Gobin, D., 1995, “Influence of Particle Parameters at Impact on Splat Formation and Solidification in Plasma Spraying Processes,” Journal of Thermal Spray Technology, 4(1), pp. 50–58. [18] Laycock, R., and Fletcher, T. H., 2015, “Independent Effects of Surface and Gas Temperature on Coal Fly Ash Deposition in Gas Turbines at Temperatures up to 1400 °C,” Journal of Engineering for Gas Turbines and Power, 138(2), p. 21402. [19] Reagle, C. J., Delimont, J. M., Ng, W. F., and Ekkad, S. V., 2014, “Study of Microparticle Rebound Characteristics Under High Temperature Conditions,” Journal of Engineering for Gas Turbines and Power, 136(1), p. 11501. [20] Delimont, J. M., Murdock, M. K., Ng, W. F., and Ekkad, S. V., 2015, “Effect of Temperature on Microparticle Rebound Characteristics at Constant Impact Velocity—Part II,” Journal of Engineering for Gas Turbines and Power, 137(11), p. 112604. [21] Delimont, J. M., Murdock, M. K., Ng, W. F., and Ekkad, S. V., 2015, “Effect of Temperature on Microparticle Rebound Characteristics at Constant Impact Velocity—Part I,” Journal of Engineering for Gas Turbines and Power, 137(11), p. 112603.

[22] Singh, S., and Tafti, D. K., 2015, “Particle Deposition Model for Particulate Flows at High Temperatures in Gas Turbine Components,” International Journal of Heat and Fluid Flow, 52, pp. 72–83. [23] Kueppers, U., Cimarelli, C., Hess, K.-U., Taddeucci, J., Wadsworth, F. B., and Dingwell, D. B., 2014, “The Thermal Stability of Eyjafjallajökull Ash versus Turbine Ingestion Test Sands,” Journal of Applied Volcanology, 3(1), p. 4. [24] Hamed, A., Tabakoff, W. C., and Wenglarz, R. A., 2006, “Erosion and Deposition in Turbomachinery,” Journal of Propulsion and Power, 22(2), pp. 350–360. [25] Kuroda, S., Fukushima, T., and Kitahara, S., 1992, “Significance of Quenching Stress in the Cohesion and Adhesion of Thermally Sprayed Coatings,” Journal of Thermal Spray Technology, 1(4), pp. 325–332. [26] Yin, C., Luo, Z., Ni, M., and Cen, K., 1998, “Predicting Coal Ash Fusion Temperature with a Back-Propagation Neural Network Model,” Fuel, 77(15), pp. 1777–1782. [27] Senior, C. L., and Srinivasachar, S., 1995, “Viscosity of Ash Particles in Combustion Systems for Prediction of Particle Sticking,” Energy & Fuels, 9(2), pp. 277–283. [28] Haynes-International, 2017, “Hastelloy X Alloy.”

5 NOTIONAL FUTURE RESEARCH

The focus of the work presented in this dissertation was primarily intended to develop tools for the turbine engine industry for dust ingestion scenarios using ARD as an analog for realistic ingestion scenarios. In general, the particulates tested were larger than what is typically experienced in a typical hot-section since the compressor section will typically pulverize larger particles. The vast majority of particulates will be less than 10 µm with a median or mean centered approximately 5 µm. In addition, slower velocities in the combustor and initial turbine sections will cause less inertial impacts instead relying on turbulent eddy diffusion and thermophoresis mechanisms. As noted previously, deposits with equivalent diameters less than 10 µm were generally not observed on the coupon surface. There are a number of potential investigation topics to better elucidate a variety of dust ingestion scenarios. The author would recommend three primary and sub-topic areas of notional research. The first topic should examine deposit formation for particles less than 19 µm ARD. A portion of this analysis should include detailed composition and crystallographic (X-Ray crystallography) analyses before, during, and after injection. The crystalline analysis is necessary since ARD is a silt soil type (per ISO 14688-1:2002) composed primarily of crystalline quartz. To complement the material analysis, the second primary topic should develop an alternative testing method and hardware. The suggested smaller particulates will have lower Stk numbers and therefore would be more likely to follow non-inertial impact processes such as thermophoresis. For comparison, the current test section causes highly turbulent flow undergoing rapid expansion and impinging on a bluff body in the current test section is not necessarily directly comparable to actual turbine vanes. The third and final recommendation is to test actual turbine vanes or equivalent vane geometries, e.g. tube or elliptical structure, to mimic a leading edge. Combined with numerical or analytical modeling, empirically establishing high SP regions relative to particle flux could provide valuable information for engine designers and operational procedures.

