Experiments and Modeling of Multilayered Coatings and Membranes. Application to Thermal Barrier Coatings and Reverse Osmosis Membranes. by MASSACHUSETTS INTflUTE OF TECHNOLOGY Jacques Luk-Cyr OCT 16 204 B.Ing., Ecole Polytechnique de Montreal (2012) LIBRARIES Submitted to the Department of Mechanical Engineering in partial fulfillment of the requirements for the degree of Master of Science at the MASSACHUSETTS INSTITUTE OF TECHNOLOGY September 2014 @ Massachusetts Institute of Technology 2014. All rights reserved. 7Q -- . - - - - - - - - redacted Signature - - - -. ... A uthor ........ Department of Mechanical Engineering August 8, 2014 Signature redacted .............-. Lallit Anand Warren and Towneley Rohsenow Professor of Mechanical Engineering Thesis Supervisor Certified by ............ Signature redacted' ....... .... David E. Hardt Chairman, Department Committee on Graduate Students Accepted by......... 2 Experiments and Modeling of Multilayered Coatings and Membranes. Application to Thermal Barrier Coatings and Reverse Osmosis Membranes. by Jacques Luk-Cyr Submitted to the Department of Mechanical Engineering on August 8, 2014, in partial fulfillment of the requirements for the degree of Master of Science Abstract In this thesis, I developed a novel methodology for characterizing interfacial delamination of thermal barrier coatings. The proposed methodology involves novel experiments-plusnumerical simulations in order to determine the material parameters describing such failure when the interface is modeled using traction-separation constitutive laws. Furthermore, a coupled fluid-permeation and large deformation theory is proposed for crosslinked polymers with a view towards application to reverse-osmosis. A systematic simulation plus experiment-based methodology is proposed in order to calibrate the material parameters of the theory. Finally, the process of reverse osmosis is studied in the context of water desalination. An experimental set-up is proposed in order to characterize the thin-film composite membranes widely used in the industry, and a preliminary set of experiments are performed. Thesis Supervisor: Lallit Anand Title: Warren and Towneley Rohsenow Professor of Mechanical Engineering 3 4 Acknowledgments First and foremost, I would like to thank my advisor Professor Lallit Anand for his guidance throughout the past two years. His dedication and desire for excellence, which recently earned him the coveted Daniel C. Drucker Medal, will be an inspiration to me from now and onward. This work would not have been possible without my current and former labmates, who have shared with me their deep knowledge and provided me with invaluable advices; special thanks to Claudio V. Di Leo, Dr. Elisha Rejovitsky, Prof. Shawn Chester, Prof. David Henann, Dr. Kaspar Loeffel, Dr. Haowen Liu and Prof. Jafar Albinmoussa. I would also like to thank Ray Hardin, Leslie Regan, Joan Kravit and Pierce Hayward whose assistance is greatly appreciated. Many thanks also go to my roommates Daniel Preston and Boris Valkov, whose presence made life at home easy and cheerful. I thank my family, parents and sister, for always being there when in need and supporting my decisions. Finally, I thank my fiance to be Thiha, who gave me everything although living hundreds of miles away. Financial support from the King Fahd University of Petroleum and Minerals in Dhahran, Saudi Arabia, through the Center for Clean Water and Clean Energy at MIT and KFUPM under project number R9-CE-08. Partial support from NSF (CMMI Award No. 1063626) is also gratefully acknowledged. 5 6 Contents 1 Thesis structure 17 I 18 Thermal Barrier Coatings 2 Introduction 19 3 Experimental characterization of interfacial properties of thermal coatings 3.1 M aterial ...................................... 3.2 Tension Delamination Experiment .............................. 3.2.1 Specimen preparation ................................. 3.2.2 Experimental procedure ......................... 3.2.3 Results of tension experiment ...... ...................... 3.3 Shear Delamination Experiment ...... ......................... 3.3.1 Specimen preparation ................................. 3.3.2 Experimental procedure ......................... 3.3.3 Results of shear delamination experiment . . . . . . . . . . . . 3.4 Asymmetric Four-Point Bending . . . . . . . . . . . . . . . . . . . . . 3.4.1 Specimen preparation ................................. 3.4.2 Experimental procedure ........................ 3.4.3 Results of asymmetric four-point bending experiment . . . . . 4 Interface traction-separation constitutive model 4.1 Cohesive Zone Modeling . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Mixed-Mode Bilinear Traction-Separation Model ................ 4.3 Implementation of an Elastic-Plastic-Damaging Traction-Separation tutive Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.1 Variational formulation of the macroscopic force balance . . 4.3.2 Summary of the interface constitutive model . . . . . . . . 4.3.3 Finite element development . . . . .... . . . . . . . . . . . 4.3.4 Two-dimensional linear cohesive element . . . . . . . . . . . 4.4 Notes on the Implementation of a Thermo-Mechanically Coupled Plastic-Damaging Traction-Separation Law . . . . . . . . . . . . . . 4.4.1 Variational formulation . . . . . . . . . . . . . . . . . . . . . 7 barrier 23 23 24 24 25 25 26 26 26 .. . . . 27 . . . . 27 27 . 28 . . . . 28 . . . . . Consti. . . . . . . . . . . . . . . . . . . . . . . . . Elastic. . . . . . . . . . 45 45 46 47 47 49 51 52 54 54 5 Estimation of the material properties in the mixed-mode bilinear traction61 separation constitutive model . 61 5.1 Material Parameters .............................. 61 .............................. 5.2 Numerical Simulations ....... 62 5.3 Calibration Results ................................ 71 6 Concluding remarks II 72 Reverse Osmosis Membranes 73 7 Introduction 8 Summary of the coupled fluid permeation and large deformation theory 77 for crosslinked polymers of [11] 78 8.1 Constitutive theory for isotropic materials ................... 78 .......................... equations 8.1.1 Constitutive 80 8.2 Specialization of the Constitutive Equations ....................... 80 8.2.1 Free-energy ................................ 82 ........................ 8.2.2 Stress. Chemical potential ...... 82 8.2.3 M obility .................................. 84 8.3 Governing Partial Differential Equations .................... 9 Calibration of the theory: comparison with experimental results 9.1 Numerical implementation ............................ 9.2 Material parameters .............................. 9.2.1 Simple compression ............................ 9.2.2 Isotropic free-swelling ................................. 9.2.3 Steady-state pressure-driven diffusion - reverse osmosis ......... 9.2.4 Limitation of the theory ........................ 87 87 . 87 87 88 89 . 91 101 10 Concluding remarks III Thin-Film-Composite Membranes: Application to Reverse 102 Osmosis in Water Desalination 103 11 Introduction 12 Characterization of TFC membranes ...................................... 12.1 Material ......... 12.2 Experimental Apparatus ............................. 12.2.1 Reverse osmosis set-up ........................... 12.2.2 Uniaxial tension .............................. 12.3 Preliminary Results ................................. 12.3.1 Reverse osmosis experiments ........................... 8 107 107 107 107 108 108 .108 12.3.2 Uniaxial tension experiment . . . . . . . . . . . . . . . . . . . . . . . 109 117 13 Concluding remarks IV 118 Appendices related to Part I A Existing approaches to investigating thermal barrier coatings A.1 On Existing Experimental Techniques ........................... A.1.1 Mode-I dominant methods ........................ ........................... A.1.2 Indentation methods ...... ....................... methods dominant A.1.3 Mode-II A.1.A Mixed-mode methods ............................ A.2 On Modeling TBC Failure ............................ 119 119 119 120 120 121 121 B Numerical implementation of traction-separation law B.1 Time Integration Procedure ........................... ........... 0. .. B.1.1 Solving for AL6. Case 1: 4&" > 0 and 4 129 . 0 0. ........... B.1.2 Solving for A8". Case 2: IV4' > 0 and < ........... B.1.3 Solving for A8P. Case 3: IkT" > 0 and <}*4N > 0 B.2 Computing the Element-Level Stiffness .......................... .......................... B.3 On the Failure of an Interface ...... B.4 Numerical Implementation in a UEL ...................... B.5 Guide to Creating an Input File for Use with a User-Element Subroutine B.5.1 Input file for UEL ................................. 125 125 127 V Appendices related to Part II C Numerical implementation of the coupled fluid permeation formation theory . . . . . . . . . . . C.1 Numerical Methodology ............ . . . . . . . . . . . C.2 Description of the Element .......... . . . . . . . . . . . C.2.1 Plane strain ................ . . . . . . . . . . . C.2.2 Axisymmetry .............. 9 130 132 135 136 137 142 145 and large de147 . . . . . . . . 147 . . . . . . . . 150 . . . . . . . . 150 . . . . . . . . 150 10 List of Figures 2-1 3-1 3-2 3-3 3-4 3-5 3-6 3-7 3-8 3-9 3-10 3-11 (a) Schematic of a TBC on a superalloy in a thermal gradient;(b) Cross-section of an APS top-coat-sprayed TBC and its associated "splats"-like microstructure;(c) Cross-section of an EBPVD top-coat-sprayed TBC and its associated columnar microstructure (adapted from [13J) . . . . . . . . . . . . . . . . . . SEM micrograph of the cross-section of a TBC: (a) an as-sprayed specimen, and (b) an isothermally exposed specimen (144h at 1100*C) . . . . . . . . . (a) distinct isothermally exposed specimen for various period of time;(b) an isothermally exposed Sulzer specimen (144h at 11000 C) . . . . . . . . . . . . (a) square 5 mm specimens of the steel/TBC assembly;(b) aluminum tension bars with 100 pm copper wires;(c) aluminum tension bar with TBC/steel specimen; (d) fully assembled tension specimen in bonding clasp with a gripped aluminum plate;(e) curing of the bond . . . . . . . . . . . . . . . . . . . . . (a) low-load single column tabletop Instron 5944 ;(b) with installed specimen;(c) schematic of the tension specimen with dimensions;(d) close-up of the tension specimen in the Instron 5944 . . . . . . . . . . . . . . . . . . . . set-up of digital image correlation system . . . . . . . . . . . . . . . . . . . . stress versus displacement curves from four tension experiments: (a) with error bars for the displacement measurements; and (b) with error bars for the stress measurements. Error bars correspond to one standard deviation on an average taken from t 40 data points. . . . . . . . . . . . . . . . . . . . . . . SEM micrographs of the tension fracture surface facing the bond-coat side of the specimen: (a) low magnification, and (b) high magnification. Fig.(b) highlights regions of exposed bond coat surrounded by rings of thermally grown oxide (TGO). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . SEM micrograph of the tension fracture surface facing the top-coat side of the specimen. The micrograph highlights a region of exposed TGO, referred to as the "TGO cap", which matches regions of exposed bond-coat on the opposing fracture surface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . schematic of the experimentally observed fracture path. . . . . . . . . . . . . (a) MTI-STX-201 diamond-wire-saw cutter from the CMSE Crystal-SEM facilities;(b) machining of the TBC specimen;(c) TBC specimen with machined top-coat "islands" . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (a) micromechanical biaxial apparatus used in the shear delamination experiments ;(b) machining of the TBC specimen;(c) schematic of the shear specimen with dimensions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 22 30 31 32 33 34 35 36 36 36 37 38 3-12 shear stress versus shear displacement curves from three shear delamination experiments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3-13 SEM micrographs of the shear delamination fracture surfaces. (a) and (b) show fracture surfaces facing the bond-coat side, while (c) shows a fracture surface facing the top-coat side of the specimen. . . . . . . . . . . . . . . . . 3-14 (a) TBC plates glued 8 mm apart on an aluminum substrate;(b) water-jet cutting operation;(c) final asymmetric beam specimen with aluminum stiffener;(d) schematic of the bending specimen with dimensions. . . . . . . . . . 3-15 micro-mechanical apparatus used in four-point bending experiments . . . . . 3-16 close-up of the bending specimen in the testing machine. . . . . . . . . . . . 3-17 asymmetric four-point bending results for three experiments. (a) force vs. displacement response, and (b) displacement vs. time response. . . . . . . . 3-18 SEM micrographs of the asymmetric four-point bending fracture surfaces. (a) Shows a fracture surface facing the bond-coat side, while (b) shows a fracture surface facing the top-coat side of the specimen . . . . . . . . . . ... . . . . . 4-1 4-2 4-3 4-4 4-5 5-1 5-2 5-3 . 68 5-4 5-5 Schematic of the bilinear traction-separation interface constitutive relation, showing (a) the pure normal response (no tangential deformation), and (b) the pure shear response (no normal deformation). . . . . . . . . . . . . . . . Schematic of interface between two bodies Bt and B-.... . . . . . .. 57 Schematic of yield surfaces for the normal and shear mechanisms. . . . . . . Four noded cohesive element . . . . . . . . . . . . . . . . . . . . . . . . . . . Four noded cohesive element . . . . . . . . . . . . . . . . . . . . . . . . . . . Simulation domain and finite-element meshes for (a) simulations of the tension experiment, (b) simulations of the shear delamination experiments, and (c) simulations of the asymmetric four-point bending experiments. The red line in each mesh highlights the cohesive elements used to model interfacial failure. Simulation fit (black line) and experimental results (gray lines) for (a) normal stress vs. normal displacement for the tension experiments and simulation, (b) shear stress vs. shear displacement for the shear delamination experiments and simulation, and (c) force vs. displacement for the asymmetric four-point bending experiments and simulation. . . . . . . . . . . . . . . . . . . . . . . Normal stress at the cohesive interface used to define the "cohesive zone length". The data is taken when the global force-displacement response is at its first peak. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Shear stress at the cohesive interface used to define the "cohesive zone length". The data is taken when the global force-displacement response is at its first peak.. ......... .............. ......................... Normal (solid line) and shear (dashed line) stress at the cohesive-interface before damage initiation as a function of the distance from the left edge of the cohesive-interface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 39 39 40 41 41 42 43 57 58 58 59 66 67 68 69 Stress-stretch curves measured via uniaxial compression test (19 = 298K) (cf.[52]) and the numerical fitting using both a neo-Hookean and Langevin m odel. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9-2 Schematic of the geometry and finite element mesh for the free-swelling problem. The horizontal AB dashed-line indicate the symmetry line while the vertical AD segment is the axisymmetry axis; only the top right quarter of the body is meshed. Adapted from [101 . . . . . . . . . . . . . . . . . . . . . 9-3 Schematic of the reverse osmosis experiment. A pressure difference Ap = po is applied on the feed side of the membrane (edge CD) to drive the fluid flux across the permeate side (edge AB). The dashed-line represent a porous support disk which prevents any deformation of the membrane at edge AB, but allows fluid to flow freely. . . . . . . . . . . . . . . . . . . . . . . . . . . 9-1 9-4 93 93 94 Simulation domain and finite-element mesh for the reverse osmosis experiment. The undeformed mesh (left) and deformed mesh (right) are shown to illustrate the amount of swelling. . . . . . . . . . . . . . . . . . . . . . . . . 94 Comparison between numerically-calculated steady-state volumetric flux versus pressure-difference curves and corresponding experimental data from [51] 95 9-6 Profiles of (a) chemical potential p; (b) mean normal pressure p; and (c) polymer fraction 5 and normalized concentration c along the thickness of the polymer membrane at steady-state. The 0 normalized position corresponds to the top of the membrane (feed side). Ap = 200 psig (1.38 MPa). . . . . . 96 9-5 9-7 Cauchy stress T22 along the thickness of the polymer membrane at steadystate. The 0 normalized position corresponds to the top of the membrane (feed side). Ap = 200 psig (1.38 MPa). . . . . . . . . . . . . . . . . . . . . Schematic illustration of reverse osmosis. Mean normal pressure, p, chemical potential p, polymer volume fraction 0 and normalized concentration 6 profiles in a dense polymer film. The direction of flux is indicated. The subscript 0 indicate the feed side at x = 0 while L indicate the permeate side at x = 1. Numerically obtained profiles (solid line) are shown against the solutiondiffusion theories' assumptions (dashed line). Ap = 200 psig (1.38 MPa). . . 9-9 Schematic of experiments and simulation for constrained-swelling experiments. The thick dashed-line indicate the solid porous boundary. The total axial swelling is denoted by A = H/HO. . . . . . . . . . . . . . . . . . . . . . . . . 9-10 Comparison between simulations and experiments for crosslinked rubber to swell in hexadecane. The"reaction pressure" is plotted against the axial swelling (A = H/HO). The final height H can be adjusted, so that each point corresponds to a distinct experiment with a pre-defined A. . . . . . . . 97 9-8 98 99 99 11-1 Typical parameters for RO elements using TFC membranes [24. . . . . . . . 104 11-2 Schematic of a reverse osmosis plant. . . . . . . . . . . . . . . . . . . . . . . 105 11-3 Schematic of a spiral wound module (SWM) RO element. . . . . . . . . . . . 105 11-4 Schematic of a TFC membrane showing the three distinct layers. . . . . . . . 105 13 12-1 Cross-section of a TFC membrane (Top). The polyamide layer is almost undistinguishable from the porous polysulfone layer. Do not confuse the black layer (top) for polyamide. The cutting was done using a sharp knife and might have smeared the surface of the polysulfone. A close-up view of the polysulfone porous microstructure (Bottom). . . . . . . . . . . . . . . . . . . . . . . . . 12-2 Experimental apparatus for measuring steady-state flux as function of applied pressure-difference. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-3 Dynamic Mechanical Analysis (DMA) apparatus for low-load tensile testing of thin-film . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-4 "Loading" and "unloading" up to 1000 psig. The membrane is initially conditioned at 300 psig for 2 hours. Each point correspond to a collection of data over 5 minutes to insure steady-state. . . . . . . . . . . . . . . . . . . . . . . 12-5 "Loading" and "unloading" up to 400 psig (bold line). The membrane is initially conditioned at 300 psig for 2 hours. Each point correspond to a collection of data over 5 minutes to insure steady-state. The data of Fig. 12-4 is also shown (Dotted line) . . . . . . . . . . . . . . . . . . . . . . . . . . . 12-6 "Loading" and "unloading" up to 400 psig (bold line) of the poly-amide/sulfone layers only. The layer is initially conditioned at 300 psig for 2 hours. Each point correspond to a collection of data over 5 minutes to insure steadystate. The data of Fig. 12-4 is also shown (Dotted line). Note that the . . . . . . . . . . . . . . . poly-amide/sulfone layer fails at about 600 psig. texture is clearly visible. "failure" (right) . A layer after 12-7 Poly-amide/sulfone A schematic of what is though to happen is shown (left). . . . . . . . . . . . 12-8 SEM of Poly-amide/sulfone layer after "failure" (Top). We can see small holes. An SEM of the metal porous disk also shows how the membrane is "extruded" through it (Bottom). . . . . . . . . . . . . . . . . . . . . . . . . 12-9 Engineering stress vs. engineering strains (%) curves for 6 different specimens. 110 111 111 112 112 113 113 114 115 A-1 (a) cross-section of TBC specimen after the tension test (adapted from [47});(b) crack induced by cross-sectional indentation (adapted from [58]) . . . . . . . 123 A-2 (a) schematic of a push-out test (adapted from [401);(b) schematic of symmetric four-point bending sample and experiment (adapted from [701) . . . . . . 124 C-1 Schematic of linear finite element with natural coordinates . . . . . . . . . . 151 14 List of Tables 5.1 Material parameters for the traction-separation model . . . . . . . . . . . . . 