5.1 TEST DUST The first primary change for dust ingestion testing focuses on the physical and material dust properties in combustion gas environment. As mentioned previously, tested particles less than 19 µm should be the focus of any future deposition-based studies. The smaller particles provide a direct comparison to realistic deposition conditions during a dust ingestion event. For example, recent experiments by Lundgreen et. al. [1] using 0-5 µm ARD has been performed in realistic conditions on a turbine vane analog. To complement the smaller particle testing size, a chemical and crystalline analysis of the dust should be analyzed as it passes through the various stages prior to depositing on to hot components. For experimental purposes, the dust changes can be analyzed in three main sections. 1) Analysis of dry ARD after exposure to realistic arid conditions. This includes exposing to low humidity and hot air up until the injection time. 2) Analysis of ARD after it is exposed to controlled combustion gases throughout the heating process. This portion of the analysis should mimic the process of actual ingested particulate. Examining composition and crystal structure changes to the particulate can provide potential insight to SP upon impact. 3) Finally, an analysis of the deposits after impact as well as extra particulates that miss the impacting surface. The purpose of this analysis is to examine the potential chemical adhesive forces between the test coupon and the subsequent deposit.

If a material-based analysis is a cost prohibitive scope limitation an alternative silt or dust should be considered as a potential candidate for SP experiments since it could be more applicable to engine manufacturers. For example, alternative dusts such as AFRL02 and AFRL03 are relatively new particulates that have been developed at the Air Force Research Laboratory to mimic the realistic effects of deposit accumulation on hot section components [2,3]. As mentioned previously, all notional alternative particulates should be less than 10 µm to mimic conditions of actual ingestion scenarios.

5.2 TESTING PROCESS The efforts of this research attempted to focus on just the impact mechanics of microparticle impacts. The current testing arrangement exposes test coupons to a pseudo jet environment resulting in an overall bluff body flow regime. Within the test section, the rapidly expanding flow causes various turbulent eddies that may cause smaller particles to re-entrain into the primary flow and impact the test coupon with a higher Stk number. Although the author finds this scenario unlikely, the additional smaller particle impacts may potentially bias SP results. Unfortunately, there is a strong connection between particle trajectories prior to impact in high temperature flows. In general, smaller particles will not necessarily rely entirely on inertial impacts for deposition due to their inherently lower Stk number. For example, Hamed et. al. [4] showed that smaller particles will depend on turbulent eddy diffusion to deposit onto the surface. Therefore, controlling or quantifying the high temperature flows near the test coupon should be considered a priority for future testing. There are two potential options, the first is modifying the current test section and the second is developing a new test section. An improved test section could allow a large variety of test coupon geometries and turbine vanes. For example, the TuRFR I and II test rigs at The Ohio State University [1] are capable of testing a variety of coupon geometries, turbine vane sections, and controlled turbulent flow conditions such as the work by Whitaker et. al. [5]. Another area of potential sub-topic is the effects of turbulent eddy diffusion within the equilibration tube. Current testing across most deposition rigs generally neglect deposition within the respective equilibration tubes or account for deposition as a function of mass. Using smaller particles, as suggested previously, will generally cause more turbulent eddy diffusion deposits as demonstrated by Li [6].

5.3 ALTERNATE TEST GEOMETRIES To complement the suggested changes in the test equipment, the test coupons or surfaces should reflect conditions similar to actual combustor or turbine components. Various cooled surfaces are a common design feature among many hot-section components across a variety of engine designs. The cooler component surface temperatures relative to the hot gas and particle temperatures can affect deposit accumulation. For example, Laycock and Fletcher [7] showed that there is a fairly non-linear response of deposition with a variety of temperature combinations when using coal ash. Ongoing testing using VTAR has implemented a cooled test coupon in an attempt to examine external deposition similar to Laycock and Fletcher [7] using 20-40 µm ARD instead of coal ash up to 1100°C. The cooled coupon test data can be potentially incorporated into the existing test matrix using the solid coupon studies presented in the majority of this work. Notionally, the larger temperature difference between the test coupon surface and the particle temperatures does appear to have non-linear deposition responses.

5.4 PRIORITY OF ANALYSIS/TESTING To summarize, subsequent deposition testing should focus on the following items in order of priority.

1) Crystalline particles (ARD) or alternatives (AFRL 02 or 03) less than 19 µm in size should be tested. Most deposition tests should focus on using particles less than 10 µm in size due to their direct applicability to actual hot-section deposits. 2) A stronger emphasis on the particle material responses should be considered depending on resources available. Examining the chemical and crystalline structure of ARD or other particles throughout the heating and deposition process could allow for additional variables to be incorporated into the SP models developed. 3) An updated test section and a variety coupons intended to examine the effects of turbulent flows and the subsequent deposits with low speed flows. Several of the notional geometries should be directly comparable to existing hot-section turbine components.