9.1 Material parameters for natural rubber obtained through mechanical testing and isotropic free-swelling in hexadecane. . . . . . . . . . . . . . . . . . . . . Material parameters used/calibrated for the pressure-driven-diffusibn simulation (Toluene). The membrane is the same natural rubber as in previous experim ents. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.2 15 63 89 90 16 Chapter 1 Thesis structure The structure of this thesis is as follows. In Part I, we discuss in detail the experimental and numerical methodology for characterizing interfacial delamination of thermal barrier coating systems. Specifically, the experimental procedure and material is discussed in detail in Chapter 3. Then, Chapter 4 discuss possible constitutive models, or traction-separation laws and numerical implementation to model the interfacial delamination of thermal barrier coatings. Chapter 5 discusses the methodology for obtaining the material parameters of a chosen constitutive model for the interface given experimental data. It is shown that we can reasonably reproduce the macroscopic response of thermal barrier coatings during interfacial delamination. In Part II, the objective is to characterize the diffusion of fluid through crosslinked polymer membranes, and accompanied large deformations through a constitutive model as well as an experimental methodology for obtaining the material parameters of such theory. These coupled problem are ubiquitous, especially in the field of reverse osmosis, where a pressure-difference drives diffusion of a fluid(solution) through a semi-permeable crosslinked polymer. In Chapter 8, we present a summary of a coupled fluid permeation and large deformation theory for crosslinked polymers. Then, representative experiments were used in order to calibrate our model and test its validity. These are shown in Chapter 9. Finally, in Part III, a brief overview of our latest work on water desalination using reverse osmosis is detailed. Specifically, in Chapter 12, we highlight our recent experimental study on thin-film composite membranes and potential work in this area. 17 Part I Thermal Barrier Coatings 18 Chapter 2 Introduction The emergence and industrial application of thermal barrier coatings (TBC) has greatly impacted the design and manufacturing of modern propulsion and electricity-generating gas turbine engines. In such applications, turbine inlet temperatures in the gas path of modern high-performances gas turbines operate at temperatures up to around 14000C. Under these harsh operating conditions where hot corrosion and oxidation significantly affect the durability of structural components specially designed high-melting-point nickel-based superalloy blades and vanes are used. These superalloys melt at around 1300*C, thus requiring additional cooling in regions operating in gas-path where temperature exceeds their melting point in order to preserve their structural integrity. Furthermore, these superalloys are also typically coated with a thermal barrier coating, which we refer as a whole to a thermal barrier coating system (TBC), which acts as a thermal insulator and oxidation inhibitor to the metallic component on which it is deposited, and serves to increase the life of the blade. A typical thermal barrier coating usually consists in a ceramic layer, called the top-coat, and a metallic layer, called the bond-coat, and is illustrated in Fig. 2(a). In present day TBCs, the top-coat layer generally consists of an yttria-stabilized zirconia (YSZ). The choice of material for this ceramic coating is no simple task, where the primary design parameter is that the coating have a low thermal-conductivity. However, when considering the entire superalloy-TBC design, the top-coat must also: i) have sufficient strain compliance so as to withstand the strains associated with thermal expansion mismatch between the coating and underlying alloys, ii) exhibit thermodynamic compatibility with the layers on which it is applied; and iii) have a stable microstructure under equilibrium conditions at high temperature. Due to these complicated design criteria, other materials such as pyrochlore structured materials, perovskite-structured oxides, lanthanide orthophosphates, and silicates have also been studied as potential substitute top-coat materials (c.f., e.g., [49J). Depending on the intended application, there two main coating technologies for YSZ aimed at creating a coating with sufficient strain compliance, these are: i) electron beam physical vapor deposition (EBPVD), and ii) air plasma spray (APS). EBPVD top-coats, Fig. 2(c) feature a feathery vertical columnar microstructure and inter-columnar gaps, with individual columns which transmission electron microscopy (TEM) studies have shown to 'In the literature, the nomenclature TBC is often used to refer only to the ceramic top coat layer of the thermal protection system. In this work, we refer to the entire layered structure, composed of a metallic bond coat and a ceramic top coat, as the thermal barrier coating (TBC). 19 contain microscopic porosities (c.f., [13}). These microscopic porosities provide the beneficial property of reduced thermal conductivity, while the inter-columnar gaps provide strain compliance. The latter technique, porous APS top-coats, Fig. 2(b), is the most commonly used in land-based gas-turbines because of their low cost. APS top-coats feature a "splat" like microstructure which is formed as individual particles are deposited at high temperatures on the surface of interest. Here, both the reduced thermal conductivity and strain compliance are a result of the pores that form between the "splats" as they are deposited. Bond-coat alloys have evolved over the years, but are all designed to be sufficiently rich in aluminum so as to form an aluminum oxide - known as the thermally-grown oxide (TGO) - during exposure to air at high temperatures. The TGO layer is important for several reasons. First, this aluminum oxide is phase compatible with YSZ. The majority of uncoated nickel-based superalloys form complex multi-layered nickel-chromium oxides which are thermodynamically unstable with YSZ (c.f., [441). Second, the formation and slow growth of alumina, on account of it having a small oxygen diffusivity, provides oxidation protection to the underlying metallic component. However, this comes at a cost. The volumetric expansion that arises from the formation and growth of the TGO, together with the stresses generated due to thermal mismatch between the ceramic and metallic layers, can result in the premature degradation of the top-coat-TGO-bond-coat interface and can ultimately lead to failure of the TBC system by spallation of the top-coat. Many different types of failure modes leading to TBC spallation have been observed in laboratory experiments, as well as upon postmortem examinations of coated turbine blades that have experienced actual service. The most widely observed delamination failure of topcoats which are APS-deposited on MCrAlY bond-coats, typically occur at the bottom of the top-coat, either just above or through the TGO, near the top-coat-TGO interface. Due to the complexity of the micro-mechanisms leading to degradation of the adherence of TBCs, presently available mechanism-based theories (c.f., e.g., [41}; [36]) are not yet sufficiently mature to allow reliable quantitative prediction of the degradation of the delamination strength and toughness of TBCs. This state of affairs is no different from many other areas where fracture mechanics is used to assess structural integrity. As remarked by [341, because models of strength and toughness are usually not sufficiently accurate for quantitative predictions - strength and fracture toughness are macroscopic properties that are measured, and not predicted. Thus, at present, an essential component of any lifetime assessment scheme for TBCs must include an experimental determination of the delamination strength and toughness of any given TBC system as function of the relevant thermal history. Fracture mechanics-based experimental approaches for measuring delamination toughness properties have been reviewed by [35]. In contrast to the experimental methods for measuring relevant properties appearing in classical fracture mechanics models of delamination, we shall our attention on measuring properties appearing in cohesive interface models of delamination failures (c.f., e.g. [4]; [17]; [45]; [46]; [7]). In such models, decohesion is regarded as a gradual phenomenon in which separation takes place across a cohesive zone, and is resisted by cohesive tractions. This methodology of modeling fracture requires the specification of interface constitutive parameters, such as the interface stiffness, peak cohesive tractions, and the fracture energy - as represented by the area under the cohesive traction-separation relation. Attractive features of this approach to model delamination fracture are that it is independent of: i) the far-field 20 geometry of the component containing the interface; ii) the specific constitutive response of the material on either side of the interface; and iii) the extent of crack growth. Indeed, the location of the evolving crack/delamination front is an outcome of the calculations based on this methodology. Cohesive laws have been built into finite element analyses by using cohesive finite elements. These surface-like elements bridge nascent cracks, and are compatible with finite element discretization of the material on either side of a potential crack. At this stage of research in the literature the material parameters appearing in interface traction-separation laws are the least well known of the required ingredients for modeling TBC delamination failure using this methodology. There are no standard experimental testing procedures for comprehensively determining the properties in traction-separation relations. The purpose of this work is to report on our novel experiment-plus-simulation-based approach to determine the relevant material parameters appearing in traction-separationtype laws, which may be useful for modeling delamination failures in TBCs. The plan of this Part is as follows. We begin by describing the novel experimental methodology in Chapter 3 as well as discussing specifics regarding the material. Then, in Chapter 4, we report on the nature and numerical implementation of traction-separation constitutive laws. Finally, given a choice of traction-separation law, in this case mixed-mode bilinear traction-separation model, Chapter 5 highlights the procedure for determining the interfacial parameters appearing in the theory based on the experimental results. 21 (b) supeum (CoogAir (a) (c) Figure 2-1: (a) Schematic of a TBC on a superalloy in a thermal gradient; (b) Cross-section of an APS top-coat-sprayed TBC and its associated "splats"-like microstructure;(c) Crosssection of an EBPVD top-coat-sprayed TBC and its associated columnar microstructure (adapted from [13]) 22 Chapter 3 Experimental characterization of interfacial properties of thermal barrier coatings A various array of experimental methods exist in order to investigate the material properties of thermal barrier coatings. A review of these experimental techniques is discussed in Appendix A. However, as previously mentioned, there are no standard experimental testing procedures for comprehensively determining the interfacial properties of thermal barrier coatings in traction-separation relations. The purpose of this section is to report on that new experimental procedure which consists three distinct experiments: i) a standard tension-delamination experiment; ii) a novel shear-delamination experiment; and iii) a novel asymmetric four-point bending mixed-mode delamination experiment. 3.1 Material The air-plasma-sprayed (APS) TBC system investigated in this work was prepared for us by colleagues at the Center for Thermal Spray Research at the State University of New York at Stony Brook. A dual-layer NiCoCrAlY bond-coat, which is made of a coarse Amdry 386-4 layer layered over dense Amdry 386-2 (Sulzer Metco Inc., USA), with a thickness of 325 pm was applied applied by high-velocity oxygen fuel (HVOF) on an Ni-based superalloy substrate, Inconel 718, of 2.5mm in thickness. The ceramic top-coat was produced using agglomerated and sintered 8wt.% yttria-stabilized-zirconia powder (Metco-204NS, Sulzer Metco Inc., USA) and plasma-sprayed on the bond-coat with an 8mm nozzle Sulzer Metco F4 MB torch. Prior to spraying, the substrate surface was cleaned with alcohol and grit blasted using #24 alumina grit at a pressure of 80 psi. The bond-coat was strategically sprayed to achieve surface roughness of approximately 7-8 pm Ra. Fig. 3-1 (a) shows an SEM micrograph of the cross-section of an as-sprayed TBC sample. Since delamination failures of TBC systems in real components occurs after some minimal time in operation, at which point a TGO layer has formed, we have chosen to develop our methodology for characterizing interfacial traction-separation properties on a specimen with an existing TGO layer. Specifically, we have investigated the properties of TBC coupons 23 which have been isothermally exposed to air at 1100*C for 144 hours. Fig. 3-1(b) shows the cross-section of such a TBC specimen taken using a HITACHI TM3000 scanning-electronmicroscope operated at 15kV. The TGO layer, approximately 5 pm thick, is clearly visible in this figure. It is worth mentioning that our choice of exposure time was, although hinted by the work of [41] and references therein, mostly arbitrary. It is known that TGO growth occurs primarily within the few days when thermally-exposed, and then reaches a quasi-saturated thickness. Hence, after exposure time greater than 144 hours, the TGO thickness does not increase significantly. However, it would be of interest to investigate TBC materials subjected to shorter thermal-exposure, whereas the interfacial properties measured would most probably differ. SEM cross-sections of TBC for various isothermal-exposure are shown in Fig. 3-2(a). Furthermore, it is important to note that the choice of isothermal exposure also depends on the TBC material. The same 144 hours temperature cycle was applied on a different TBC material fabricated by Sulzer Metco 1 and resulted in an incredibly damaged interface, as shown in Fig. 3-2(b). A closer analysis under energy-dispersive X-ray spectroscopy (EDS) 2 showed that the "dark interface" is actually void, indicating spallation of the top-coat from the bond-coat. 3.2 3.2.1 Tension Delamination Experiment Specimen preparation One of the simplest and most widely used methods to determine the bond-strength of an interface is a tension experiment(c.f., ASTM standard C633-79, [31}; [59]). The specimens for such an experiment were prepared by first bonding the TBC onto a 3 mm thick 1018 steel substrate, ceramic face down, using a commercial Araldite AW 106/HV 953 epoxy. The bond was then cured at 150*C for 20 minutes according to the manufacturers specification 3. A water-jet machine was then used to create 5 mm square specimens of the steel/TBC assembly, which is shown in Fig, 3-3(a). Next, aluminum tension bars with a cross section of approximately 12 x 40 mm were prepared by gluing 100 pm diameter copper wires, as shown in Fig. 3-3(b). The specimen were then bonded on both sides onto these aluminum tension bars using the same adhesive and held in a bonding clasp to insure proper alignment, Fig. 3-3(c)-(d). Note that a 10 mm thick aluminum plate was clasped with the specimen in the bonding clasp and further tied using c-clamps so as to avoid difficulties during the experimental set-up, which we will discuss shortly. Furthermore, the copper wires were used in order to insure a uniform 100 pm thick layer of adhesive between the bonded parts (c.f., [15]). Finally, the whole bonding clasp was inserted into the oven and cured using the same curing specification. A schematic of a sample prepared in this manner with key dimensions 1 Amdry 997 bond-coat VPS-sprayed using a F4VB gun and grit blasted with corundum followed by a vacuum diffusion heat treatment at 10800C for 4 hours. Then, a ceramic top-coat (Metco 204C-NS) is APS-sprayed using a F4MB gun. 2 JEOL5910 SEM available at the CMSE 3 Sun Electronic Systems, Inc.TM, maximum temperature of 315*C. 24 is shown in Fig. 3-4(c). 3.2.2 Experimental procedure The experiment was carried on a displacement-controlled, low-force single column tabletop Instron 5944, shown in Fig. 3-4(a). This machine has a maximum load capacity of 2 kN where the force is measured with a 0.5% accuracy and an applied vertical displacement resolution of 0.094 pm. Furthermore, due to the small scale nature of experiment, the improved alignment of the grips 4 makes this machine suitable for our purposes. The specimen was removed from the bonding clasp and carefully installed in the tension machine, ensuring the best alignment possible between the grips. This is where the 10 mm thick aluminum plate comes in handy, since any misalignment of the grips might create a bending moment on the interface high enough so as the break the interface. Then, the tension-bars were spray-painted to produce a speckle pattern for measuring their relative displacements using a digital-image-correlation (DIC) apparatus 5. At this point, the c-clamps are carefully removed along with the aluminum plate. The resulting set-up is shown in Fig. 3-4(b). Fig. 34(d) shows a close-up image of the specimen with the speckle pattern in the tension machine. The customized camera set-up used to measure the displacement via DIC is illustrated in Fig. 3-5 and consists in two Manfrotto 454 micrometric positioning sliding plates that allows vertical and horizontal micro-positioning of the camera. A Mitutoyo lens was used 6 together with a high resolution 5 MP digital camera 7 with a 2448 x 2048 pixels resolution. The experiments were carried by imposing a nominal displacement-rate of 1 pm/s. The number of captured images for the DIC can be varied for any given experiment, but for the experiments described in this chapter, four frame was taken every second. Such an experiment allows for a measurement of the interfacial stiffness and strength in a pire tensile mode. The experimental results are presented in the following section. 3.2.3 Results of tension experiment The normal stress versus normal relative displacement curves obtained from tension experiments are shown in Fig. 3-6. Four TBC specimens were tested and their response is reasonable consistent. The normal strses is computed by dividing the measured applied force by the area of the TBC specimen. It can be observed that the response is essentially linear until a peak stress-level is reached, at which point the interface fails abruptly. Fig. 36(a) and Fig. 3-6(b), respectively, show the error bars in the displacement measurements, and the stress measurements. We attribute the major part of the substantial scatter in the displacement measurements shown in Fig. 3-6(a) to the inherent noise both in the encoded 4 5 Instron 5940 Series specification manual available at www.instron.us Vic-2D version 4.4.1, Correlated Solutions, Inc. TM. The software uses a cubic B-spline interpolation algorithm to track the movement of the grayscale within any selected pixel over the course of time, and several points on both the top and bottom of the interface are selected. An extensometer can be drawn across the interface and the algorithm will track the pixel position of the selected point over the spectrum of "deformed" photos taken 6 Model 378-802-2. 5X magnification, 0.14 numerical aperture, 34 mm working distance, 40 mm focal length. 7 Model GRAS-50S5M-C from Point Grey Research 25 for the stepper-motor used for displacement actuation in the Instron 5944 machine and in the numerical algorithm behind the DIC, where sub-micron displacements are measured. However, as per the machine's load measurement accuracy and precision, the error bars in Fig. 3-6(b) are within 1% of the measured load. Figs. 3-7(a) and (b) show representative SEM micrographs of the bond-coat side fracture surface of the TBC specimen obtained after a tension experiment. Even at low magnification, Fig. 3-7(a), the TGO is clearly visible as the dark annular regions in the micrograph. At a higher magnification, Fig. 3-7(b) reveals regions of exposed bond-coat separated from the top-coat by an annular TGO layer. Fig. 3-8, shows an SEM micrograph of the top-coat side fracture surface of the same specimen. Here, we see a dark circular region of exposed TGO, known as the TGO "cap", which mates one of the exposed bond-coat regions shown in Fig. 3-7. As shown schematically in Fig. 3-9, the micrographs in Figs. 3-7 and 3-8 reveal that the fracture path proceeds in an alternating fashion between the top-coat, the TGO, and the bond-coat, and is always near or at the top-coat-TGO-bond-coat "interface". We thus conclude that the data in Figs. 3-6 from our tension experiments reflect the fracture properties of the "interface" and not of either the top-coat or the bond-coat alone. 