5.5 REFERENCES [1] Lundgreen, R., Sacco, C., Prenter, R., and Bons, J. P., 2016, “Temperature Effects on Nozzle Guide Vane Deposition in a New Turbine Cascade Rig,” GT2016-57560, ASME Turbo Expo 2016, Turbomachinery Technical Conference and Exposition, ASME, Seoul, South Korea, Vol. 5A, p. V05AT13A021. [2] Phelps, A. W., and Pfledderer, L. M., 2014, “DEVELOPMENT OF A NATURALISTIC TEST MEDIA FOR DUST INGESTION CMAS TESTING OF GAS TURBINE ENGINES,” Engineering Conferences International, pp. 1–2. [3] Murugan, M., Ghoshal, A., Walock, M., Nieto, A., Bravo, L., Barnett, B., Pepi, M., Swab, J., Pegg, R. T., Rowe, C., Zhu, D., and Kerner, K., 2017, “Microstructure Based Material-Sand Particulate Interactions and Assessment of Coatings for High Temperature Turbine Blades,” Volume 2D: Turbomachinery, ASME, p. V02DT48A009. [4] Hamed, A., Tabakoff, W. C., and Wenglarz, R. A., 2006, “Erosion and Deposition in Turbomachinery,” Journal of Propulsion and Power, 22(2), pp. 350–360. [5] Whitaker, S. M., Prenter, R., and Bons, J. P., 2015, “The Effect of Freestream Turbulence on Deposition for Nozzle Guide Vanes,” Journal of Turbomachinery, 137(12), p. 121001. [6] Li, M., 2010, “Eddy Impaction As An Ash Deposition Mechanism: A Theoretical And Experimental Investigation,” Brigham Young University. [7] Laycock, R., and Fletcher, T. H., 2015, “Independent Effects of Surface and Gas Temperature on Coal Flyash Deposition in Gas Turbines at Temperatures Up to 1400°C,” ASME Turbo Expo 2015, Volume 3: Coal, Biomass and Alternative Fuels; Cycle Innovations; Electric Power; Industrial and Cogeneration, ASME, Montreal, Canada, p. V003T03A010.

6 APPENDIX

6.1 DATA REDUCTION PROCESS From each test case, the Sticking Probability (SP) is empirically modeled as a function of impact vectors, surface temperatures, and particulate sizing. Figure 6-1 is the process for data reduction for each test condition. Impact vectors and surface temperatures are matched to deposits depending on the relative locations on the test coupon after each test. Surface temperatures are estimated from two thermocouples embedded in the test coupon near the deposited surface. Impact locations are extrapolated using a hybrid PIV-CFD technique [1]. This technique requires multiple iterations depending on nominal particulate sizing for each test case. A variety of image processing techniques are employed to maximize the number of particles potentially impacting the coupon surface. Section 5.1.2 elaborates on the hybrid PIV-CFD technique utilized for this research. Deposits are identified using a semi-autonomous object identification process. Each image autonomously selects potential particles and then is altered by an analyst as necessary to ensure accuracy. This method was compared between multiple analysts for consistency with careful training. A full autonomous system was determined to have inconsistent accuracy depending on variations in coupon surface color and deposit color depending on test conditions. Section 5.1.1 elaborates on the automatic object recognition and processing method.

Single Test Particle Tracking

Acquire PIV data near Particle Tracking the impact location of and Deposition the test coupon

Deposition 2D CFD steady state velocity flow field of Reduce PIV data test section Collect microscopic images of coupon surface

Identify the initial Lagrangian Particle Tracking: Identify particle deposits position and velocities Extrapolate the particle relative to location of the of each particle trajectory downstream test coupon

Identify particle hits, Separate deposits misses, rebounds, and based on sizing double bounces

DATA FILE: Deposit DATA FILE: Particle hits, data relative to misses, rebounds and coupon location double bounces

Determine Sticking Identify Hit Rate Convert raw data Probability relative to particle (Impact Efficiency) scatter into impact vectors and sizing based on particle size regression models

DATA FILE: Collated Sticking probability relative to impact vectors DATA FILE: Hit Rate and surface temperatures and impact regression models

Figure 6-1: Flowchart of data reduction process for a single test.

6.1.1 Deposition Deposition data is acquired using microscopic imaging of the coupon after each exposure. Since the coupon is not significantly smaller than the flow path from the equilibration tube in VTAR, each location along the coupon is exposed to a different particle impact condition. Therefore, deposits must be quantified relative to the location on the coupon to correlate to the various particle impact vectors across the surface. The process for deposit data acquisition is shown in Figure 6-2. To ensure statistical strength, three horizontal

82 sample image rows were acquired across the coupon. Each image size to about 698 µm by 522 µm at 20x magnification using a Zeiss Axio Vert.A1 with an AxioCam MRc5. Each image was acquired manually and spacing between each image was relatively equal. This process resulted in approximately 50 to 70 images per sample image row.

Post Test Coupon

Setup microscope for 20x 20x magnification: 698µm x 522µm magnification at maximum Max resolution: 2584x1936 pixels acquisition resolution

Acquire images starting at the leading Each image should be edge and move towards the trailing focused on the surface so the edge with relatively equal spacing particles appear blurry

DATA FILE: 50 to 70 images per sample row

DATA FILE: All images Repeat for 3 sample image rows for deposit identification

Figure 6-2: Posttest coupon data acquisition for deposits.

After all the data was acquired per test, each image was processed using a series of filters and blurring functions. Figure 6-3 shows the image processing methodology. The process begins by calibrating several randomly selected images from the image sample row data set. The calibration process was used to select the constant values for each appropriate image filter. For example, the number of blurred pixels per region would have to be adjusted for a set of images. Since there is a significant texture and color differential across the entire coupon and between data sets, the filter constants are saved for each image. The process is repeated until the user is satisfied with the settings. The user then processes the data by validating each image for accuracy and either adding or removing selected particles accordingly. A data file (typically “.mat”) is output from each sample image row.

DATA FILES: 50-70 images per sample row

Calibration Acceptable Process settings?

Yes Convert color image to grayscale Process all images

Gradient of grayscale image Each image is validated for accuracy

Blur or smooth image Add or remove misidentified particles on each image

Compliment image then convert to black and white DATA FILE: Particle shape, size, location, calibration filters per image for each row

Remove small artifacts and smooth edges (blurring)

Fill in holes for “white” regions and identify regions

Figure 6-3: Process for identifying deposits per each sample image row

All the sample image row data files were combined into a single data file where the data is processed based on particle size. Figure 6-4 shows the process to combine each sample image row into a single data file. The raw deposit data is compared to a user defined input of desired particle bin sizes. Each bin size correlates to a relative diameter of particles based on the sample’s respective size distribution. Testing for

this study assumes that the particle deposit sizes are not significantly different than the original sample bin size. The final output from the deposit analysis is a single data file which identified particle deposits based on individual size relative to the location on the test coupon.

DATA FILE: Particle DATA FILE: Particle DATA FILE: Particle shape, size, location, shape, size, location, shape, size, location, calibration filters per calibration filters per calibration filters per image row image row image row

Combine sample image row data files (.mat files)

Separate particle deposits DATA FILE: Sand bin based on sizing separation sizing

DATA FILE: Particle deposits per size and location on coupon

Figure 6-4: Combining data files from each sample image row and separating particle deposits based on size then output a collated data file to correlate to impact vectors.

6.1.2 Particle Tracking Correlating deposits to impact vectors is ultimately needed to calculate a SP relative to multiple independent variables. Particle trajectories are semi-empirically derived using a hybrid PIV-CFD technique utilized in several prior studies at Virginia Tech [2–6]. Figure 6-5 is the process to acquire PIV data and to reduce the available data to find the initial positions and velocities of the particulate in the flow field. Unlike traditional PIV techniques that use small particulates (typically titanium dioxide) to map a flow field, this method uses the initial positions and velocities than extrapolates the various positions forward and backward in time. The particle trajectories are extrapolated using a Lagrangian particle tracking algorithm similar to numerical simulations. From each test case, PIV data is acquired using a twin head Litron Nd:YAG laser emitting a 532 nm wavelength at 135 mJ into the test section through a quartz window on the side of the test section that illuminates the horizontal midline of the test coupon. The PIV system is limited to 7.4 Hz image pair acquisition rate. The pulse times used for Boulanger et. al. [7] are typically around 40 µs and subsequent testing for validation reduced that time to 20 µs and 6 µs to reduce the trivial solutions during the Lagrangian particle tracking algorithm. The collected data is processed using Dantec® software to find the minimum pixel values for both the first and second frame image pair. The resulting image pair contains the minimum pixel value across all images in the sample data set. Each image in the data set is then subtracted by the minimum image for both the first and second frame. This process eliminates extra ambient light across the data set and helps identify particles that were not immediately visible in the raw images. This process can be repeated using alternative software such as MATLAB®. The next portion of the PIV image process utilizes several MATLAB® scripts to find the initial positions and velocities of potential particles. The original images after subtracting the minimum pixel images are effectively grayscale with a range of 0 to 255 per pixel. Each image is converted to black and white using a cutoff filter from the Image Processing Toolbox within MATLAB®. A sub-script identifies each potential particle and trajectory within a user-defined region of interest that is part of the preprocessing script. Varying the black and white filter cutoff values for all image pairs increases the probability of finding potential particles. Duplicate particles are removed per image for all the black and white filter cutoffs. Each potential particle and initial trajectory is collated and placed into a data file for particle trajectory extrapolation using the Lagrangian particle tracking algorithm.

Single Test Case

DATA FILES: PIV images during sand injection

Posttest image Calculate image Calculate the RAW DATA FILES: Processed processing using minimum across images minus the image files Dantec® program suite the image set minimum image

Setup image orientation Select the coupon Rotate images to Automatically select the region and datum points for leading and trailing sync with the actual in front of the test coupon processing edge points coupon test angle normal to the bulk flow

Process all image pairs in the data set No

Apply black/white Identify each particle and Altered image image processing subsequent trajectory Processed all black/white limits? pair limits (particle_track.m) Yes

No Last image pair in data set?

Yes

DATA FILE: Initial position and velocity vectors of notional particles

Figure 6-5: Process to identify initial position and velocity vectors of particles for each test case using PIV image data.

The tracking algorithm requires three data sets, the initial position and velocity trajectories (discussed previously), particle bin sizing, and a CFD velocity flow field. Figure 6-6 is the Lagrangian particle tracking data reduction process to extrapolate particle trajectories relative to particle size. The CFD velocity flow field is used to calculate the relative drag on a particle and extrapolate the fluid forces for each time step and subsequent or prior physical location in the flow field. This portion of the analysis is iterated for each particle sample bin size. For example, particle with an equivalent diameter between 19.0 µm and 21.8 µm in the ARD sample will correspond to particle impact trajectories that assume a size of 20.4 µm. The nominal size is used to calculate the local drag forces on a particle in the flow. Particle impacts, rebounds, double impact locations, and overall impact efficiency (hitting rate) are output from the Lagrangian particle tracking algorithm can be collated for statistical regression model development. For each metric, specifically particle impacts and rebounds, an optimal regression model is selected as a function of relative horizontal location on the test coupon. The regressions were typically first

87 or second order equations. Those equations are then applied in the next section where the impact vectors are correlated to the deposit locations on the coupon.