3.3 3.3.1 Shear Delamination Experiment Specimen preparation Guided by the experiments of [62], a novel experiment was developed and used to measure the interfacial failure properties in shear. Specimen preparation was as follows. First, a TBC plate was placed ceramic face down, onto a sacrificial aluminum substrate and a waterjet cutter was used to cut small 3 x 6 mm TBC coupons (the cutting jet entering through the superalloy side). This procedure prevents the brittle ceramic to shatter due to the impact of the abrasive jet. Then, multiple 300 pm wide and 350 pim deep top-coat "islands" were carefully machined by using an MTI-STX-201 diamond-wire-saw 8. Fig. 3-10(a)-(b) shows the diamond-wire-saw and the actual cutting of the TBC specimen into top-coat "islands". A sample specimen (not the one used for the experiments) is shown in Fig. 310(c) for illustration purposes. 3.3.2 Experimental procedure The shear delamination experiments were conducted in a flexure-based precision biaxialmicro-mechanical testing apparatus shown in Fig. 3-11(a). Details regarding this apparatus may be found in [22]. Briefly, a Burleigh Inchworm actuator 9 which can travel over 6.35 mm with a 4 nm resolution is used to drive the TBC specimen against a steel blade. Velocities ranging from 4 nm/sec up to 1.5 mm/sec can be achieved while applying a maximum axial load of 15 N. For our purposes, the steel blade was moved at a constant velocity of 0.5 pm/sec until the top-coat "islands" were completely sheared off. The tangential force applied on the steel blade was measured using a flexure-based shear load-cell, which has a resolution 8 9 Available at the CMSE Crystal-SEM facilities Inchworm Motor IW-700 Series with Controller 6000ULN Series by Burleigh 26 TM. of 225 piN, while the relative displacement between the steel blade and the base of the top-coat "islands" is measured using DIC (c.f., Fig. 3-11(b)) where the speckle pattern is shown). Fig. 3-11(c) illustrates a schematic of the sample with dimensions and the steel tool. Note that the specimen shown in Fig. 3-11(c), up to three experiments may be performed since there are three distinct top-coat "islands" whose shear delamination response may be measured. A similar camera set-up as the one illustrated in Fig. 3-5 was used. 3.3.3 Results of shear delamination experiment Fig. 3-12 shows shear stress versus tangential displacement curves for three shear delamination experiments. The shear stress is computed by dividing the measured reaction force by the area of the top-coat-bond-coat interface of the island. Given the inherent local variability of the interfaces in TBCs, the three TBC samples that were tested showed reasonably consistent traction-displacement responses in shear 10. Observe that the traction-displacement response in shear consists of an initial linear region, followed by a region of nonlinearity which leads to a "plateau"-like region of "inelastic" deformation, which subsequently ends with complete shear delamination. SEM micrographs of the bond-coat side fracture of the specimen obtained after the shear delamination experiment are shown in Figs. 3-13(a) and (b), where the two lighter colored regions in Fig. 3-13(a) correspond to machining marks caused by the diamond-wiresaw during sample preparation. A higher magnification image of the bond-coat side fracture is shown in Fig. 3-13(b). Again, as in the tension experiments, the TGO is clearly visible as a dark annular region surrounding a region of exposed bond-coat. On the corresponding top-coat side fracture surface, Fig. 3-13(c), the TGO "caps" previously discussed are also observed. The SEM micrographs for the shear experiments (Figs. 3-13), are consistent with those for the tension experiments (Figs. 3-7 and 3-8) and clearly suggest a similar fracture path which proceeds in an alternating fashion through the bond-coat, the TGO, and the topcoat, and is always near or at the interface (Fig. 3-9). Thus, we conclude that in the shear experiments-as in the tension experiments-the measured traction-displacement response reflects the interfacial delamination response of the top-coat-TGO-bond-coat interface. 3.4 3.4.1 Asymmetric Four-Point Bending Specimen preparation Finally, guided by the experiments in the literature (c.f., e.g., [55]; [67]; [70]), four-point bending experiments on asymmetric beams were conducted in order to characterize the mixed-mode delamination response of a TBC interface. Specimen preparation is as follows. First, two TBC plates measuring 15 x 7 mm and 15 x 15 mm (which have been prepared by water-jet machining, see previous section) are bonded 8 mm apart onto an aluminum substrate, which is used as a metal stiffener, using the same epoxy adhesive (Araldite AW '0 Note that in contrast with Fig. 3-6 in tension, the experimentally measured response in shear is less noisy because of the precision of the micromechanical apparatus used to conduct the latter experiments. 27 106/HV 953) and curing specifications 1 . An illustration of this step is given in Fig. 3-14(a). A second water-jet cutting operation is then used to cut 3 mm thick beams from the layered structure, see Fig. 3-14(b). The result is an asymmetric beam-bending specimen, Fig. 314(c), which is shown schematically with all the its dimensions in Fig.3-14(d). Note that in Fig. 3-14(c), the excess of glue on the right-hand-side of the specimen (on the outside), does not affect our measurement of the interfacial properties since the fracture is expected to initiate in the inner part of the left-hand-side of the beam, as indicated by the red arrow. 3.4.2 Experimental procedure For this experiment, we used the flexure-based micro-mechanical testing apparatus shown in Fig. 3-15. Details regarding this testing machine may be found in [251 and [26]. Briefly, an electromagnetic voice-coil actuator which has a stroke of 12.7 mm and a maximum continuous stall force of 86.2 N is used in order to apply a normal loading with a 0.5 mN resolution. The top rollers and bottom rollers have a span of 9 and 26 mm, respectively, with the top rollers centered in between the bottom rollers. A close-up view of the specimen in the experimental apparatus is given in Fig. 3-16. The relative displacement between the top and bottom rollers is measured using DIC, while the reaction force on the top rollers is measured using the flexure-based load-cell. The digital image camera used (QImaging, Retiga 1300i, Fast 1394, with Nikon Nikkor Lens) here was positioned on a tripod approximately 0.5 m away. 3.4.3 Results of asymmetric four-point bending experiment Fig. 3-17(a) shows the force versus displacement curves for three asymmetric four-point bending experiments. The load initially increases in a linear fashion up to the point where the strain energy available is sufficient to initiate cracking at the interface. At this critical load, crack initiation is observed near the top-coat/bond-coat interface and the behavior becomes non-linear with an observed load plateau. The critical load is ~~24 N and is reached after a relative roller displacement of ~ 150 pm. The load plateau corresponds to a region of rapid and unsteady crack propagation along the interface. When crack growth reaches a steady-state regime (corresponding to the end of the plateau), the load continues to increase as the remaining uncracked beam is deformed in bending. It is important to note that the machine used in this experiment imposes a load on the beam during bending. Thus, if the beam loses its load carrying capacity, the machine will rapidly push the specimen through to its next stable configuration. This is what is experimentally observed to occur, and is in accordance with the experimental results of [68]. This phenomenon is better understood by plotting the displacement versus time behavior as shown in Fig. 3-17(b). During the fracture process, a sudden displacement jump from ~ 150 pm to ~ 250 pm can be observed which suggests an unstable crack burst. Figs. 3-18(a) and (b), respectively, show SEM micrographs of the bond-coat-side and top-coat-side fracture surfaces. Consistent with our previous observations for the tension and shear experiment, the TGO is visible on the bond-coat side as dark annular material 'IThe stiffener increases the elastic energy available for delamination. 28 surrounding exposed bond-coat material. On the opposite side, the top-coat fracture surface again shows dark TGO "caps". For all three experiments, the approximate size of those interfacial defects are of the order of 50 pm. 29 (a) TGO-+ (b) Figure 3-1: SEM micrograph of the cross-section of a TBC: (a) an as-sprayed specimen, and (b) an isothermally exposed specimen (144 h at 1100*C) 30 I12h3 (a) (b) Figure 3-2: (a) distinct isothermally exposed specimen for various period of time;(b) an isothermally exposed Sulzer specimen (144 h at 1100 C) 31 (a) (b) (c) (d) (e) Figure 3-3: (a) square 5 mm specimens of the steel/TBC assembly; (b) aluminum tension bars with 100 pm copper wires;(c) aluminum tension bar with TBC/steel specimen;(d) fully assembled tension specimen in bonding clasp with a gripped aluminum plate;(e) curing of the bond 32 tensio bar Top coat I 1018 Stee3 I --... -substrate tnsion >nr = (a) - superallo -- (c) (d) (b) Figure 3-4: (a) low-load single column tabletop Instron 5944 ;(b) with installed specimen;(c) schematic of the tension specimen with dimensions;(d) close-up of the tension specimen in the Instron 5944 33 Figure 3-5: set-up of digital image correlation system 34 14 1210 - Cr' Cr' ARJ.v off 84-D 642 0 -0.5 0.5 1 1.5 Normal displacement (m) (a) 14 12- 10 E2 U1 0 86 42 0 -0.5 0.5 1 1.5 Normal displacement (pIm) (b) Figure 3-6: stress versus displacement curves from four tension experiments: (a) with error bars for the displacement measurements; and (b) with error bars for the stress measurements. Error bars correspond to one standard deviation on an average taken from ~ 40 data points. 35 Top coat BondT coat ca exposed - . 500m . . . .GO 50OJm (a) (b) Figure 3-7: SEM micrographs of the tension fracture surface facing the bond-coat side of the specimen: (a) low magnification, and (b) high magnification. Fig.(b) highlights regions of exposed bond coat surrounded by rings of thermally grown oxide (TGO). - ----------- ;'7 TopO coat_Z; Bond a 50 pm Figure 3-8: SEM micrograph of the tension fracture surface facing the top-coat side of the specimen. The micrograph highlights a region of exposed TGO, referred to as the "TGO cap", which matches regions of exposed bond-coat on the opposing fracture surface. Figure 3-9: schematic of the experimentally observed fracture path. 36 (a) (b) (c) Figure 3-10: (a) MTI-STX-201 diamond-wire-saw cutter from the CMSE Crystal-SEM facilities; (b) machining of the TBC specimen; (c) TBC specimen with machined top-coat "islands" 37 loadcell flexure actuator Burleigh inchworm actuator 300 pm 3mm Top coat(b) Bond coat steel tool 350 Am --.-..-.......... . - -- ~325 Am -- (c) superalloy substrate Figure 3-11: (a) micromechanical biaxial apparatus used in the shear delamination experiments ;(b) machining of the TBC specimen;(c) schematic of the shear specimen with dimensions. 38 . 121 1086 42 0 0 0.5 1 1.5 2 Shear displacement (pm) Figure 3-12: shear stress versus shear displacement curves from three shear delamination experiments. fracture surfaces exposed nd coat Top coatav Bond coat 500pm 50Apm achining marks (a) (b) TGO 'cap' Top coat Bond coat 50 pAm (c) Figure 3-13: SEM micrographs of the shear delamination fracture surfaces. (a) and (b) show fracture surfaces facing the bond-coat side, while (c) shows a fracture surface facing the top-coat side of the specimen. 39 (a) (b) (c) 3 m - adumninum stiffener Top coat 350 pm substrate 7 MM substrate 7mm 8 n= Bond coat a325 pm 15 mm (d) Figure 3-14: (a) TBC plates glued 8 mm apart on an aluminum substrate; (b) water-jet cutting operation; (c) final asymmetric beam specimen with aluminum stiffener; (d) schematic of the bending specimen with dimensions. 40 Figure 3-15: micro-mechanical apparatus used in four-point bending experiments. Figure 3-16: close-up of the bending specimen in the testing machine. 41 40 35 30 - 25 20 2 15 10 5 U 0 100 200 300 400 Displacement (jtm) (a) 400 300 200- 0 - S100- 200 400 600 800 1000 1200 Time (s) (b) Figure 3-17: asymmetric four-point bending results for three experiments. displacement response, and (b) displacement vs. time response. 42 (a) force vs. Top coat Bond coat neoed L hO (a) Top.'e coat Bond,. coat 50 Ism (b) Figure 3-18: SEM micrographs of the asymmetric four-point bending fracture surfaces. (a) Shows a fracture surface facing the bond-coat side, while (b) shows a fracture surface facing the top-coat side of the specimen. 43 44 Chapter 4 Interface traction-separation constitutive model Guided by the experimental observations shown in Chapter 3, we now take a step back and discuss possible numerical tools for characterizing interfacial properties of thermal barrier coatings. The following chapter focuses on the so-called interface traction-separation laws and their numerical implementation in a finite element software. 4.1 Cohesive Zone Modeling In order to model failure of the top-coat-TGO-bond-coat interfaces in TBC systems, the cohesive zone model is a convenient approach that relates displacement jumps across the interface at the crack tip with the tractions on the interface. This type of model has been successfully applied in recent years to a number of decohesion and fracture problems (c.f., e.g., [451, [62], [7], [34]) and we believe it has great promise in the modeling of failure of TBC systems. Typically, a cohesive interface is introduced to the finite element discretization of the problem of interest, through the use of special interface elements which obey an interface traction-separation law. Importantly, cohesive elements also differer from regular continuum elements, in that they may have zero initial thickness in the direction normal to the interface. Thus, when the cohesive elements are undamaged, the cohesive interface approximates an un-cracked portion of the body. This also motivates the necessity of a traction-separation relation, rather than a standard continuum constitutive law, for the description of the constitutive behavior of these cohesive elements 1. The traction-separation constitutive relation provides a phenomenological description of the complex microscopic processes that lead to the formation of new traction-free crack surfaces. Such cohesive interface models describe fracture as a separation process occurring at the crack tip where debonding is assumed to be confined to a small region called the cohesive zone. Instead of using classical macroscopic 'Cohesive elements may also have a non-zero thickness in the direction normal to the interface, and such cohesive elements may be modeled with standard continuum constitutive laws. However, these non-zero thickness elements are best suited for the description of adhesive joint type problems, rather than fracture type problems. 45 fracture properties to describe crack nucleation and propagation, the interface tractionseparation relation usually includes a cohesive strength and a cohesive work-to fracture. Once the local strength and local work-to-fracture criteria across an interface are met, decohesion occurs naturally across the interface, and traction-free cracks form and propagate along element boundaries. With a view towards modeling the top-coat-TGO-bond-coat interface response which was experimentally measured in Chapter 3, we herein present three different traction-separation constitutive theories, each having their own interesting characteristics. 4.2 Mixed-Mode Bilinear Traction-Separation Model Many different traction-separation-type models have been proposed in the literature. To fix ideas, consider the schematic of the pure-mode bilinear traction-separation interface constitutive relation (for a two-dimensional situation) shown in Fig. 4-1 (e.g. [8]). With respect to this figure, (tN, 6N) and (tT, ST) represent the normal and tangential components of the traction vector t and the separation vector 6 at a point of the interface. The parameters KN and KT represent the elastic stiffnesses of the interface for normal and shear separation, respectively. Damage is taken to initiate when the following criterion is satisfied max DN - } TI(O -1. 2 (4.1) Here, * The parameters t and 4t denote the values of the interface strengths in the normal and shear directions, respectively. Also, (x) is the Macauley bracket used to describe the ramp function with value 0 if x < 0 and a value x if x > 0. Thus, no damage is presumed to occur under a purely compressive loading, (tN < 0, tT = 0), at the interface. Under continued loading, damage grows until final fracture occurs when the following simple mixed-mode criterion is satisfied: GN GT -- + -- = 1. G*N GT (4.2) Here, * The parameters GN and GI are two additionalmaterialproperties, which respectively represent the fracture energy of the interface for pure normal and pure shear separations. 2 With respect to the particular traction-separation law considered here and depicted schematically in Fig. 4-1, "initiation of damage" refers to the initiation of microstructural defects at particular values of the normal and tangential tractions which lead to the degradation of the elastic stiffnesses KN and KT. 46 Finally, unloading subsequent to damage initiation is assumed to occur linearly towards the origin. Reloading also occurs along the same linear path until the "softening envelope" is reached. Then, upon further loading, damage will continue until final fracture according to (4.2). This interface model has already been implemented has a built-in feature in the finiteelement analysis package Abaqus/Standard [53]. In such a model, the material parameters that need be determined are {KN, KT, to, (4.3) More details regarding the specific of this traction-separation law can be found in [8] and [53]. Remark. It is important to note that this traction-separation model does not account for frictional sliding in shear after failure of the interface. If after failure the two surfaces of a failed interface come into contact, then such effects are easily account in Abaqus/Standard [531 by allowing for a frictional contact interaction, with a constant Coulomb friction coefficient p. 4.3 Implementation of an Elastic-Plastic-Damaging TractionSeparation Constitutive Law In the previous section, we have discussed the basic concepts of a simple bilinear tractionseparation constitutive model available as built-in features in finite-element softwares. However, interface separations, which are ubiquitous in nature, cannot always be represented by such a simplified bilinear model, and thus more guided constitutive laws must be developed. Here, we present a methodology for implementing arbitrary cohesive traction-separation laws (TSLs) for cohesive elements as a user-element subroutine in the widely-used general-purpose finite element program Abaqus/Standard [53]. We begin by deriving the variational formulation of the governing equations for cohesive modeling in Section 4.3.1. Then, as an example, we will consider the implementation of an elastic-plastic-damaging traction-separation constitutive law based on the work of [541 and [60], which is summarized in Section 4.3.2. The fully-implicit time-integration procedure required for implementing the chosen model is given in Appendix B. In Sections 4.3.3 and 4.3.4 details of the general finite element implementation for cohesive elements are discussed. Furthermore, a discussion about the main aspects of writing a user-element element subroutine for implementing the present traction-separation law is given in Appendix B. 4.3.1 Variational formulation of the macroscopic force balance The displacement solution variables are governed by the partial differential equation for the balance of momentum, the strong form of which, in the current configuration, along with 47 appropriate boundary conditions is given by: divT + b = 0 on Tn = t on u =u on Bt, S 1, (4.4) S2 , I where Bt denotes the body in the current configuration, and S, and S2 are complementary subsurfaces of the boundary BB of Bt. On the surface S we prescribe surface traction t = Tn, and on the surface S2 we prescribe displacements u. Here, T is the Cauchy stress and b are the generalized body forces. The div operator is the spatial divergence in the current configuration. Let q be a virtual variation in the displacement u. The weak form of the balance of momentum is obtained by multiplying (4.41), by 77, which yields (4.5) b-. d =0. .fdTqdv+ Further, using the divergence theorem we may write divT -7 dv = f J Tn -7 da - j T:gradq d, (4.6) where grad is the gradient with respect to the current configuration. Then, (4.42) may be written as T:gradldv+j b -dv=O. Tn,qda-J j OBt (4.7) Bt B j Tn-qda- j T grad 17dv+ Tn -. 