DATA FILE: Initial position DATA FILE: CFD and velocity vectors of velocity flow field notional particles

DATA FILE: Sand bin separation sizing

Select a nominal particle size

Extrapolate each identified particle No trajectory downstream

DATA FILE: Particle All particle trajectories, impacts, misses, sizes used? Yes rebounds, double impacts, and impact efficiency per size

DATA FILE: Particle impacts, Select the optimal regression misses, rebounds, double model for each impact bounce regression models per trajectory per size size Figure 6-6: Determine particle trajectories, impact efficiency, relative to particle size using initial trajectories and the CFD velocity flow field. Also establish regression models for particle impacts and rebounds relative to paritcle size.

6.1.3 Sticking Probability Using Particle Tracking and Deposition Data All data from particle tracking and deposits thus far have been a function of horizontal location on the test coupon. Establishing the SP as a function of particle impact conditions (velocity, temperature, etc.), eliminates the dependence on relative coupon location. The process for calculating are separated into two major paths shown in Figure 6-7, (1) develop a model of deposits as a function of local conditions of temperatures and impact vectors across the coupon surface and (2) calculating particle hitting rate relative to the total injected. Deposits are quantified as a function of their size, local impact vectors, surface temperature, particle (or flow) temperature prior to impact relative to their location on the coupon. The hitting rate is calculated relative to nominal particle size from the injected sample bin sizing. The hitting

88 rate calculation is described in detail in Chapter 4. The final is then developed using the hitting rates combined with the deposit model. The process reduces the data to a simple multi-linear curve-fit. The methodology can be directly linked to a physical process since all known common variables are included in the model for this study.

DATA FILE: Flow and Local coupon surface temperature

DATA FILE: Particle impacts regression models per size

DATA FILE: Particle DATA FILE: Particle impact hitting rate deposits per size and in region of interest location on coupon per size

Deposits per size bin, coupon location, and surface temperature

Calculate the Sticking Probability per (1) particle size bin (2) relative impact vectors (3) surface temperature (4) particle/flow temperature

Figure 6-7: Combining 3 to 4 data files allows for SP to be calculated independent of location on the test coupons.

6.2 MAXIMUM NOTIONAL STICKING PROBABILITY Prior numerical and semi-numerical studies have estimated that SP to be significantly higher than observed in this study. For example, Singh and Tafti [8]as well as Barker et. al. [9] have SP that are unusually high compared to experimental observations from this study in Chapter 4. Singh and Tafti [8] estimating a SP of 0.05 to 0.28 from 950°C to 1050°C respectively. Barker et. al. [9] observed a SP of about 0.10 at high normal impact velocities at 1100°C. Both studies produced SP that are abnormally high based on experimental observations in Chapter 4. From this study, Table 6-1 is number of particles for 10 g or 5 g injections and subsequently, Table 6-2 contains the number of particles that may possibly impact the coupon based on coupon angle relative to the coupon area and equilibration tube flow exit size (76.2 mm in diameter). Approximately 99% of the injected mass is between 10 µm to 40 µm based on the sample analysis. Depending on the coupon angle, and the sample injected, Table 6-2 provides the maximum mass that may impact the test coupon.

Table 6-1: Particle metrics and number of particles calculated from testing

Estimated Particle Density (/) 2,842.86 Coupon Area () 2,419.35 Number of Particles in 10 g between 10-40 µm 473,844,702 Number of Particles in 5 g between 10-40 µm 236,922,351

Table 6-2: Maximum possible number of impacts based on geometry and ARD sample distribution.

50° 80° Angle proportion [() ] 0.76604 0.98481 Maximum Number of Impacts (10 g) 192,570,107 247,563,358 Maximum Number of Impacts (5 g) 96,285,054 123,781,679 Maximum Impacting Mass (10 g) 4.064 5.224 Maximum Impacting Mass (5 g) 2.032 2.612

The calculation in Table 6-3 assumes that the number of particles and mass for a monolayer vary with the particle size. The relative impact efficiency based on mass from Table 6-2 and Table 6-3 implies that a for a 5 g injection at 50°, a maximum SP of 0.10 would produce a monolayer. Likewise, a nominal value of approximately 25 µm would result in a maximum of approximately 0.05. The experimental data in Chapter 3 never reached full surface saturation to develop a monolayer from any portion of the test coupons. Therefore, based on the coverage ratios in Chapter 3 and the values up to 0.05 in Chapter 4 are realistic.

Table 6-3: Maximum potential number of particles and mass based on particle size for a monolayer of deposits.