17da - j 8Be~ Bt- b -dv+ T: grad7dv + I T+n+,+da=, JI+ JB B+ B JOB+ ( We now consider the body Bt to be composed of two bodies B+ and B7 separated by an interface I, see Fig. 4-2. We may then apply (4.44) separately to both B+ and B7 which yields: b . vdv + f T~n- - 71fda = 0, JBt-J- where I represents the locus of the crack interface in the current configuration. The unit normal to I+, n+, points from B+ to B- and n- is the unit normal to I-, pointing from Bj to Bt. Adding together equations (4.45) yields, JB Tn -7da- T::grad77dv+ tJBJIBtJ b - qdv- ft-[[7711da= 0, (4.9) 49 where [[77]] =7l+ - q- represents the jump in the virtual displacement q across the interface, and t is the interfacial traction. Here we have made use of the fact that Bt = Bt + Bt, 8Bt = OBt + OB, and I+ = I-, the law of action and reaction T+n+ = -T-n-, and the facts that n+ = -n- and T+ = T- = T at the interface I. At the interface, writing n = n-, 48 the interfacial traction is t = Tn. Further, since the variation v vanishes the boundary S2 where the displacements are prescribed, J(b - (4.10) t- [[71 da =0. t1da- - T : grad) dv + Bt Isi 11 The first two terms in equation (4.47) arise from the classical mechanical equilibrium of the body Bt without a cohesive interface, while the last term represents the contribution from the presence of a cohesive interface. The first two terms are taken care of by Abaqus/Standard. Here we concentrate on contribution from the last term in the weak statement (4.47), and to do so we need to specify a suitable traction-separation constutive law between the traction t and the displacement jump (4.11) S= [u]= u+ - U-, which we consider in the next section. 4.3.2 Summary of the interface constitutive model We consider two bodies Bt and B- separated by an interface I, see Fig. 4-2. Let {ei, 62, e 3 } be an orthonormal triad, with 6^ aligned with the normal n - n7 to the interface, and {e 2 , e} in the tangent plane at the point of the interface under consideration. We assume that the displacement jump may be additively decomposed as ([54],[601) 6 = (4.12) e + 6P, where 6' and 6, respectively, denote the elastic and plastic parts of 8. Additionally, for later use we also introduce a second decomposition of 6 into normal and tangential parts, 8= 6N+bT, 6N = (non)b = (6.n)n = 6 Nn, 6 r = (1-non)6 = 6- 6 N, (4.13) where the scalars 6 N and 6T represent the displacement jump in the normal and tangential directions, respectively. We are concerned with interfaces in which the elastic displacement jumps are small, but the plastic displacement jumps may be arbitrarily large. Following [541, for small elastic displacement jumps we assume t = K6' = K(6 - 6P), (4.14) with K the interface elastic stiffness tensor, taken to be positive definite. We consider an interface model which is isotropic in its tangential response, and take K to be given by K=KNn 0 n + KT(1 - n 9 n), (4.15) with KN > 0 and KT > 0 the normal and tangential elastic stiffness moduli. The interface traction t may also be decomposed into normal and tangential parts, 49 tN and tT, respectively, through t=tN+tT, tN=n (9 nI) t = (t - n) n =tNfl tT(1 n D n) - t= t - tN. (4.16) The quantity tN represents the normal stress at the interface, and we denote the magnitude of the tangential traction vector tT by T -tT~ (4.17) and call it the effective tangential traction, or simply the shear stress. We take the elastic domain in our elastic-plastic model to be defined by the interior of the intersection of two convex yield surfaces. The yield functions corresponding to each surface are taken as Pi(tI si) _< 0, i = N, T, (4.18) and henceforth we identify the index i = N with the "normal" mechanism, and the index i = T with the "shear" mechanism. The scalar internal variable sN represents the deformation resistance for the normal mechanism, and sT represents the deformation resistance for the shear mechanism. In particular, we consider the following simple specific functional form for the yield functions: 4N = tN - SN < 0, T = +(-tN) - ST 0, (4.19) where p represents a friction coefficient. The surface 4i = 0 denotes the i yield surface in traction space, and "*N nN n N T, - = '*TT T,_ + /In), (4.20) denote the outward unit normals to the yield surfaces at the current point in traction space, see Fig. 4-3. The equation for e, the flow rule, is taken to be representable as a sum of the contributions from each mechanism mN +6 mT, 3?>O, "of = 0,N=with +mN= nN, mT -. (4.21) Note that since mT # nT, we have a non-normal flow rule for the shear response.3 Finally, during inelastic deformation, an active mechanism must satisfy the consistency condition 3iD = 0 when =, (4.22) which serves to determine the inelastic deformation rates 6i' when inelastic deformation occurs. 3 Such a non-normal flow rule for the shear response is common in interface models for friction, where there is strong effect of the compressive normal traction on the resistance to plastic flow, but the plastic flow in shear is essentially non-dilational. 50 Next, let iP = 5|(C)dC, (4.23) define the accumulated plastic displacements for each of the two individual mechanisms, and further let sp = V(6N))2 + a(6TP)2, (4.24) define an equivalent relative plastic displacement, where a represents a coupling parameter between the normal and shear mechanisms. In the theory under consideration, the interfacial resistances si are allowed to soften according to a simple linear damage rule si = s,,o( - D), (4.25) where si,O denotes their initial value, and 0 D if S < sp,. - (4.26) P iS .P <<5 denotes a damage parameter in the range 0 < D < 1. In other words, the interface deforms plastically in a perfectly-plastic fashion until a critical value . for the equivalent relative plastic displacement is reached. Then, the interface incurs damage until ultimate failure at P = S7f, at which point the damage parameter D = 1. In a numerical simulation, an interface after failure (D = 1) is not able to carry tensile traction, but for compressive traction, the response to penetration is purely elastic, and under such circumstances the compressive normal stress goes up quickly with penetration depth. As to the shearing response of an interface after failure, its shearing resistance is purely frictional when tN is compressive; otherwise, the two surfaces across a failed-interface are free to slide over each other without any resistance. Further discussion on interfacial failure is detailed in Appendix B. Details about the time-integration procedure of the said constitutive model is given in the Appendix B. 4.3.3 Finite element development In this section, details are given regarding the general finite element development for cohesive elements. Henceforth, boldfaced upper-case letters (e.g. N, L) denote matrices while boldfaced lower-case letters (e.g. t, u) denote column vectors. Further, in a finite element discretization, the integration point quantity for the displacement jump 6 defined in Section 4.3.2 will henceforth be denoted by 6 (cf. equation (4.29)). We begin by defining local and global cartesian coordinate systems, see Fig. 4-4. The global coordinate system (X, Y) is a fixed cartesian coordinate system, while the local coordinate system (x, y) is aligned with the cohesive element in the current deformed configuration of the body. Then, * The nodal displacement vector in the global coordinate system, denoted by u, may be 51 transformed to a local nodal displacement vector -aby a rotation matrix R through4 (4.27) i = Ru. The local nodal displacement jumps S can then be computed from the local nodal displacements fi through 6 = LU, (4.28) where L is the local displacement-separation matrix (again specific to the cohesive element type chosen). The local displacement jumps 6 at the integrationpoints within the elements are then interpolated from the local nodal displacement jumps 6 through 8=N6, (4.29) where N is the shape function matrix. Finally, use of eqs. (4.27) and (4.28) in (4.29) leads to the following relationship between the local displacement jumps at integration points and the global nodal displacement = Bu, where B = NLR, (4.30) with B the global displacement-separation relation matrix. In summary, given the nodal displacement vector u in a fixed global coordinate system, the matrix B is used to compute the displacement jumps 6 at the integration points with respect to a local coordinate system, i.e. the current configuration, which rotates with the deforming cohesive element. Based on the aforementioned approximate displacement jump field, we define the elementlevel displacement residual f= (4.31) fBtda. The element stiffness is then given by KuU = Of, (4.32) which using the residual defined in (4.31) is aBt~d Kt a Oh a3u [BatB =9 - BT 86 Bda. (4.33) It is the traction-separation constitutive law discussed in Section 4.3.2 that relates the local interface traction t to the local displacement jumps 6 at each integration points. 4.3.4 Two-dimensional linear cohesive element We will now consider the specific case where the cohesive interface is modeled using twodimensional linear cohesive elements each having four nodes whose numbering follow the 4 The rotation matrix R depends on the specifics of the cohesive element chosen and will be specialized in Section 4.3.4. 52 conventions of the finite element code Abaqus/Standard 6.10 in which it is implemented. Each node has two degrees of freedom, as shown in Figure 4-5. As defined in equation (4.27), the global nodal displacements u are transformed to local nodal displacements fi via a rotational matrix R. In the theory developed here, the cohesive elements are extended to be used at finite deformations. The mapping from reference to current configuration is done via this rotation matrix R 4 R= ( A(m (4.34) 1T=1 where each diagonal block [A(m is the 2 x 2 transformation matrix defined by: 5 A(m) [ cos sin ] - sin O cosO jI (4.35) The angle 0 represents the angle between the global (X - Y) and local (x - y) coordinates system of the cohesive element. The local coordinate system is obtained by taking the mid points between opposing nodes in the deformed cohesive element, that is between node 1-4 and node 2-3, and drawing a straight line. The local nodal displacement jumps 8 are obtained from the local nodal displacement as follows (c.f. Fig.4-4 for the relationship between 6 and fi) 1 =u7 - f 1 , 2 = i8 -u 2, 3 =f 5 -L, 4 = Ej- i4. (4.36) Based on the above relations, the local displacement-separation relation matrix L relating the nodal quantities = Lii is given as 1 0 0 0 0 -1 0 0 0 0 -1 0 0 0 0 -1 0 0 1 0 0 1 0 00 1 0 0 0 (4.37) 1 0 0- The interpolated displacement jumps 6 = N8 at each integration points in the cohesive element are obtained from the local nodal separation quantities in conjunction with the shape function N, N =[ 0N2A 0 N1 0](4.38) 0 N2' where the linear shape function in the normalized coordinate C axe given as N = 2 ,+CN2 = (4.39) 2 In our specific case, we have chosen the integration points to be positioned at both ends of 5 and C = 1, respectfully. From equations (4.27) through The direct sum (symbol) of two matrices A and B is defined as A@B= 53 . the cohesive element, i.e. at C = -1 (4.29), the global displacement-separation relation matrix B = NLR is expressed as: B= B -Ncoso N1 sin0 -Nsin -Ncos0 -N 2 cosO -N 2 sinG N2 sinO -N 2 cosO N2 cosG -N 2 sinO N2 sinG N2 cosO N1 cosO -NsinG N1 sinG N1 cosG J (4.40) 4.4 Notes on the Implementation of a Thermo-Mechanically Coupled Elastic-Plastic-Damaging Traction-Separation Law In this section, we further our discussion by including thermal effects at the interface (e.g. heat flux) leading to a thermo-mechanically coupled theory for cohesive elements (cf.[27], [481). Only the variational formulation will be discussed as the numerical implementation is identical as before. This model is, to some extent, independent of the bulk by certain assumption regarding the damage. 4.4.1 Variational formulation The thermo-mechanical problem of a solid containing cracks must have a solution for which mechanical and thermal equilibrium is satisfied. Hence, we present a simple theory for cohesive zone modeling where the displacement solution variables as well as temperature solution variable are the only degrees of freedom. In the absence of an internal heat supply, the strong form of the coupled partial differential equations for the balance of momentum and balance of energy, in the currentconfiguration, along with appropriate boundary conditions, are given by divT+b=0 on Bt, Balance of momentum Tn=t on Sx, u= on St, (4.41) c{ div(igradV) on Bt, Balance of energy - grad - n=q on Sq, on S9, where Bt denotes the body in the current configuration, S. and St are complementary subsurfaces of the boundary BB of Bt, (Sx n St = 0) on which the displacement u and traction t are prescribed. Similarly, S and Sq are complementary subsurfaces of Bt of Bt, (s, n s q= o) on which the temperature V and heat flux q are prescribed. Here, T is the Cauchy stress, b are the generalized body forces, r is the material's thermal conductivity and c is the specific heat per unit spatial volume. The div operator is the spatial divergence in the current configuration. In (4.41), the coupling between the two PDE's will arise through the the constitutive equation for the conductivity r. The numerical solution to the PDE for the momentum balance (4.41), has already been discussed in Section 4.3.1 and will thus not be the focus of this discussion. Here, details are given with regards to solving the balance of energy (4.41)2 of solid containing cracks which are modelled using cohesive elements. 54 Let w be a virtual variation in the temperature 0. The weak form of the balance of energy is obtained by multiplying (4.41)2 by w, which yields J cw dv- (4.42) div(xgrad V)w dv = 0. where "grad" is the gradient with respect to the current configuration. Further, using the divergence theorem, and considering a spatially constant scalar conductivity X, we may write f div(rgradV)w dv = OB,wrgrad O-n da - f, rgrad O - grad w dv, (4.43) (Agrad 9- n)w da = 0. (4.44) Then, (4.42) may be written as L chw dv+ grad 9- grad w dv- We now consider the body Bt to be composed of two bodies B+ and B- separated by an interface I, see Fig. 4-2. We may then apply (4.44) separately to both Bt and B- which yields: f B+ f chw dv + t c~w dv + JBt rgrad V -gradw dv - f~i Kgrad?- gradw dv - JOB+ (rgrad V - n)w da - f(q+ .n+)w+da = 0, J+ . t LB 7 (,rgrad V n)w da - jl-(q- n-)wda = 0 (4.45) where I represents the locus of the crack interface in the current configuration. The unit normal to I+, n+, points from Bt to BT and n- is the unit normal to I-, pointing from Bto B+. Similarly, the interface heat flux q* represents the heat flux across the interface from one side to the other, along direction nI, respectively. Adding together equations (4.45) yields, fBcwdv JB V(grad-n)wda+q- [[w]da=0, (446) j where [[w]] = w+ - w- represents the jump in the virtual temperature w across the interface, and q is the interfacial heat flux. Here we have made use of the fact that Bt = Bt + B-, OB = oBt + Bt, and I+ = I-, the continuity condition q+ - n+ = -q- - n-, and the facts that n+ = -n- and q+ = q- = q at the interface I. At the interface, we write n - n-. Further, since the variation w must vanish at the boundary S, where the temperature is prescribed, f(chw +Kgrad o - gradw) dv55 qw da + q - [[w]] da = 0, (4.47) The first two terms in equation (4.47) arise from the classical energy balance of the body Bt without a cohesive interface, while the last term represents the contribution from the presence of a cohesive interface. The first two terms are taken care of by Abaqus/Standard. Here we concentrate on the contribution from the last term in the weak statement (4.47), and to do so, we need to specify a suitable constitutive law between the heat flux q and the temperature jump (4.48) 0-0-, 0 = [[11= which must take into account the nature of the cohesive elements. As is standard, a constitutive theory for heat transfer across a cohesive interface, which complements the mechanical elastic-plastic-damaging traction-separation law presented in Section 4.3.2, to which we refer, would need to be specified. 56 tN' 0 N tT T0 tT KT N I 6N (b) (a) Figure 4-1: Schematic of the bilinear traction-separation interface constitutive relation, showing (a) the pure normal response (no tangential deformation), and (b) the pure shear response (no normal deformation). Bn -n -------- Figure 4-2: Schematic of interface between two bodies B- and B-. 57 tN nNI L =0 SN Figure 4-3: Schematic of yield surfaces for the normal and shear mechanisms. global i = local Ru ) (u5 ,U 0 (6i 5 , fi 6 ) u 04, i!8 )j i y ( 3 - 4 64 7 2 (fA 3 ,ii 4 ) ) (fi 1 ,fi 2 W. X Figure 4-4: Four noded cohesive element 58 3 yx 4 4 x 2 3 1 =+2 1 V 8 Figure 4-5: Four noded cohesive element 59 X 60 Chapter 5 Estimation of the material properties in the mixed-mode bilinear traction-separation constitutive model The exact nature of the traction-separation constitutive law describing the TBC interface is not well known at this point. However, due to its simplicity, the mixed-mode bilinear assumption described in Section 4.2, along with the battery of experiments presented in Chapter 3 can be used in order to determine the interfacial properties. Hence, the purpose of this chapter is to describe a straightforward methodology, using finite element simulations and experimental results, for determining the interfacial properties of said traction-separation constitutive relation. 5.1 Material Parameters In order to determine the material parameters {KN, KT, to I to, Gc , G } involved in the bilinear traction-separation constitutive model described in Section 4.2, we have used Abaqus/Standrad [53] to conduct finite-element analyses to simulate the tension, shear and asymmetric bending experiments discussed in Chapter 3. We iteratively varied the values of these parameters so as to obtain acceptable matches of the simulated stress (load)-displacement results with the corresponding experimentally measured results from the three experiments. 5.2 Numerical Simulations In all finite-element simulations discussed below, we made the following modeling assumptions: * A plane-strain assumption is used to approximate all of the experiments. 1 'Plane-strain simulations are used only to reduce the complexity and computational time associated with simulating the actual three-dimensional geometries of the experiments. We have also performed simulations of these experiments using a plane-stress assumption, and the results from these simulations (not shown) were nearly identical to those for the plane-strain simulation shown here. 61 * The top-coat, bond-coat and superalloy substrate are modeled as linear elastic and discretized using four-noded fully integrated elements (CPE4), while the cohesive zone, i.e. the bond-coat/top-coat interface, is modeled using linear four-noded elements (COH2D4) with zero initial thickness. * The top-coat is taken as having a Young's modulus E = 37 GPa and Poisson's ratio V = 0.18, while te bond-coat and supperalloy substrate have E = 200 GPa and v = 0.29 [68] * For interfaces which have failed and can possibly come into contact after failure, such as in the shear and bending experiments, we have used a value of t = 0.25 for the friction coefficient between failed interfaces. The specifics regarding the simulations of each experiment are as follows: * The finite-element mesh for the tension experiment is shown in Fig. 5-la. The 1018 steel stiffener is modeled as linear elastic (E = 200 GPa, v = 0.29) and also discretized using CPE4 elements. The bottom surface AB (bottom of the superalloy) is fixed in space, while a tensile displacement is applied on the top surface CD (top of the 1018 steel). e The finite-element mesh for the shear experiment is shown in Fig. 5b. In these simulations, the bottom surface AB of the superalloy is fixed in space, while the steel tool (modeled as an analytical rigid surface) is displaced horizontally at a constant velocity against the top-coat 'island" until failure of the interface occurs. * The finite-element mesh for the asymmetric bending experiment shown in Fig. 5c. In these simulations, the aluminum stiffener is modeled as an elastic-perfectly plastic material with Young's modulus E = 69 GPa, Poisson's ratio v = 0.34 and yield strength Y = 258 MPa, and discretized using CPE4 elements. 2 Both top and bottom rollers are modeled as analytical rigid surfaces. The top rollers are fixed in space while the bottom rollers are constrained to move only in the vertical direction at a constant velocity. To prevent rigid body motion, a single node immediately underneath the right top roller is constrained from moving horizontally. 5.3 Calibration Results The material parameter estimation procedure that we followed is described below: (i) First, the material parameters defining the normal stiffness and strength were estimated by simulating the tension experiment as KN %50 to70 MPa/Lm, tN = 12to l4 MPa. We note that the value of t' can be read directly from the experiment results shown in Fig. 3-6 however, the stiffness KN requires simulating the experiment since the 2 The aluminum stiffener deforms plastically during the experiments. 62 experimental data includes not only the stiffness of the interface but also that of the substrate, bond-coat, top-coat and steel stiffener. (ii) Second, the material parameters defining the tangential stiffness and strength were estimated by simulating the shear experiment as KT; 75 to 125 MPa/pm, 4 ~~8 to 10 MPa. As in the tension experiment, the value of tr can be read directly from Fig. 3-12, while estimation of KT requires simulating the experiment. (iii) Using the estimates for the material parameters {KN, tON, KT, 4} from the tension and shear experiments, we used the simulations of our asymmetric four-point bending experiments to estimate values for {Gr, GIT}. The steps (i), (ii) and (iii) outlined above were repeated until an acceptable fit to the stress (load) vs. displacement data from all three experiments was obtained. Fig. 5-2a-c show comparisons of our simulation results with the three different experimental measurements. The simulations reproduce the experimental measurements with reasonable accuracy. Of course, due to the nature of the pure tension and shear experiments, the stain-softening branch is not observed in the experiments, and are thus not fitted well by the simulations. Therefore, the softening portion of the simulated results is only partially shown in Fig. 5-2a and b as a dashed line. The behavior of the softening branches in tension and shear are controlled by the fracture energies Gi and GC which are fit based solely on the asymmetric four-point bending experiments. The material parameters for the bilinear traction-separation law used to obtain these fits are listed in Table 5.1. Table 5.1: Material parameters for the traction-separation model tN (MPa) G (J/m2 ) KT (MPa/pm) t4 (MPa) 60 13.5 100 100 9 G (J/m 2 ) KN (MPa/pm) 100 Jb = 2GcN and 6f =2 , Based on the presumed cohesive law, we may define intrinsic normal and tangential cohesive lengths T No respectively. For the pure mode behaviors shown schematically in Fig. 4-1, these length scales correspond to the values of the normal and tangential displacement jumps when the interface completely loses its load-carrying capacity. For the material parameters listed in Table 5.1 f = 22.2pm. = 14.8pim, -b 63 Following [561, we may define yet another length scale usually referred to as the "cohesive zone length" and denoted by 1CZ. The cohesive zone length ICZ is defined as the distance from the crack tip to the point where the maximum cohesive traction is attained. Using the bending simulation results shown in Fig. 5-2c, we compute the normal stress at the cohesive interface when the global force displacement response is at its first peak (which corresponds to a displacement ~ 140 pm in Fig. 5-2c). The result is shown in Fig. 5-3, and from the simulation we measure a cohesive zone length of CZ = 1.2 mm. Further, this length scale is usually predicted (cf. [56]) from an equation of the form CZ= ME (5.1) (tN 2 where E is the Young modulus of the surrounding material, and M is a parameter that depends on both the interface model and the geometry. In the work of [56], the parameter M is determined from simulation. Using our simulated results, and E = 37 GPa, we find a suitable value of this parameter to be M = 0.06. We note that eq. 5.1 is derived based on an ideal situation in which the crack is opened in pure Mode-I while surrounded by a homogeneous material of stiffness E (cf. [321). In such cases the parameter M is usually in the range 0.2 to 1.0. The "cohesive zone length" is important because it is this length which must be appropriately discretized with cohesive finite-elements in order to achieve accurate results. In our simulations, we use cohesive-elements with size 1e = 27.5 pm, which leads to 1cz N =- le 40 elements in the cohesive zone. This number of elements in the cohesive zone is significantly larger than the conservative minimum of N = 10 suggested in [56]. Thus, our finite-element mesh with respect to the discretization of the cohesive-interface is sufficiently fine. We also note that in our shear and tension simulations, the cohesive-element sizes are le = 20 pm and le = 18 pm, respectively, and thus give an even finer discretization of the cohesive zone. It is worth mentioning that if we use the value of M determined purely based on the normal parameters and the normal stress computed from the bending simulation, we can compute a tangential cohesive length . C Z = 0.06E G =2.7 mm. The simulated tangential cohesive zone length, see Fig. 5-4, is 2.4 mm, which is very close to this value. Some remarks: 1. Note that wit the exception of the values of the values of tT and t4 which are easily estimated directly from the experimental data, the iterative fitting procedure described 64 above and applied to the no-linear problem at hand yields values for the other material parameters which are inherently non-unique. The specific values for the material parameters {KN, KT, G"N, GT} shown in Table 5.1 represent only one possible fit to the experimental data. 2. Our "shear delamination" experiments do not induce pure tangential relative displacement 6 T across the interface. We have found through our simulations that the relative contribution of the normal displacement 6 N to the shear tangential displacement 3 r in the experiment is controlled by the position of the steel tool relative to the "top-coat island". Specifically, the vertical point of contact between the steel tool and the topcoat "island" (se close-up in Fig. 5-1b) induces a moment o the island which induces a normal relative displacement 6 N, and a corresponding normal traction tN. Fig. 5-5 clearly show that our shear delamination experiment is not a pure shear experiment, but exhibits some non-negligible mode-mixity.NOte that the fact that the delamination experiment is not a pure shear experiment does not affect our material parameter estimation procedure. 3. Note also that in both the experimental and simulated shear stress vs. displacement curves shown in Fig. 5-2b, after the initial linear response there is some amount of non-linearity. This non-linearity is a direct result of the geometry of loading, which as mentioned above, leads to a combined tangential and normal relative deformation across the interface. The mixed-mode deformation leads to contributions to the damage evolution by both the tangential and normal modes, and in return, results in a nonlinear behavior of the resulting shear stress vs. shear displacement curve. 4. The traction-separation law considered here is not sufficiently rich to account for the plateau-like region observed in the shear delamination experiments; cf. Fig. 5-2b. A more refined interface constitutive model, such as the one presented in Section 4.3 which allows for some inelastic deformation, might be required in order to capture the full extent of the apparent "plateau". 5. With regards to the asymmetric four-point bending experiments, Fig. 5-2c,the simulations predict that there is a decrease in the load-carrying capacity of the system after crack initiation, which appears to be in disagreement with the experimental results. However, this discrepancy is a result of the fact that the simulation was carried out under "displacement control" while the experiment was conducted under "force control". Thus, the simulations are able to probe states of the system which in the experiment are unstable and thus jumped over. We note that the force level for the "Maxwell-line" computed from the simulation data shown in Fig. 5-2c is 22.2 N; this value lies within the experimentally measured upper value of 25N and lower value of 20 N. 65 C D 1018 steel stiffene ,,-''top coat elements } coat bondcoativ superalc )y substrat e A B (a) top coat island steel tool (analytical surface) cohesive elements IH-H-H-HI +R::V superalloy substrate IIIrII MIM H I I I I! iI iI iI Ij Ii iI iI iI II Q i i ! i ! i ! I i i H !i i ! i i i i i i A B (b) top coat cohesive elements top rollers (analytical surface) aluminum stiffener bond coat superalloy_ substrate _.. bottom rollers (analytical surface) (c) Figure 5-1: Simulation domain and finite-element meshes for (a) simulations of the tension experiment, (b) simulations of the shear delamination experiments, and (c) simulations of the asymmetric four-point bending experiments. The red line in each mesh highlights the cohesive elements used to model interfacial failure. 66 - 12 12 - .12 .1. 14 e10 10 8 rn8 6 6 0 Simul ation --- Expei riments 1 1.5 S2 0 1w C 0.5 4 . -- 2 . 4 2 ol 2.5 0.5 --- Simula tion ---- Experi mnents 1 1.5 2 2.5 Displacement (Am) Displacement (pm) (b) (a) 40 Simulation -35 - Experiments 30 25 C.) 0 20 100 20 30 15 10 5 0 100 200 300 400 Displacement (jpm) (c) Figure 5-2: Simulation fit (black line) and experimental results (gray lines) for (a) normal stress vs. normal displacement for the tension experiments and simulation, (b) shear stress vs. shear displacement for the shear delamination experiments and simulation, and (c) force vs. displacement for the asymmetric four-point bending experiments and simulation. 67 6 4 2 0 -2 -4 0 Z -6 .CZV -8 -10 C 2.5 5 7.5 10 D istance from left edge (mm) Figure 5-3: Normal stress at the cohesive interface used to define the "cohesive zone length". The data is taken when the global force-displacement response is at its first peak. 6 PL4 2 _ -CZ UD -10 $.4 t 5 0 10 Distance from left edge (mm) Figure 5-4: Shear stress at the cohesive interface used to define the "cohesive zone length". The data is taken when the global force-displacement response is at its first peak. 68 10 OT N 8 Op 01 6 4 ;.D. -4-D 2 -4 0 250 500 750 1000 Distance from left edge (pm) Figure 5-5: Normal (solid line) and shear (dashed line) stress at the cohesive-interface before damage initiation as a function of the distance from the left edge of the cohesive-interface. 69 70 Chapter 6 Concluding remarks We have developed a novel experiment-plus-simulation-based methodology for characterizing interfacial delamination properties in a representative traction-separation model which should be useful in simulations of delamination failures of TBCs. The experiments consist of load-displacement measurements obtained from three different experiments: (i) a standard tension-delamination experiment; (ii) a novel shear-delamination experiment; and (iii) a novel asymmetric four-point bending mixed-mode delamination experiment. SEM observations of the fracture surface of isothermally exposed (144h at 1100*C) TBC specimens subjected to the tension, shear and bending experiments revealed that interfacial delamination of TBC proceeds with a crack path that travels in an alternating fashing through the top-coat, the TGO and the bond-coat. Which confirms that the experimentally measured load-displacement curves reflect the interfacial properties of the topcoat/TGO/bond-coat interface and supports the idea that a TBC delamination failure may be modeled as an "interfacial failure" process using traction-separation-type models. The experimental methodology proposed here, together with the traction-separation models, provide a pragmatic and straightforward method for determining interfacial properties of micron-dimensioned multilayered structures such as TBCs. However, much remains to be done. Specifically: * Substantial further experimentation is required to determine how the interfacial properties evolve as a function of thermal history. The use of a thermo-mechanically coupled model such as the one suggested in Section 4.4 might be required. * Further experimentation is also required in order to shed light on (a) the nature of the mode-mixity of interfacial failure, and (b) the role of friction for interfaces which have failed and can possibly come into contact after failure. The use of a "custom-made" cohesive element with a more complex constitutive response, as the one proposed in Section 4.3.2, might be required. 71 Part II Reverse Osmosis Membranes 72 Chapter 7 Introduction The transport of gases, vapors, and liquids through polymeric membranes has received a great deal of attention in the past decades and is of importance for a variety of engineering and biomedical applications (cf. [431,[21]). For example, when a patient is in a state of renal failure, hemodialysis is used as an artificial kidney operation to achieve the extracorporeal removal of waste product such as creatinine, urea and free water from the blood. The separation process, or fluid removal, is achieved via ultrafiltration through a semi-permeable polymeric crosslinked membrane by altering the hydrostatic pressure of the dialysate compartment, causing the free water and some dissolved solutes to move across the membrane along a created pressure gradient. This mode of operation is often referred to as reverse osmosis where the mechanics of transport is molecular diffusion of penetrant molecules dissolved in the membrane. Many transport models have been proposed to describe the process of reverse osmosis. However, one of the most useful and most widely used has been the solution-diffusion mechanism first introduced by [42]. Their simple formulation involved a linear model where the "permeate" flux increases linearly and without limit with an ever increasing pressuredifference. Although unrealistic, their model has proven very useful in qualitatively describing water desalination through reverse osmosis using cellulose acetate membranes. With recent membrane technology advances, reverse-osmosis processes are becoming more affordable and accessible, and their application now range from simple fresh water production to pharmaceutical needs. Such human-oriented applications require an understanding of the mechanisms that affect the transport of organic and non-organic solutions through polymer membranes, which the classical solution-diffusion model cannot provide. For this reason, notable attempts at reformulating the solution-diffusion model have been proposed over the past decades, see [611 and references therein. In fact, one of the limitation of the classical solution-diffusion model is well described in the work of [51]. They observed that the flux of solvent forced across a polymeric membrane increases non-linearly as the applied pressuredifference increases, until a critical pressure-difference is reached; at which point the flux "saturates". In order to explain this phenomenon, [501 suggested that nonlinearities in the pressure-difference vs. flux relationship originates from: (i) the diffusing fluid's molecule size, or partial molar volume; high partial molar volume leads to more apparent nonlinearities; (ii) the active presence of the membrane when highly swollen (high fluid content); and (iii) the concentration dependence of the diffusion coefficient. However, in his proposed theory, [50] 73 neglected the last two contributions, essentially disregarding the presence of the crosslinked polymer membrane. In a recent paper by [57], the authors approached the second point, viz. the active presence of the membrane, by assuming a priori a pressure profile over the membrane thickness. They then used jump balances to describe the fluid-membrane system and derived a flux equation which depends on the pressure gradient, which ironically is assumed from the get go. Although a step forward compared to their predecessors, [57] still do not fully answer the questions, namely, for instance, the variation of the pressure profile across the membrane from it's "dry" to highly swollen state. Although useful in applications where the polymeric membrane does not deform, the applicability of the existing solution-diffusion models (and reformulations) are limited when the amount of swelling and mechanical deformation is non-negligible, as it is the case for most reverse osmosis applications involving polymeric membranes (cf. [23],[9,[52]). Furthermore, very few studies exist regarding the exact nature of the diffusion coefficient, or mobility; and there is yet no agreement on its specific form. Hence, the objective of this paper is to make use of the coupled fluid diffusion and large deformation theory for crosslinked polymers developed by [11] among others, with a view towards application to reverse osmosis. We believe that this theory, which encompasses aspects of: 9 the thermo-mechanically coupled theory for fluid permeation in elastomeric gels (cf. [11]). An essential kinematical ingredient of our theory is the multiplicative decomposition F = FF', with Fs = A51, As > 0, of the deformation gradient F of the crosslinked polymeric membrane into elastic and swelling parts F' and F, respectively, with the swelling taken to be isotropic with a swelling stretch A; * the Flory-Huggins theory for the free energy change due to mixing of the fluid with the polymer network (cf. [191,[16]). This theory contains an interaction parameter X, which characterizes the interaction between the fluid and the underlying polymer network; * the non-Gaussian statistical-mechanical model which accounts for the limited chain extensibility of polymer chains in order to model the change in configurational entropy of crosslinked polymers; and, complements the current solution-diffusion theories in that the contribution of the crosslinked polymeric membrane comes in naturally without the need of the current underlying assumptions. We further describe the diffusing mechanism using Yasuda's version of "free-volume" theory ([66]) where diffusion coefficient vary exponentially with the fluid concentration. In addition, we propose a systematic numerical finite element simulations plus experimentsbased methodology for characterizing the material parameters of such a theory. Highlights of this methodology includes (a) mechanical simple compression testing; (b) isotropic stressfree steady-state free-swelling experiments; and (c) steady-state pressure-induced reverseosmosis experiments. Using the latter experiment, we show that that some assumptions of solution-diffusion theories, notably profiles of pressure, chemical potential and concentration through the membrane are not accurate when large swelling are considered. 74 In what follows, we first summarize the thermodynamically-consistent constitutive theory of [11]. In Section 8.2, we further specialize the constitutive equations, and discuss a specific set of useful equations for reproducing experimental results. The governing partial differential equations and initial/boundary conditions are reviewed in Section 8.3. The theory has been numerically implemented in the commercial finite element software Abaqus/Standard [531 by writing special user-elements which couple mechanical deformation and fluid permeation. The methodology for characterizing the material parameters in the theory is described in Section 9.2. The calibration of the theory allow us to compare our numerical results with the theoretical assumptions of the solution-diffusion models. Finally, we close with concluding remarks in Section 10. 