Particle Size Number of Mass of (µm) Particles for a Monolayer Monolayer (g) 10.0 3.080(10) 0.04585 14.5 1.456(10) 0.06649 20.4 7.402(10) 0.09354 23.3 5.674(10) 0.1068 26.7 4.321(10) 0.1224 34.3 2.618(10) 0.1573 40.0 1.925(10) 0.1834

6.3 ARD SOFTENING TEMPERATURE AND VISCOSITY CALCULATIONS Viscosity of ARD at elevated temperatures is calculated using the relationships presented by Senior et. al. [10] and Yin et. al. [11]. Singh and Tafti [8] uses the same methodology to calculate the viscosity as a function of temperature to numerically model SP. The viscosity relationship presented in Senior et. al. [10] depends on a critical temperature where the viscosity relationship differs between a “low” and a “high” temperature. Eqn. (6-1) and Eqn. (6-2), from Yin et. al. [11], uses mole fractions to calculate the critical temperature. Mole fractions were determined by the nominal ISO distribution of ARD. Table 6-4 shows that ARD is primarily composed of silicon and aluminum oxides. The resulting critical temperature is 1089.6°C, Eqn. (6-3).

= 100 (10.75 +13.03 −5.28 −5.88 −10.28 ) +3.75 (6-1) + 453

= 100 1 − ( + + ++) (6-2)

Table 6-4: Mole fraction of ARD sample for testing.

Component Mole Fraction SiO2 0.7810 Al2O3 0.0799 Fe2O3 0.0143 Na2O 0.0315 CaO 0.0407 MgO 0.0243 TiO2 0.0041 K2O 0.0242

= 1362.8 = 1089.6 ° (6-3)

The critical temperature is used to calculate a critical viscosity between low and high temperature regimes. Eqn. (6-4) is the relationship between particle viscosity and temperature for viscosities from 104 to 108 Pa-s [10]. The critical temperature calculated previously provides the inflection point between the low and high temperature regimes discussed as follows.

(6-4) log = + 10 10

Where is temperature in Kelvin and is viscosity in Pa-s. Terms and are calculated using Eqn. (6-5) through Eqn. (6-12). The subscripts and correspond to the high and low temperature regimes around the critical temperature. Terms and require a nonbridging oxygen to tetrahedral oxygen network parameter calculated in Eqn. (6-5) and Eqn. (6-6). Similar to the critical temperature relationship before, is calculated from the mole fractions of each component of the ARD from Table 6-4. +++ + − − (6-5) = + + + 2 (6-6) =0.054468

Eqn. (6-7) through Eqn. (6-11) are used to calculate for the low and high temperature regimes. For the low temperature regime, , is a piecewise function depending on . For ARD, Eqn. (6-7) and Eqn. (6-10) are used for the high and low temperature ranges, respectively.

(6-7) = −2.81629 −0.46341 −0.35342 = −0.982 −0.902473 (6-8) ≥1.3 (6-9) =2.478718 −0.902473 −2.662091 0.2 ≥ >1.3 (6-10) =9.223 −0.902473 −36.3835 0 ≥ >0.2 =9.223 −0.902473 (6-11) <0

Equations (6-12) to (6-14) and Table 6-5 are used to calculate the B term which are used by the previous equations to calculate viscosity.

= + + + + + + + + + + (6-12) + (6-13) = + (6-14) = + +

Where terms are coefficients from Table 6-5, is a relative mole fraction of calcium oxide, and is a relative mole fraction of silicon oxide. Terms and use the mole fractions from Table 6-4.

Table 6-5: Coefficients for compositional dependence of .

High temperature Low temperature −224.98 −7563.46 636.67 24431.69 −418.70 −17685.4 823.89 32644.26 −2398.32 −103681.0 1650.56 74541.33 −957.94 46484.8 3366.61 −146008.4 −2551.71 104306.0 387.32 21904.63 −1722.24 −68194.8 1432.08 48429.31

The critical viscosity for ARD, Eqn. (6-15), is calculated as follows. Singh and Tafti [8] used a similar critical viscosity calculation as a reference point for numerically modeling SP. For the purposes of this study, the critical viscosity point provides a similar role for empirically estimating SP.

= 6.9582 (10 ) ∙ (6-15)

The previous calculations can also be used to calculate ARD viscosity across a range of temperatures. Figure 6-8 depicts the viscosity changing logarithmically relative to temperature. The range of viscosities generally falls in the low-temperature regime associated with the previous equations.

ARD Viscosity 108

107

106

105 1280 1290 1300 1310 1320 1330 1340 1350 1360 1370 Temperature (K)

Figure 6-8: Viscosity of ARD across the range of test temperatures in this study.

For non-dimensionalization derivations, a linear regression of ARD viscosity simplifies the comparison for the temperature range between 1000°C and 1100°C. Eqn. (6-16) is the first order linear regression. There is minimal difference between the first and second order regressions for the temperature range of interest as shown in Figure 6-9.

log =30.8628 −0.01836437 (6-16)

Noonal ARD Viscosity 7.6 Original Logarithmic Regression 7.4 Linear Regression

7.2

6.8

6.6

6.4

6.2

5.8

5.6 1280 1290 1300 1310 1320 1330 1340 1350 1360 1370 Temperature (K)

Figure 6-9: Viscosity of ARD from the original logarithmic regression compared to a linearized regression.