75 76 Chapter 8 Summary of the coupled fluid permeation and large deformation theory for crosslinked polymers of [11] Abstract In this paper, we point out the major limitations of the classical solution-diffusion model and subsequent reformulations for modeling pressure-driven diffusion processes such as reverse osmosis. The objective is to make use of the continuum-level, thermodynamically-consistent theory for the semi-permeable crosslinked polymer membrane theory of [11] which takes into account coupled fluid permeation and swelling of the membrane to describe reverseosmosis applications. We make use of Yasuda's version of the "free-volume" theory ([66]) to describe the diffusing mechanism. This theory is numerically implemented and we propose a systematic simulation plus experiments-based methodology for characterizing the material parameters involved in such theory. Highlights of this methodology includes the following experiments: (a) mechanical simple compression; (b) three-dimensional isotropic steadystate free swelling; and (c) steady-state reverse osmosis experiments. Using our numerical simulation capability, we find that the underlying assumptions of solution-diffusion theories, namely the profiles of pressure, chemical potential and concentration through the polymer membrane are not accurate when swelling and deformation of the semi-permeable membrane system. Keywords: Solution-diffusion theory; Reverse osmosis; Large deformations; Diffusion; Polymer membranes 77 8.1 Constitutive theory for isotropic materials In this section,' we summarize the coupled fluid permeation and large deformation constitutive theory for crosslinked polymers of [11], specialized for isothermal deformations at fixed temperature V. The theory relates to the following basic fields: x = X(Xt) F=VX, motion J=detF>O deformation gradient F = F'F' multiplicative decomposition of F Fe, Je = detFe > 0 elastic distortion FS, J = detF8 = (1+ QcR) swelling distortion F= ReUe polar decomposition of F1 Ce =FeTF* elastic right Cauchy-Green tensor T = TT Cauchy stress Me = JFTTFe- T TR = JTF T Mandel stress Piola stress free energy density per unit reference volume OR entropy density per unit reference volume CR number of moles of fluid per unit reference volume C number of moles of fluid per unit current volume fluid molar volume; fluid chemical potential grad A spatial gradient of the fluid chemical potential j spatial fluid flux vector 8.1.1 Constitutive equati ons 1. Free energy 1aR = ?R 'CeC R), (8.1) where Ice represents a list of the principal invariants of the right Cauchy-Green tensor Ce. 2. Mandel stress. Cauchy stress. Piola stress 'Notation: We use standard notation of modern continuum mechanics. Specifically: V and Div denote the gradient and divergence with respect to the material point X in the reference configuration; grad and div denote these operators with respect to the point x = X(X, t) in the deformed body; a superposed dot denotes the material time-derivative. Throughout, we write F*I- = (F)-1 , F~T = (F)~T, etc. We write tr A, sym A, skw A, AO, and sym 0 A respectively, for the trace, symmetric, skew, deviatoric, and symmetricdeviatoric parts of a tensor A. Also, the inner product of tensors A and B is denoted by A: B, and the magnitude of A by JAI = v/A: A. 78 The Mandel stress is defined by [ 9IPR(ICR)I M=*=2C (8.2) 0OCe> T = 2J-Fe &OR(Ce, E'Ce CR) F . The Cauchy stress is, (8.3) Also, the Piola stress is given by T, = JTF T . (8.4) Let jf J-1 (tr Me) P=3 P - 3 Je (trT). (8.5) G (86 (8.6) define a mean normal pressure. 3. Chemical potential The partial derivative of the free-energy OjNR(ICe,C) OCR represents the chemical potential of the fluid. 4. Evolution equation for F The evolution equation for the swelling distortion F is F8 = D8 F8 , (8.7) where D' represents the stretching due to swelling of the material due to absorbed fluid molecules and is given by = j3J'~l 1. (8.8) The swelling is related to the concentration of fluid molecules in the material through J = 1 + 92c, (8.9) where Q is the volume of a fluid molecule - or molar volume, (presumed to be constant). 5. Fluid flux The spatial fluid flux j is presumed to obey a Fickian-type law j = -m grad p 79 (8.10) where m is a scalar mobility coefficient, which in general is an isotropic function of the stretch and the fluid concentration. Specialization of the Constitutive Equations 8.2 We list below the specialized form of the constitutive equations proposed by [11]. 8.2.1 Free-energy We consider a separable free-energy of the form PR) = ORMehaNI (ICeCR) 'OR (C, + RPmixing (IcC (8.11) R Here: (i) VRm,,anica is the contribution to the change in the free energy due to the deformation of the polymer network which arises from an "entropic" contribution. Let /tr- 1C,(8.12) N define an effective stretch. Using (8.1)3 and (8.1)7, (8.12) can be written as ._ 1 (1+ QcR)2/3 trCe = Vj2/3 trCe. (8.13) Then, from classical statistical mechanics models of rubber elasticity (cf. [3]), we choose the following estimate for the entropy change due to mechanical stretching - for large values of A it is necessary to use non-Gaussian statistics to account for the limited extensibility of the polymer chains, therefore, rl~me~aacal = -NRkBA2 [(+) +ln si)( ) sin 0 + NRkB In J, (8.14) with kB the Boltzmann constant, and L ( C #fl ,and = 4)(8.15) where ' is the inverse of the Langevin function (x) = coth(x)-x'. This functional form for the change in entropy involves two material parameters: NR, the number of polymer chains per unit reference volume, and AL, the network locking stretch 2. 2 The network locking stretch AL is related to the number of links n in a freely-jointed chain by AL = 80 vf. Next, we take a to also have an energetic component CR = Js' K (In Je)2 ,(8.16) where K is the bulk modulus of the crosslinked polymer. This term in (8.16) is a contribution meant to reflect changes in the internal energy associated with volumetric mechanical deformation of the swollen polymer. The term within brackets represents a contribution to the free energy measured per unit volume of the intermediate space, and multiplication by P converts this to a contribution per unit reference space. Then, using (8.12) through (8.16), and writing (8.17) Go 4f NRkBO, for a temperature-dependentground-state shear modulus, we obtain the estimate PRneanical -(4 = GoA2 + . AL -Go (ii) ) )AO snh - In8 L sih~o (8.18) In J+ P[J K (In Je)2 is a chemical free-energy related to the mixing of the fluid into the host polymer. Following Flory (cf. [16],[191) theory for the contribution to the free energy due to mixing, we write 1faxing /ORa = pca+Rt9c (in (1 __ c__ 1 + QCR + X 1 (8.19) , 1 + fCR where pt is the chemical potential of the unmixed pure solvent, R is the universal gas constant, X is an interaction parameter representing the dis-affinity between the polymer and the fluid. We note that * The "In" term in (8.19) represents the entropy of mixing - the fluid with the polymer - from regular solution theory, which measures the increase in the uncertainty about the locations of fluid molecules when they are interspersed. e The "X" term in (8.19) represents the enthalpy of mixing the fluid within the polymer network. Therefore, when X is positive, it opposes spontaneous mixing of fluid with polymer. In the literature on swelling of crosslinked polymers, the quantity d2= (1+ Ca)~1 = (A)3 = j9~1, 0< # 1, is called the polymer volume fraction. The dry state corresponds to represents a swollen state. 81 (8.20) # = 1, and # < 1 Using (8.18) and (8.19) the total free energy function which accounts for the combined effects of mixing, swelling and elastic stretching is V[(t =#OA1 ipa=Go~~i --AL + JS K (In Go ()P I #+ In Je)2 sinh P + PcR +RcR In -n-G AL -#oJ 3 sinh fo +x1 (8.21) 8.2.2 Stress. Chemical potential The Cauchy stress tensor is given by T = J- [Go((B - (ol)] + Jel~K(In Je)1. (8.22) and (8.23) where (f(i L-(I), Since L- 1 (x) -+oo as x -+ 1, the stretch-dependent shear modulus G (A/AL) -+ 1, The Piola stress, T. = JTF T , is given by Go( -+ oo as TR = Go((F - (oF~ T ) + J8 K(In Je)F-T. (8.24) Also, from (8.6) the chemical potential y is given by A = pi + Rd (ln(1 - 0) + 0 + X02) - KQ(In Je) + KQ(ln J*)2 (8.25) For a detailed derivation of these quantities, we refer to readers to the work of [11]. 8.2.3 Mobility Here we depart for the specialized form considered by [20] and [101. These authors considered a mobility of the power-law form m Oc c (8.26) with n > 1 a constant; this choice for the dependence of m on c models an increase in the mobility of fluid permeation as the polymer network is "opened" by an increase in the local fluid content. This relationship is more conveniently expressed in terms of the polymer volume fraction, viz. 0 = (1+ QcR)~1, 51, 0<0< (8.27) so that, using the fact that c = J-1 cR, . m Oc 82 (8.28) More specifically, [201 proposed the following form for the mobility, M=M (8.29) where M > 0 has units of [m2 /s]. Although good predictions can be obtained using (8.29), not much is known experimentally about the exact nature of the dependence of m on either the stretch or the fluid content. For simple interstitial diffusion the mobility is often taken to be a linear function of the concentration, Dc m = -c (8.30) Ro') with D a constant diffusivity parameter. However, following [14], in solution-diffusion theories the diffusivity is taken to be a function of the average "free volume" of Vf of the polymer/solvent system, D = Do exp [_ij, (8.31) where Do and a are adjustable constants. [66] further assumed that the free volume vf of the polymer/diluent system is proportional to their solvent uptake, so that the solvent in the system provides effectively all of the free volume, (8.32) - Vf OC Ca so that D= Do (exp [-a , (8.33) in which case DD is to be interpreted as the self-diffusion coefficient of the diluent in itself. We note that when the polymer is "dry", q5 = 1, (8.34) yields D = 0. Hence, to allow for a small but finite diffusivity when 4 = 1, we modify expression for D to the form D = Do (exp - 0+-o a (8.34) where -y < 1, so that when q5 = 1, the diffusivity of the diluent in the dry polymer is given by Do x y. Substituting (8.34) into (8.30), and writing the spatial concentration c in terms of the polymer volume fraction, leads to m =Do(1-4) (exp [-a ( JH#R, 1t Henceforth, we take the mobility to be given by (8.35). 83 +-y 4 . (8.35) 8.3 Governing Partial Differential Equations The governing partial differential equations, when expressed in the deformed body, consist of: 1. The local force balance in the absence of body forces and inertial forces, is divT = 0, (8.36) with T given by (8.3). 2. The local balance equation for the fluid concentration, 6 = -divj, (8.37) which using (8.20) may be written in the form JQ4 - divj = 0, (8.38) in which the fluid flux j is given by (8.10), and the chemical potential A is given by (8.25) We also need initial and boundary conditions to complete the theory. Let S, and St be complementary subsurfaces of the boundary OB of the body B in the sense 0B = S. U St and S. n St = 0. Similarly, let S and S be complementary subsurfaces of the boundary: B = S, USj and S, n Sj = 0. Then, for a time interval t E [0, T], we consider a pair of boundary conditions in which the displacement is specified on S and the surface traction on S u=ii on S, x (0,T), Tn = t (8.39) on St x (0,T), and another pair of boundary conditions in which the chemical potential is specified on S, and fluid flux on S p on S, x (0, T), j-n=j on Sjx(0,T),( with n, t, i and prescribed functions of x and t, and the initial data u(X,0) = uo(X) and p(X, 0) = o(X) in B. (8.41) The coupled set of equations (8.36) and (8.38), together with (8.39), (8.40) and (8.41) yield an initial boundary-value problem for the displacement u(X, t) and the chemical potential p(4X, t). In applications, for the case in which the environment consists of a pure and incompressible liquid, the boundary condition on chemical potential j is given by (cf. [501 and references therein) P" + fpg, (8.42) 84 where p.a = pf - p, is the gauge pressure of the environment and 1P is a corresponding integration constant (identical to (8.19)) which depends on the reference pressure chosen and the fluid. 3 Remark Eqn. (8.42) assumes a pure and incompressible fluid. If instead, one would consider the general case of a compressible solution of i = 1, 2... n phases, then the chemical potential of the each phase i would be given by A, =/pi + RV In c, + where wi is a pressure-dependent molar volume. 3 We take p0 = 0 at p, the atmospheric pressure. 85 widp, (8.43) 86 Chapter 9 Calibration of the theory: comparison with experimental results 9.1 Numerical implementation For our numerical simulation, we make use of the implementation of [11] in the finite element software Abaqus/Standard [53]. Special four-noded isoparametric quadrilateral axisymmetric (CAX4) user-elements which couple mechanical deformation and fluid permeation are used. Therefore, in order to model near incompressibility of the elastic deformation, we make use of fully-reduced-integrated elements, both in the displacement and chemical potential degree of freedom. These lower order elements have lower accuracy and thus, it is important to have a sufficiently refined mesh. 9.2 Material parameters In this Section, we study representative examples of application of the theory as found in the literature in order to calibrate our constitutive theory. We study (a) mechanical simple compression experiment; (b) isotropic three-dimensional free-swelling equilibrium of a crosslinked polymer in an unconstrained, stress-free state (cf. [52}); and (c) axisymmetric pressure-difference-driven diffusion of organic solvents across crosslinked polymeric membranes for steady-state diffusion(cf. [51]). With the material parameters therein calibrated, we study axisymmetric directional swelling equilibrium (cf. [52], [38}). Through the following case studies, namely, (a) simple axisymmetric compression; (b) unconstrained isotropic steady-state free-swelling [52]; and (c) steady-state reverse osmosis experiments [51]; we will use the experimental data in order to calibrate the material parameters of our theory, namely {Go, K, AL, x, Do, a, y}. 9.2.1 Simple compression We begin by estimating the mechanical properties of natural crosslinked rubber through uniaxial compression data collected by [52]. As observed in the work of [2], it is reasonable 87 to assume that the behavior of crosslinked natural rubber can be modeled using a nonGaussian (Langevin) theory with a ground-state shear modulus Go and locking parameter AL. As shown in Fig. 9-1, we find that G o - 0.43 MPa, AL ; 1.5. (9.1) Note that a uniaxial tension test should be carried on in order to probe the response over a broader range of stretch and get more representative values for the mechanical parameters. To model the incompressibilityof the elastic deformation, we choose the bulk modulus K so that,' K = 2000. (9.2) G 9.2.2 Isotropic free-swelling Following the experiments of [52], we simulate isotropic, steady-statefree-swelling of crosslinked natural rubber in hexadecane (C16 Hs4) at V = 355K. As it is done in [51], we use this experimental data in order to obtain the "best-fit" dis-affinity parameter "X" that leads to the appropriate amount of swelling. We consider the fluid to be incompressible with constant molar volume (9.3) Q = 3 x 10~4 m3 /mol. The crosslinked rubber has an initial height of HO = 2 mm and a diameter of HO = 2 mm; see Fig. 9-2. Due to the symmetry of the problem, we only model one-fourth of the geometry. The initial finite element mesh consists of 400 CAX4R elements. * For the mechanical boundary conditions, the nodes along edge AB are prescribed to have displacement component U2 = 0 for symmetry, while the nodes along the axis of axisymmetry AD are prescribed to have u1 = 0; the edges BC and CD are taken to be traction free. * For the chemical boundary conditions, the edges AB and AD (the symmetry edges) are prescribed a zero fluid-flux, and on the edges BC and CD the chemical potential p is prescribed as (9.4) (t)= , where p0 is the integration constant (or referential chemical potential of the surrounding solvent). 2 * The initial condition for the chemical potential of the dry polymer is taken to be p(Xt = 0) = -13403.86J/mol. (9.5) This initial condition is computed using (8.25), with p0 = OJ/mol, 4 = 0.999,0 = 355K, J' = 1.0 and x = 0.425. Note that this initial condition depends on the value of the X-parameter and is recomputed if the value of "X" changes. With 0e, A, and J approximate elastic incompressibility, we choose K to be as high as possible. 'To 2 As previously mentioned, we set this value to p0 = 0. 88 Table 9.1: Material parameters for natural rubber obtained through mechanical testing and isotropic free-swelling in hexadecane. Go (MPa) K (GPa) AL X 0.43 0.86 1.5 0.425 denoting the equilibrium values of the polymer volume fraction A, and the volume ratio J, we find that at steady-state ke = 0.48, A, = 1.28, 4, the isotropic stretch J = 2.09. (9.6) The equilibrium swelling ratio of Je equal to 209%, compares with 210% measured in the experiments of [521. The material parameters obtained so far are listed in Table 9.1. Remark It is worth noting that due to differences in the models available in the literature, caution must be taken when comparing our value of X = 0.425. For instance, [521 obtained X = 0.29 using a modified Flory-Huggins free-energy which underestimate the contribution of the enthalpic contribution, while [51] obtained X P 0.425 using a neo-hookean description for the stretching of the polymer chains. Remark For an incompressible elastomeric material modeled using classical Gaussian chainstatistics, [101 show that the equilibrium free-swelling may be calculated from, Ri9 (ln(1 - 4e) + (e + X) + G'neo (01/ 3 -- Oe) = 0. (9.7) where Go,. is the ground-state shear modulus in the neo-hookean theory. Hence A,, and also J, may also be determined, Using x = 0.425 and Go, = 0.7 MPa, we find #e = 0.3549, A, = 1.4, Je = 2.8. (9.8) Thus, the results clearly show that the non-Gaussian theory, which accounts for the effects of limited chain extensibility, predicts a significantly smaller amount of swelling at equilibrium than the classical Gaussian-statistics-based-theory. 9.2.3 Steady-state pressure-driven diffusion - reverse osmosis We consider the problem of pressure-difference-driven diffusion of fluid across a polymer membrane - commonly known as reverse osmosis. This last study allows the calibration of the remaining parameters of our theory, namely {DO, a, y}. The experiments of [51] involve forced-diffusion of various organic liquids (mostly fuels) through a thin crosslinked natural rubber membrane at ?9 = 303K. A schematic of the 89 Table 9.2: Material parameters used/calibrated for the pressure-driven-diffusion simulation (Toluene). The membrane is the same natural rubber as in previous experiments. Parameter Toluene o-Xylene iso-Octane Q (m 3/mol) x Do (m 2 /s) a S1x 1 x 10-4 0.425 3.7 x 10-6 4.5 1.21 x 10-4 0.408 4.3 x 10-6 4.7 1.69 x 10-4 0.572 12.8 x 10-6 5.0 10- 4 1 x 10- 4 1 x 10- 4 geometry for this problem is shown in Fig. 9-4. As illustrated, a thin polymer membrane (LO = 0.0265 cm thick) is supported by a fixed porous rigid plate which prevents deformation of edge AB, but allows the fluid to flow freely. A pressure-difference Ap = Po is then applied across the membrane deforming and steady-state fluid flux is reported for the swollen membrane of thickness L. The mesh consists of 1500 axisymmetric CAX4R elements. o For the mechanical boundary conditions, the nodes along edge AB, BC and AD are prescribed to have displacement component ul = 0, while the nodes along edge AB also are prescribed to have u 2 = 0; the edges CD are taken to be traction free. o For the chemical boundary conditions, the edges AD and BC are prescribed a zero fluid-flux, and on edges AB and CD the chemical potential p is prescribed as per (8.42), viz. (9.9) i = IP+ pga, where pga is the gauge pressure on each side of the membrane, respectively. * The same initial conditions as in Section 9.2.2 are used. The material parameters {DO, a, y} were calibrated in order to fit the experimental results of [51] for Toluene, o-Xylene and iso-Octane and are given in Table 9.2. Additionally, the molar volume Q and X-parameter for the three solvents were taken directly from [51], and are also listed. The fit is shown in Fig. 9-5. The solid lines show steady-state volumetric solvent flux (in units of m 3 /m 2s) versus Ap (in psig) curves for the three organic solvents across the crosslinked natural rubber studied in Section 9.2.1-9.2.2. 3 The theory nicely reproduces the nonlinear dependence of the flux on the driving-pressure difference for all three solvent-polymer pairs. 3 Note that the conversion factors for pressure are: 1 MPa = 145.0377 psi. 90 Profiles of pressure, chemical potential, polymer fraction across the polymer membrane As previously mentioned, solution-diffusion models circumvent explicitly modeling the membrane and its mechanical contribution by making various hypotheses regarding the profiles of pressure, chemical potential and polymer fraction across the polymer membrane. Here, we show that most of these assumptions, although convenient for one-dimensional problems, limited swelling and simple geometries, are erroneous. Figs. 9-6 show the profiles of the chemical potential p, mean normal pressure i and polymer fraction # and spatial concentration e obtained from our simulations using Toluene, across the membrane at steady-state for a pressure-difference of Ap = 200 psig, plotted on the deformed geometry of the membrane. The normalized concentration E is simply given by (9.10) c C )= Cmax where cm. is the highest concentration reached within the polymer at the particular time. In Figs. 9-6a-c, it can be observed that both the chemical potential and polymer volume fraction (and normalized concentration) distribution across the membrane is not linear. In fact, the chemical potential decreases non-linearly from the feed side (normalized position 0) to the permeate side (normalized position 1). Similarly, the polymer volume fraction increases nonlinearly from the feed to permeate side (so that the solvent molecule concentration in the deformed polymer network decreases from the feed to permeate side). Figs. 9-6b shows the profile of the mean normal pressure (recall (8.5)) which decreases almost linearly from the feed to permeate side. It is important to note that the mean normal pressure on the feed side is not equal to the feed pressure. Only the vertical T2 2 stress, shown in Fig. 9-7, is directly equal to the feed pressure. Because of equilibrium only the T22 stress is constant throughout the thickness of the membrane. These profiles obtained numerically are compared to the solution-diffusion assumptions in Fig. 9-8. The left side, x = 0 corresponds to the feed while x = 1 (right) is the permeate side. The numerically-calculated profiles obtained (solid lines) are plotted against those assumed in solution-diffusion theories (dashed lines). It can be observed that the pressure profile on the feed side "jumps up" from the actual feed pressure "jumps down" on the permeate side while decreasing almost linearly across the membrane. The chemical potential and concentration profiles differ from the solution-diffusion assumptions in their nonlinearity across the thickness of the membrane. We emphasize that in our coupled diffusion theory we do not need to assume anything about the profiles of {p, It, 6} across the membrane. In our theory, these are outcomes of the field equations, constitutive equations, and the boundary conditions of the theory. 