6.4 DYNAMIC MODULUS OF ELASTICITY OF HASTELLOY® X The elastic modulus of deposits and surface interactions depends on elastic and plastic deformation during the impact process. Numerical modelling from relevant literature focuses on the elastic-viscous-plastic properties associated with impact mechanics. Section 6.3 discusses the viscous properties of ARD at the associated flow and particle temperatures in this study. This section summarizes the dynamic modulus of elasticity for HASTELLOY® X for the surface temperatures experienced during testing. Table 6-6 are the values available from Haynes International in 2017 [12]. The 950°C properties were obtained from Varela et. al. [13].

Table 6-6: Dynamic modulus of elasticity of Hastelloy® X across a broad temperature range from Haynes International [12] and Varela et. al. [13].

Temperature Young’s Modulus (°C) (K) (GPa) (Pa) 25 298.15 205 2.05(10) 100 373.15 202 2.02(10) 200 473.15 198 1.98(10) 300 573.15 192 1.92(10) 400 673.15 187 1.87(10) 500 773.15 180 1.8(10) 600 873.15 173 1.73(10) 700 973.15 165 1.65(10) 800 1073.15 157 1.57(10) 900 1173.15 148 1.48(10) 950 1223.15 140 1.4(10)

The material property data in Table 6-6 was fit to a regression function derived in Equation (6-17). The regression has an of 1.00.

= −37,437.6 −11,330,632.63 + 211,480,830,511.59 (6-17)

Where istemperature in Kelvin and is dynamic modulus of elasticity in Pascals. For all testing in this study, the surface temperature of the HASTELLOY® X is between 800°C to 1000°C. Between those temperatures, the modulus decreases linearly from about 156 GPa to about 136 GPa as shown in Figure 6-10.

HASTELLOY® 154

152

150

148

146

144

142

140

138 1100 1150 1200 1250 Temperature (K)

Figure 6-10: Dynamic modulus of elasticity for HASTELLOY® X for the range of surface temperatures within this study.

For non-dimensionalization derivations, a linear regression simplifies the comparison for the temperature range between 800°C and 1000°C. Eqn. (6-18) is the first order linear regression. There is minimal difference between the first and second order regressions for the temperature range of interest as shown in Figure 6-11.

= 262,879,445,255.776 −99,170,473.51 (6-18)

HASTELLOY® 154 1st Order Regression 2nd Order Regression 152

150

148

146

144

142

140

138 1100 1150 1200 1250 Temperature (K)

Figure 6-11:Dynamic modulus of elasticity for HASTELLOY® X comparison for first and second order regressions.

6.5 DIMENSIONAL ANALYSIS Deposit formation from particle impacts depend primarily with the particle kinetic energy, particle physical properties, and substrate physical properties. The impact process can be separated into two main categories, elastic or inelastic deformation caused by momentum and the adhesion forces between the surface and particle. Table 6-7 contains all the relevant primary variables used for non-dimensionalization using the Buckingham Pi theorem. The repeated variables are highlighted in yellow. The repeated variables were selected based on multiple iterations between optimizing model development discussed in Chapter 4.

Table 6-7: Common independent and controlled variables associated with deposition. Variables highlighted in yellow are repeated variables for the Buckingham Pi theorem.

Variable Base Units Description Normal impact velocity Tangential impact velocity Impact velocity magnitude Flow/Particle temperature

Surface temperature Critical temperature Particle equivalent diameter Deposit area (function of particle diameter) Particle mass, function of particle density and diameter μ Particle viscosity at specific temp μ Particle critical viscosity [8] - Sticking probability Dynamic modulus of elasticity of the coupon

The primary independent variables for testing were flow-particle temperature, surface temperature, impact velocity vectors, and particle size. The primary role of temperature for particle impact mechanics is to determine the physical state of both the particle and the surface. The dynamic modulus of elasticity and viscosity are common terms used to describe either the substrate and the particle, respectively. Typically, when the modulus is used for modeling purposes, there is an accompanying modulus for the particle. Fortunately, the viscosity of the particle is more readily available based on empirical data from Yin et. al. [11]. Combining both viscosity of the particle and the dynamic modulus of elasticity with the substrate provides a unique non-dimensional term for estimating through statistical regression development.

Eqn. (6-19) to Eqn. (6-22) are the final non-dimensional terms for predicting . Chapter 4 has the final statistical regression equation for reference. Additional non-dimensional terms were derived but were discovered to be statistically insignificant for the appropriate multi-linear regressions. Variations of Π, Π, and Π were evaluated and checked for statistical significance. Table 6-8 is a sample list of the variations for the non-dimensional terms in Eqn. (6-20) to Eqn. (6-22). The Π and Π terms were strongly influenced by probabilistic modeling. For example, if is equal to for a perfectly normal impact, it would maximize the likelihood a particle will deposit. Likewise, Π was based on the probabilistic modeling from Singh and Tafti [8]. The Π term that contains viscosity and modulus of elasticity has the most interdependency with Π and Π due to the repeating variables of velocity and viscosity. Selection of

Π terms was the most challenging since the appropriate viscosity term and velocity term would provide the most statistical significance.

Π == (Π, Π, Π) (6-19)

(6-20) Π =

(6-21) Π =

(6-22) Π =

Table 6-8: Variations of non-dimensional terms used for model optimization.