9.2.4 Limitation of the theory The material parameters for natural rubber in hexadecane are listed in 9.1. As a representative example, we study the steady-state constrained swelling experiments, [521, of natural rubber in hexadecane at V = 355K. The schematic, shown in Fig. 9-9, shows the initial polymer (ABCD) of height Lo constrained in all but the vertical directions. Only edge EE is 91 porous so that the fluid may wet the polymer only on edge CD. The polymer swells only in the vertical direction until it reaches edge EE, at height L, which is a solid mechanical boundary. At this point, additional absorption of fluid molecules by the polymer will create a "reaction pressure" on edge CD (which is now superimposed to edge EE), which is measured as well as the axial stretch (A = L/Lo). The mesh consists of 400 axisymmetric CAX4R elements. * For the mechanical boundary conditions, the nodes along edge AB, BC and AD are prescribed to have displacement component ui = 0, while the nodes along edge AB also are prescribed to have u2 = 0; the edges CD are initially taken to be traction free. e For the chemical boundary conditions, the edges BC and AD are prescribed a zero fluid-flux, and on the edges AB and CD the chemical potential p is prescribed as g)= 0. (9.11) * The same initial conditions as in Section 9.2.2 are used. * When the polymer's edge CD comes into contact with the solid mechanical boundary EE, the "reaction pressure" is measured as the vertical reaction force divided by the constant cross-sectional area of the polymer. The final height L can be adjusted, so that each point corresponds to a distinct experiment with a pre-defined A. The results are plotted against various pre-determined values of axial stretch (A = L/Lo) in Fig. 9-10. We can capture the overall decaying trend of the "swelling pressure" with increasing axial stretch A. The high discrepancies discrepancies between our results and the experiments of [52] suggests that the model may be not be perfectly adequate. As observed by [521, the differences may arise in part due an over-contribution of the Flory-Huggins energy of mixing to the total free energy. Using a chemical free energy similar to (8.19), they were unable to reproduce the experimental data - introducing a modified Flory-Huggins energy of mixing which underestimate the enthalpic contribution allowed them to fit the data within reasonable accuracy. 92 0 -0.2 Cn 4-a U3 -0.4 '-o -0.6 ---Langevin Go = 0.43, AL = 1.5 -experiments 0 -n ._nlI neo-Hookean G = 0.7 MPa . . . .. 0.9 0.8 1 A Figure 9-1: Stress-stretch curves measured via uniaxial compression test (9 = 298K) (cf.[52]) and the numerical fitting using both a neo-Hookean and Langevin model. Solvent D A C "" B 2 3 1 Figure 9-2: Schematic of the geometry and finite element mesh for the free-swelling problem. The horizontal AB dashed-line indicate the symmetry line while the vertical AD segment is the axisymmetry axis; only the top right quarter of the body is meshed. Adapted from [10]. 93 s>TvenT Pga - PO - - D D Pga = PO C polymer membrane L polymer membrane LOt C A A B __Pa =0 permeable plate sogven AI_ =0 B steady-state initial problem Figure 9-3: Schematic of the reverse osmosis experiment. A pressure difference Ap = po is applied on the feed side of the membrane (edge CD) to drive the fluid flux across the permeate side (edge AB). The dashed-line represent a porous support disk which prevents any deformation of the membrane at edge AB, but allows fluid to flow freely. D D C C L LO A B A B Figure 9-4: Simulation domain and finite-element mesh for the reverse osmosis experiment. The undeformed mesh (left) and deformed mesh (right) are shown to illustrate the amount of swelling. 94 1 0.8 CD 0.60.4 U Toluene *0o-Xylene V iso-Octane -Simulation - * 02 100 200 300 400 Ap (Psig) Figure 9-5: Comparison between numerically-calculated steady-state volumetric flux versus pressure-difference curves and corresponding experimental data from [51] 95 - ----- LA Is Kri I ..DV 2.8 2.7 100 2.6 2.5 50 2.4 0.2 2.3 0.8 0.6 0.4 Normalized Position 0 0.2 0.4 0.6 0.2 0.4 0.6 0.8 1 Normalized Position (b) (a) 1 0.581 0.98 -e 0.56[ 0.96 ti 0.54 0.94 polymer fraction 4 -- -normalized concentration a - ni 0 0 0.6 '- - - - 0.2 0.4 0.8 10.92 Normalized Position (c) Figure 9-6: Profiles of (a) chemical potential I; (b) mean normal pressure p; and (c) polymer fraction q and normalized concentration c along the thickness of the polymer membrane at steady-state. The 0 normalized position corresponds to the top of the membrane (feed side). Ap = 200 psig (1.38 MPa). 96 U -0.5- -2- -2.5 0 0.2 0.4 0.6 0.8 1 Normalized Position Figure 9-7: Cauchy stress T22 along the thickness of the polymer membrane at steady-state. The 0 normalized position corresponds to the top of the membrane (feed side). Ap = 200 psig (1.38 MPa). 97 PO =0 -PL feed side % % permeate side C %. %. %. go - 40% L =0 % % flx solvent - solvent m x=0 x=1 Figure 9-8: Schematic illustration of reverse osmosis. Mean normal pressure, p, chemical potential y, polymer volume fraction # and normalized concentration Z profiles in a dense polymer film. The direction of flux is indicated. The subscript 0 indicate the feed side at x = 0 while L indicate the permeate side at x = 1. Numerically obtained profiles (solid line) are shown against the solution-diffusion theories' assumptions (dashed line). Ap = 200 psig (1.38 MPa). 98 porous boundary solvent k I -J E / 1 I., A L L polymer LO B porous boundary * A Figure 9-9: Schematic of experiments and simulation for constrained-swelling experiments. The thick dashed-line indicate the solid porous boundary. The total axial swelling is denoted by A = H/Ho. 5 m 4 * experiments *simulation -S Ue Ci2 3 S 2 U S * 1 U U, UL 1 1.2 1.4 A 1.6 1.8 Figure 9-10: Comparison between simulations and experiments for crosslinked rubber to swell in hexadecane. The "reaction pressure" is plotted against the axial swelling (A = H/HO). The final height H can be adjusted, so that each point corresponds to a distinct experiment with a pre-defined A. 99 100 Chapter 10 Concluding remarks We have specialized the general thermodynamically-consistent continuum-mechanical theory for the coupled fluid-permeation and large-deformation behavior of crosslinked polymers of [11], and in particular introduced a new physically-motivated form of the mobility function in (8.35). We have implemented the theory in the finite element program Abaqus/Standard [53] by writing specialized user-elements, and have proposed a systematic numerical simulations plus experiments-based methodology for characterizing the material parameters appearing in such theory. Highlights of this methodology are using: (i) mechanical simple compression; (ii) isotropic steady-state free-swelling; and (iii) steady-state pressure-driven reverse-osmosis across a polymeric membrane, in combination with suitable numerical simulations in order to calibrate material parameters in the theory, notably the "free-volume"-based fluid mobility. Furthermore, we have shown that solution-diffusion theories make assumptions regarding profiles of pressure, chemical potential and concentration that are not accurate when mechanical swelling and deformation of the membrane are taken into account. However, much remains to be done. Specifically: * Our preliminary results show that our free energy of mixing might be inaccurate for the problem at hand. As mentioned in [52], a modified version of the Flory-Huggins model which reduces the entropy of swelling might be needed. * With a view towards modeling reverse-osmosis with applications towards water desalination, this theory must be complemented in order to allow multi-component diffusion through the membrane, such as water and salt molecules. 101 Part III Thin-Film-Composite Membranes: Application to Reverse Osmosis in Water Desalination 102 Chapter 11 Introduction Only about 0.5% of the 1.4 billion cubic kilometers of the planet's water is accessible fresh water, the rest being sea or brackish waters. For this simple reason, water desalination plays an increasingly important part in today's industry as at least 26 countries do not have sufficient water resources to sustain agriculture and economic development, and at least one billion people lack access to safe drinking water (cf.[33 and references therein). In 2009, the total desalination capacity was around 66.4 million cubic meter per day of fresh water, providing for over 300 million people and is expected to reach 100 million cubic meter per day by 2015. Membrane-based reverse osmosis (RO) or thermal-based multi-stage flash (MSF) and multi-effect distillation (MED) constitute over 90% of the global desalination capacity. Other technologies such as electrodialysis account for only a small fraction of the desalination capacity and is often reserved for niche application such as groundwater treatment. Although widely used in the middle eastern part of the world, MSF and MED processes require large footprints due to the necessity of on-site electricity-generating facilities. In the rest of the world, RO plants are the preferred alternative since they require nearly four times less total energy to produce a single cubic meter of desalinated water than the thermal-based plants. Here, we focus on the membrane-based RO processes and more specifically membranes. However, before jumping into the fray, it is helpful to outline the main steps in water desalination. A schematic of the infrastructure of an RO plant is shown in Fig. 11-2. Water uptake is first screened in step 1 where dirt and sediment particles larger than 5 pm are removed. The water is then treated with chemicals to remove chlorine, bad taste and odor and its pH is adjusted. The RO separation process happens in step 3, where a thin-film composite membrane is used to remove at least 95 % of total dissolved solids down to 0.0001 pm. The clean water is then transported to our home's faucets. One of the most crucial, and most expensive item in RO plants is the reverse osmosis element. Different configuration exists for RO elements (e.g. hollow-fiber module), however, the current industry standard is the spiral wound module RO element shown in Fig. 11-3. Its configuration is somewhat counterintuitive; the feed water (e.g. salt water) is pumped through the feed channel spacer, which then diffuses through the RO membrane into a permeate spacer. The permeate water (e.g. clean water) then flows through the permeate spacer in a spiral direction all the way to the center core - the permeate collector tube. This said, the quality of the separation process depends almost exclusively on the RO membrane, 103 which is a broad field by itself. In fact, many studies exist on cellulose acetate membrane and nanostructured membranes using carbon nanotube, direct-growth or interfacially embedded zeolite and many more (cf.[33], [23], [21], [24], [1]). However, the most widely used is the thin-film composite (TFC) membrane. A schematic is shown in Fig. 11-4; the membrane consists of " an ultra-thin semi-permeable salt rejection layer of ~ 0.3 prm thick fully aromatic polyamide; " a thin porous supporting layer of ~~45 pm thick poly(ether)sulfone; and * a porous non-woven base fabric ~~100 pm thick polyester. Also of importance are the operating parameters of these RO elements. Typical parameters for reverse-osmosis using TFC membranes are shown in Fig. 11-1. What is interesting to note is that industry standard limits the maximum operating pressure to approximately 8 MPa. The reasons behind this are not well known, and is thus the objective of this chapter. We tackle this problem by experimentally characterizing the mechanical properties of RO membranes as well as experimentally reproducing operating conditions of simplified reverseosmosis processes. In what follows, we first describe the experimental apparatus used in order to characterize TFC RO membranes, Section 12.1 and 12.2. Then, Section 12.3 discusses some preliminary results which open the way for further investigations. Parameter Seawater RO Brackish water RO RO permeate 12-15 (open water 12-45 (groundwater) intake) 15-17 (beach well) 5500-8000 600-3000 20% per year Every 2-5 years 5% per year Every 5-7 years 35-45 75-90 5.5-7 99.4-99.7 5.5-7 95-99 flux (W/m2 -h) Hydrostatic pressure (kPa) Membrane replacement Recovery (%) pH Salt rejection (%) Figure 11-1: Typical parameters for RO elements using TFC membranes [24]. 104 STGMR RO NEVARANE CIR= K oenuia *VW~dVIDLM reesWsmsspAt Fiur 11-2 Scemti of aue 1r F1 of rAt -in FM UT FED OUTMAW N Principte of RO Memibrane Filter FED CHAWL SPACER Figure 11-3: Schematic of a spiral wound module (SWM) RO element. Reverse Osmosis Membrane Feed water Ultra -thin Salt Rejection Laye Crosslinked Fully Aromatic Polyamide 0.3um V2 Supporting Layer Polysulfone 45um I FE -SEM Photograph of RO Membrane (UHR -FE -SEM) x Base Fabric Non-wover Fabric Polyester 1.O1umn Product Water Figure 11-4: Schematic of a TFC membrane showing the three distinct layers. 105 106 Chapter 12 Characterization of TFC membranes 12.1 Material As mentioned previously, the TFC RO membrane consists of a layer of polyamide, poly(ether) sulfone and polyester. More specifically, we carried our experiments on TY82E18-Toray membranes supplied by Sterlitech. Due to the non-conductive nature of polymer membranes, observations of the microstructure were done under an SEM after sputtering 100 nm gold particles on the surface of interest. 1 Fig. 12-1, shows an SEM picture of the cross section of the TFC membrane. The polyester (bottom) layer is clearly seen, as well as the polysulfone (middle). However, the polyamide layer is undistinguishable. A closeup view of the polysulfone shows the porous microstructure. Note that the polyester layer of the TFC membrane can be "peeled" from the poly-amide/sulfone layers. However, the poly-amide/sulfone are so thin that separation is very difficult. 12.2 Experimental Apparatus The objective is characterize the response of the TFC RO membranes during operating conditions. We propose to do so through the following experiments: (a) steady-state flux measurements using a reverse-osmosis apparatus; and (b) mechanical uniaxial tension experiments. Following these experiments, SEM pictures allow the observation of the microstructure, which hopefully will increase our understanding of the problem. 12.2.1 Reverse osmosis set-up In order to monitor the steady-state flux with respect to applied pressure-difference, we use a reverse osmosis device similar to what [511 used. As shown in Fig. 12-2, a pressure cell (HP4750, Sterlitech Inc.) is used. The cylinder is filled with a solution (e.g. salt water) which is forced through a TFC RO membrane located at the bottom of the cell (supported by a porous metal disk) by a feed pressure from the top. The feed pressure is monitored using a digital pressure gauge (EW-68334-25, Cole-Parmer Inc.) and is supplied through a nitrogen gas cylinder. The whole system sits on a magnetic stirring plate (H4000-H, Biomega and 'The sputtering was under using an evaporator at the CMSE facilities by Dr. Yong Zhang 107 Scilogex Inc.) which allows a stir-bar assembly located within the cell (not shown) to rotate and prevent concentration polarization of the solution on the membrane's active surface (active area is ; 15 cm 2 ). The permeate flux is then collected and dynamic measurements are made using a digital balance (ML203E, Mettler-Toledo Inc.) with 0.001 g resolution. Note that we have used two models of pressure cell. The HP4750, shown in Fig. 12-2, is rated for feed pressures up to 1000 psig, while the HP4750X (not shown) allows us to probe pressures up to 2500 psig. 12.2.2 Uniaxial tension Uniaxial tension of thin films is carried to characterize the mechanical properties of TFC membrane and of the individual layers. 2 At first, an Instron tabletop tensile machine (see Part I) was used to test the TFC. However, difficulties in specimen preparation resulted in large scatter in the results. Instead, a dynamic mechanical analysis (DMA) machine was used (DMA Q800) to carry on low-load tensile experiments, see Fig. 12-3. Small rectangle specimen of roughly 2 cm x 1 cm are needed. An interesting feature of this apparatus is that it allows the testing of specimen in a submerged environment, which proves useful especially in the field of RO membranes. 12.3 Preliminary Results 12.3.1 Reverse osmosis experiments The experimental procedure is as follows: 1. The membrane is conditioned in water at 300 psig for 2 hours as specified by the manufacturer. 2. The cell is then depressurized ("unloading"). 3. The cell is pressurized ("loading") to the desired pressure and collection of data over 5 minutes interval begins. Note that prior to collecting the data, we make sure that the pressure is stable, and that the flux is steady. Fig. 12-4 shows the flux measurements for such an experiment. The "loading" was done until a pressure of 1000 psig was reached. Then, "unloading" ensued to 100 psig followed by another "loading" to 900 psig (there was no more water left for the 1000 psig measurement). We clearly see non-linearities at around 600 psig and subsequent hysteresis upon "unloading". However, when reloading, there is almost a linear relationship between the flux and pressuredifference all the way up to 1000 psig. In Fig. 12-5, a similar experiment was carried, however, loading and unloading were limited to 400 psig. Astoundingly, we see close to no hysteresis. Furthermore, the loading-unloading path coincides with the initial loading path mentioned earlier, the polyester support layer can be peeled of the poly-amide/sulfone layers. However, the polyamide cannot be peeled from the polysulfone layer is they are too thin. 2As 108 of the previous experiment. This suggests that something is definitely happening at higher pressures. A similar experiment was carried on the poly-amide/sulfone layers only (the polyester layer was peeled off). The results are shown in Fig. 12-6. Interestingly, at about 300 psig, the flux-pressure-difference response becomes non-linear and the flux increases rapidly with increasing pressure-difference. Moreover, it is noted that at 600 psig, the poly-amide/sulfone layers "failed"; the flux becomes a steady flow and a reduction in pressure difference does not decrease the flow (permanent holes in the membrane). Fig. 12-7 shows the "failed" poly-amide/sulfone layer. Upon subsequent inspection of the membrane, a texture is clearly visible; it is though that the membrane is forced through the pores of the metal support disk (e.g. blowing a balloon) and permanent deformation ensues until failure at a "critical pressure". SEM pictures were taken of both the permeate side of the membrane and the metal porous disk and are shown in Fig. 12-8. In light of these pictures, it might be possible that the membrane is "extruded" through the metal porous disk. 12.3.2 Uniaxial tension experiment Tensile experiments were done on multiple 1 cm x 2 cm samples of the poly-amide/sulfone layer only. Note that in the peeling of the polyester, we may have damaged some samples, which would explain the scatter in the data shown in Fig. 12-9. However, the response seem to be of an elastic-plastic-type with initial yield strength of approximately a, = 7 MPa. Recall that in our steady-flux experiments, unloading after reaching 1000 psig ;:: 7 MPa led to large hysteresis or "permanent plastic deformation". 