-term Versions Π Π Π

The Π term is effectively an interaction term between the particle and the surface. Particle mass and cross section area terms are still important for model development. Unfortunately, due to statistical cross- correlation, both mass and area can be replaced with particle diameter. Eqn. (6-23) and Eqn. (6-24) are examples of the Π terms using particle cross section area and mass. Variations replacing viscosity and velocity terms are also acceptable formats.

(6-23) , =

(6-24) , =

6.6 BOUNDARY LAYER ON THE SOLID TEST COUPON The boundary layer developed on the surface of the solid coupon can affect the likelihood of particle deposits since any drastic changes in flow vectors may affect the impact trajectories and subsequent deposit probability. Direct measurement of the boundary layer was not feasible for the testing performed in this study. Therefore, an analytical solution was implemented. Eqn. (6-25) is the turbulent boundary layer thickness assuming an isothermal plate [14]. Although the isothermal condition does not directly apply for this study, the boundary layer estimation provides relative context for particle impact results. Figure 6-12 shows the range of boundary layers for a variety of parallel bulk flow velocities of dry air at atmospheric pressure and 1050°C. Although the estimated boundary layer thickness is not directly applicable, it indicates that the maximum boundary layer is relatively small compared to the overall length of the coupon. Therefore, one can assume that the boundary layer effects of particle impacts are minimized.

/ (6-25) =0.37

Figure 6-12: Boundary layer thickness of flow across the solid test coupon for a variety of potential flow velocities.

100

6.7 REFERENCES [1] Reagle, C. J., Delimont, J. M., Ng, W. F., Ekkad, S. V., and Rajendran, V. P., 2012, “A Novel Optical Technique for Measuring the Coefficient of Restitution of Microparticle Impacts in a Forced Flowfield,” ASME Turbo Expo 2012, Volume 7: Structures and Dynamics, Parts A and B, ASME, p. 1. [2] Reagle, C. J., Delimont, J. M., Ng, W. F., and Ekkad, S. V., 2013, “Study of Microparticle Rebound Characteristics Under High Temperature Conditions,” Journal of Engineering for Gas Turbines and Power, 136(1), pp. 011501–011501. [3] Reagle, C. J., Delimont, J. M., Ng, W. F., and Ekkad, S. V., 2014, “Study of Microparticle Rebound Characteristics Under High Temperature Conditions,” Journal of Engineering for Gas Turbines and Power, 136(1), p. 11501. [4] Reagle, C. J., Delimont, J. M., Ng, W. F., Ekkad, S. V., and Rajendran, V. P., 2013, “Measuring the Coefficient of Restitution of High Speed Microparticle Impacts Using a PTV and CFD Hybrid Technique,” Measurement Science and Technology, 24(10), p. 105303. [5] Delimont, J. M., Murdock, M. K., Ng, W. F., and Ekkad, S. V., 2015, “Effect of Temperature on Microparticle Rebound Characteristics at Constant Impact Velocity—Part I,” Journal of Engineering for Gas Turbines and Power, 137(11), p. 112603. [6] Delimont, J. M., Murdock, M. K., Ng, W. F., and Ekkad, S. V., 2015, “Effect of Temperature on Microparticle Rebound Characteristics at Constant Impact Velocity—Part II,” Journal of Engineering for Gas Turbines and Power, 137(11), p. 112604. [7] Boulanger, A. J., Hutchinson, J., Ng, W. F., Ekkad, S. V., Keefe, M. J., Xu, W., Barker, B. J., and Hsu, K., 2017, “Experimental Based Empirical Model Of The Initial Onset Of Sand Deposits On Hastelloy-X From 1000°C To 1100°C Using Particle Tracking,” GT2017-64480, ASME Turbo Expo 2017: Turbomachinery Technical Conference and Exposition, ASME, Charlotte, North Carolina, United States, p. V02DT48A015-V02DT48A015. [8] Singh, S., and Tafti, D. K., 2015, “Particle Deposition Model for Particulate Flows at High Temperatures in Gas Turbine Components,” International Journal of Heat and Fluid Flow, 52, pp. 72–83. [9] Barker, B. J., Hsu, K., Varney, B., Boulanger, A., Hutchinson, J., and Ng, W. F., 2017, “An Experiment-Based Sticking Model for Heated Sand,” ASME Turbo Expo 2017, pp. 1–11. [10] Senior, C. L., and Srinivasachar, S., 1995, “Viscosity of Ash Particles in Combustion Systems for Prediction of Particle Sticking,” Energy & Fuels, 9(2), pp. 277–283. [11] Yin, C., Luo, Z., Ni, M., and Cen, K., 1998, “Predicting Coal Ash Fusion Temperature with a Back- Propagation Neural Network Model,” Fuel, 77(15), pp. 1777–1782. [12] Haynes-International, 2017, “Hastelloy X Alloy.” [13] Varela, L. A., and Stewart, C. M., 2016, “Modeling the Creep of Hastelloy X and the Fatigue of 304 Stainless Steel Using the Miller and Walker Unified Viscoplastic Constitutive Models,” Journal of Engineering Materials and Technology, 138(2), p. 21006.

101

[14] Bergman, T. L., Lavine, A. S., Incropera, F. P., and DeWitt, D. P., 2011, “Fundamentals of Heat and Mass Transfer, 7th Edition,” John Wiley & Sons, Inc.

102