109 Figure 12-1: Cross-section of a TFC membrane (Top). The polyamide layer is almost undistinguishable from the porous polysulfone layer. Do not confuse the black layer (top) for polyamide. The cutting was done using a sharp knife and might have smeared the surface of the polysulfone. A close-up view of the polysulfone porous microstructure (Bottom). 110 I feed pressure porous stainless steel support disk (15 cm 2 active area) Figure 12-2: Experimental apparatus for measuring steady-state flux as function of applied pressure-difference. Figure 12-3: Dynamic Mechanical Analysis (DMA) apparatus for low-load tensile testing of thin-film. 111 30 25 20 15 CO) 10 CO~ 5 0 200 400 600 800 1000 Ap (Psig) Figure 12-4: "Loading" and "unloading" up to 1000 psig. The membrane is initially conditioned at 300 psig for 2 hours. Each point correspond to a collection of data over 5 minutes to insure steady-state. 3 0O .... Conditioned at 300 PSI 25- 20 15 U) t.- 10 a) -4- 5 01 0 200 400 600 800 1000 Ap (Psig) Figure 12-5: "Loading" and "unloading" up to 400 psig (bold line). The membrane is initially conditioned at 300 psig for 2 hours. Each point correspond to a collection of data over 5 minutes to insure steady-state. The data of Fig. 12-4 is also shown (Dotted line) 112 30 25- S20 4S,15 ~105 01 0 200 400 600 800 1000 Ap (Psig) Figure 12-6: "Loading" and "unloading" up to 400 psig (bold line) of the poly-amide/sulfone layers only. The layer is initially conditioned at 300 psig for 2 hours. Each point correspond to a collection of data over 5 minutes to insure steady-state. The data of Fig. 12-4 is also shown (Dotted line). Note that the poly-amide/sulfone layer fails at about 600 psig. porous metal support thin poly-amide/sulfone layer Figure 12-7: Poly-amide/sulfone layer after "failure" (right) . A texture is clearly visible. A schematic of what is though to happen is shown (left). 113 30 um Figure 12-8: SEM of Poly-amide/sulfone layer after "failure" (Top). We can see small holes. An SEM of the metal porous disk also shows how the membrane is "extruded" through it (Bottom). 114 12 . . . .... 10 - 'd . 8 -- 6bfl 0 10 20 30 40 Engineering Strains (%) Figure 12-9: Engineering stress vs. engineering strains 115 (%) curves for 6 different specimens. 116 Chapter 13 Concluding remarks In this Section, we have begun characterizing the thin-film composite membranes widely used in water desalination using RO processes. However, these results are recent, and few other data is available in the literature. A lot of work still needs to be done, namely (i) further investigation of possible "compaction" of the porous microstructure at a "critical pressure"; (ii) effect of adding NaCl in the feed solution on the quality of the separation process; and (iii) development of a theoretical framework for the coupled diffusion of multiple components (i.e. salt and water molecules) coupled to elastic-plastic deformation for TFC membranes, as suggested by these preliminary experiments. 117 Part IV Appendices related to Part I 118 Appendix A Existing approaches to investigating thermal barrier coatings A.1 On Existing Experimental Techniques A variety of mechanical testing techniques have been used to characterize the interfacial properties of TBC systems, with most of the experiments being carried out at room temperature. These techniques can be grouped into four major categories: i) Mode-I dominant experiments such as tension and budding experiments; ii) Indentation methods; iii) Mode-II dominant methods such as direct shear, barb pull-out, and push-out techniques; and iv) Mixed-mode - a combination of mode-I and mode-II - methods such as asymmetric fourpoint bending. In what follows, we briefly review the relevant existing techniques in each of these four main categories. A.1.1 Mode-I dominant methods One of the most widely applied methods for characterizing interfacial strength of TBC in a mode-I set-up is the tension test following the ASTM standard C633-79 (c.f., [31]). In this method, the TBC specimen is glued between two extension bars and subjected to tensile loading until failure of the top-coat-bond-coat interface. The interfacial strength can then be calculated from the ultimate loading force and the area of the fracture surface. One difficulty in applying this technique to coatings having hight strength, is that failure occurs in the adhesion layer between the specimen and the extension bars. In order to to circumvent this problem, [59] introduced a pre-crack at the interface and derived an asymptotic analytical formula to evaluate the interfacial toughness. However, extreme care must be used when interpreting this data since observations (c.f., [471) have shown that this approach can lead to a fracture path entirely within the top-coat as shown in Fig. A.2(a), thus giving a measure of the ceramic top-coat toughness, rather than an interfacial property. Alternatively, [691 used a buckling technique to measure the mode-I interfacial toughness. They monitored the budding of the coating while subjecting the TBC to in-plane compression which they used to estimate the interfacial fracture properties. However, as noted by the authors, difficulties with this technique can lead to repeatability issues in the experimental results. 119 A.1.2 Indentation methods Indentation techniques have also been used for determining interfacial properties of TBC systems. [65] performed Rockwell indentation tests on the top-coat of as-sprayed and isothermally-exposed TBC specimens. Unexpectedly, they observed that indentation of the heat-treated samples resulted in less delamination area than as-sprayed samples, suggesting an increase in fracture toughness for coatings initially subjected to elevated temperatures. Further, image analysis of the cross-section after failure showed that in both cases, fracture induced by indentation occurred in the top-coat and not at the interface. This suggests that the experiment does not represent the likely failure path expected during normal operating conditions. [631 indented the cross-section of a TBC specimen directly at the top-coat-bond-coat interface. They found that although the strength of the ceramic top-coat increased with thermal cycling due to sintering, the interfacial strength is comparable to that of an assprayed TBC specimen. [58] conducted similar cross-sectional indentation experiments and found the interface fracture toughness to be dependent on thermal cycling and showed an apparent increase within the first few cycles. However, they observed that the top-coat serves as a dominant energy release mechanism in TBCs, leading to potential large overestimation of the interface fracture strength when measured via indentation since more often than not, cracks usually appeared in the top-coat before than at the interface (Fig. A.2(b)). To our knowledge, all indentation techniques aimed at determining interfacial properties are constrained to measuring properties within a small area and might thus not be suitable for representing the global behavior of the material 1. In addition, one major shortcoming of these methods is that failure of the interface tends to occur simultaneously with failure of the top-coat, thus producing experimental results which do not only reflect the interfacial properties A.1.3 Mode-II dominant methods Various methods have been developed in order to measure the shear properties of top-coatbond-coat interface. [39] and [37] did push-out tests where a TBC specimen is forced to pass through a narrow channel where the ceramic top-coat is sheared off. A schematic of the method is shown in Fig. A.2(a). As observed in mode-I experiments, the measured delamination toughness was found to significantly decrease as the growth of the TGO layer progressed. However, these methods require highly refined specimen preparation and buckling of the top-coat can have a significant influence on the results. These shortcomings limit their application in toughness characterization experiments. Alternatively, [62] developed a direct shear test in which a 1mm x 1mm top-coat island was prepared on top a metal substrate and subsequently sheared off by a steel tool blade. The experimental results where then used in a finite element analysis to determine the 'This general shortcoming of indentation techniques is usually overcome by automatically performing a grid of test over a relatively large area of the material. In the case of indentation experiments aimed at probing material properties near an interface, the location of the indenter is crucial and thus would be extremely difficult to automate. This limits the number of indentations that can be performed within a reasonable time frame, and thus limits the area whose properties can be probed. 120 fracture energy and mode-II stress intensity factor. Although still requiring refined specimen preparation, this method does not have the shortcoming of potential buckling of the topcoat. Thus, it provides a relatively simple and direct method of obtaining mode-II interface properties. A.1.4 Mixed-mode methods Modified bending tests are become increasingly popular since they allow the characterization of TBC interface properties under mixed-mode loading. [18 measured in-situ interfacial crack initiation and subsequent propagation by loading simply supported beams using steel micro-tips. Initially, single-center-point loading is used to induce a vertical crack that extends through the ceramic top-coat to the interface. Then, the single-point load is horizontally translated causing the crack to propagate along the top-coat-bond-coat interface. This approach forces the TBC to crack in a controlled fashion, thus avoiding the appearance of segmented cracks. [30] and [681 did modified four-point bending experiments where a metal stiffener was first glued on the top of the ceramic top-coat. A notch was then introduced by cutting through the stiffener, to the interface as illustrated in Fig. A.2(b). Although this modified method successfully suppresses the formation of segmented cracks, we have found through experimentation with this technique, that sample preparation, especially the cutting of a notch to a very precise depth, is extremely difficult. Inaccuracies in the notch depth may damage the interface and may, upon subsequent loading, cause cracks perpendicular to the interface to initiation before delamination occurs. [55], [70] and others have used symmetric four-point bending configurations. In such experiments, due to the inherent variability of the properties of the interface and the specimen loading, they found it difficult to get the delamination to initiate and propagate in a completely symmetric manner In a similar fashion, [67] glued two TBC specimens onto a steel stiffener to form an asymmetric beam. Due to the asymmetry, the delamination crack naturally proceeded towards one end of the beam, rather than asymmetrically as has been observed by [64], and no prenotching is required. This method further benefits from having the metal stiffener act as an energy storing mechanism, which aids in driving the delamination once crack initiation starts. This method allows for simpler sample preparation as well as finer control of the experiments since the metal stiffener can be tailored, both in its geometry as well as in its material properties, to complement the machine which will be used in testing the sample. A.2 On Modeling TBC Failure The development of suitable constitutive theories for modeling failure of TBC systems, and a robust numerical implementation of said theories, is a necessity in future development and deployment of TBC systems in practice. Important efforts at modeling TBC system failure include the works of [6], [5], [281 and [29] and references therein 2. In these works, 2 Note that some of these papers focus on EBPVD sprayed coatings; however, the methodology discussed therein equally applies for plasma-sprayed coatings. 121 the authors investigate failure of TBC systems by considering a representative unit-cell of the bond-coat-top coat interface which can capture a single surface imperfection. A typical surface imperfection in a plasma-sprayed coating is in the order of 10 pm, thus a typical unit cell considered by these authors consists of a similar amount of interfacial area. When modeling the degradation and failure of a macroscopic part, i.e. an entire turbine blade, such simulations quicIdy become untractable due to the computational expense required to resolve the surface roughness of the interface over a large body. Even when the interface roughness is not resolved in detail, modeling the oxidation of the bond coat requires resolving the oxide layer which is also in order of 1 pm (e.g. [41]). Thus, we require a theoretical and computational technique which can be suitably applied to model degradation and failure of a TBC system at a the macroscopic length scale such as that of a turbine blade. To this end, we adopt a description of the TBC system in which we do not explicitly model oxidation; rather, we assume that oxidation manifests itself only indirectly by changing the resistance of the TBC system to spallation. With such an approach, the computational requirements of resolving the oxidation layer are relaxed, and simulations of macroscopic objects with larger dimensions become tractable. 122 TOOost (a) 600pm (b) Figure A-1: (a) cross-section of TBC specimen after the tension test (adapted from [47]);(b) crack induced by cross-sectional indentation (adapted from [58]) 123 Load Pushout blook TSC Support block Substrate speciffmSupport btck* (a) P/2 ~0X4 Notch P/2 Stffkaer h=~--(2 W -2.7 __Sabstrate 4 + bond cost L1-20 50 b-~-3.3 B Unit: nun (b) Figure A-2: (a) schematic of a push-out test (adapted from [40]);(b) schematic of symmetric four-point bending sample and experiment (adapted from [70]) 124 Appendix B Numerical implementation of traction-separation law B.1 Time Integration Procedure A fully-implicit time-integration procedure for the interface constitutive model presented in Section 3.3.2 is detailed in this section. Let [0, T] be the time interval of interest. We assume that at time tn. C [0, T1 we are given {5, D., sin}, as well as 6b and Ab, (B.1) where A6 is a displacement jump increment which occurs over a time interval At, the timeintegration procedure in this displacement driven problem is then to calculate , si,n 1}, {t+1,6+ 1, D at time tag = t + At. (B.2) For later use, the deformation resistance is integrated as (B.3) 8 S,n+1 = si,o(l - Dn+1), where the damage parameter at the end of the time increment is given by Sp - Per and where the relative equivalent plastic displacement is = ( )+ a(6 i )2. (B.5) The flow rule, reiterated here from (4.21) 5f =MmN +6TmT. 125 (B.6) is integrated using an implicit Euler scheme bn+ 1 =5 = 6pn+ AbP(B7 6 +In+At 6.P P(B.7) where A8P is the plastic increment in the displacement jump. Then, using the constitutive equation for the tractions (4.14) we may write tn+1 = K [4n+1 - 45+11 = K [6n - 6P + Ab - A ] (B.8) = K [8 e'trial - AP] where we have defined a trial elastic displacement jump 6e'tr" = quantity at the start of the integration procedure. Further, let 6 n+1 t&a = K6t" - 6b which is a known (B.9) be the trial traction, such that (B.8) may be written as tn+1 = tt'al - K [A6P]. (B.10) During plastic flow, according to the yield condition (4.21)3, we have that the yield functions (4.19) must be equal to zero, that is tN,n+1 - ItT,n+1 + A(-tNn+1) - N,n+1 = 0(B.11) Tn+1 = 0. In this particular implementation, for simplicity, we set the constitutive friction coefficient p = 0. 1 Note that in writing (B.11), we have used the fact that for a two-dimensional deformation problem ;i = ItT,n+1 IFinally, using equations (B.3), (B.8) and (B.9) in (B.11), we obtain the stress update equations t*a tt' - KNA N - SN,O (1 - Dn+1 ) = 0, - KTAbr - ST,o (1 - Dn+1 ) = 0. where the damage Dn+1 , which is also a function of the plastic displacement jump increments, is given by (B.4). We also define trial values for the yield functions N cjgiaL SNO, (B.13) It|alI _ STO. LIt is worth noting that when cohesive elements are used to model the interface between two surfaces, one may use particular surface interactions in order to add a friction to the tangential sliding behavior of those surfaces. Such surface interactions are built into Abaqus/Standard. 126 such that we may determine if a mechanism is deforming plastically based on the following conditions, if I,"a < 0 then we have an elastic step, if V' ' > 0 then we have a plastic step. Depending on which mechanism or mechanisms are deforming plastically during the time integration step, we will have different procedures for solving the stress update equations (B.12) for the plastic displacement jumps A6P. The different possible cases, and how to solve these, are outline in Sections B.1.1 through B.1.3. Once the plastic displacement jump increments A6PN and AbT. have been found, we need to update all other necessary variables in the theory. First, according to (B.7) and (B.5), the plastic displacement jumps and the equivalent plastic displacement are updated as N,n+1 - N,n =P~~ - P~ + NI and S N 2 +2. (B.15) << (B.16) + A41, The damage variable, through (B.4), is simply computed as f - , Dn+ = ?+ if S cr < on SPI. Finally, the deformation resistance is updated as SNn 1 = SN,0(1 - Dng), (B.17) sT,n+l = ST,0(1 - Day), and the tractions are updated through tN,n+1 =N''' - KNAJPN2 (B.18) tTn+1 = t;ial - KTAP sign(tr'a), Solving for Ab1 . Case 1: V"" > 0 and <D" < 0 B.1.1 In this case, only the normal mechanism is deforming plastically, the tangential mechanism is elastic, and AbP = 0. Substituting (B.4) into (B.12)1 yields N- KNAbN - N,O + SN,O (+ ( fcr 127 +- (B.19) which must be solved for APN. Let us define a, = pe- S!, a2 = t - a4 = 1 - N,O - SN,a KN 2 a1 al-K 2, (B.20) SN,O a,5 = 2 N,n a6 = a3 + 6 1 2 a ~~ 2 NS SN,O such that (B.19) maybe written as a quadratic equation a4 ANi 2 (B.21) + a5 A6N +a 6 = 0. and the normal plastic displacement jump increment A3gN is simply found through the roots of (B.21) as (B.22) - 4 a4 a6 /aj 2a A6P, = a5 4 In order for the plastic displacement jump increment computed through (B.22) to be a valid solution to the problem, it must satisfy two conditions, namely APN -a,(a > 01(B.23) KNA6) > 0- 2 - The first condition simply states that the plastic displacement jump increment must be positive, as required in the theory. The second condition is equivalent to the requirement that the damage variable D,+ 1 > 0. If both conditions (B.23) are satisfied then the solution (B.22) is accepted. If the second condition (B.29) 2 is violated, it means that there is no damage during this step, in which case the normal displacement jump is simply computed from ttN =bN = 128 SN ~N, KN (B.24) B.1.2 > 0 and <bijal < 0 Solving for A6P. Case 2: <bia' In this case, only the tangential mechanism is deforming plastically, the normal mechanism is elastic, and APN = 0. Substituting (B.4) into (B.12)2 yields + A6P) 2 +a(6~n +- tTI I - KTA6T - sTo + sTO f .kN (B.25) er which must be solved for APT. Let us define a1 = 65 - SP5 a 2 =|ttial | s ,O - - 8 sO a1 ,2 + a 6P 2 a3=6n )2 a 4 = a - (a azi = 2 (B.26) ST'O =2 (a~ +KTa2 a~) ap4n + S2 TIO a6 = a3 - (a) such that (B.25) may be written as a quadratic equation a4 Ag 2 + a6 A6P + a6 = 0. (B.27) and the tangential plastic displacement jump increment A6P is simply found through the roots of (B.27) as A6; = a,5 + k Va- 2a 4 4 a4 a 6 (B.28) In order for the plastic displacement jump increment computed through (B.28) to be a valid solution to the problem, it must satisfy two conditions, namely AbT> 01 (B.29) -a,(a2 - KTAFT) > 0. The first condition simply states that the plastic displacement jump increment must be positive, as required in the theory. The second condition is equivalent to the requirement that the damage variable D,+ 1 > 0. If both conditions (B.29) are satisfied then the solution (B.28) is accepted as a valid solution. If the second condition (B.29) 2 is violated, it means that there is no damage during this step, in which case the tangential displacement jump is 129 simply computed from ItiIST'o T KT Ab B.1.3 (B.30) Solving for AP. Case 3: <bial > 0 and <1%7ia > 0 In this case, both the normal and the tangential mechanisms are deforming plastically. We begin by rewrite (B.12) as tNi' - SN,0 KNAN + Dn+1 = 0, sN,O It I SN,O sTo - (B.31) 0 KTA+D _ ,+ Dn+1 = 0, ST,O STO which must be solved simultaneously for APN and AP. Let us define ttNial 8 - N,O SN,O a2 = ItTP' I (B.32) 8 - TO STO such that (B.31) may be written as KNA6 N a1 + SN,O a2 Dn+1 = 0, (B.33) + KTA 5 T,O Dn+ 1 = 0. Then, subtracting (B.33) 2 from (B.33) 1 yields a- - N AP -a SN,O 2 + -AbP. = 0. (B.34) ST,O Further, let a3 = as (a1 - a2)sN,o =KN KTsN,O a4 = KNO (B.35) KN ST,0' such that, (B.34) may be written as A oN= a3 + a4 A5T. 130 (B.36) Substituting (B.36) into (B.31) 2 and rearranging yields -a. (a 2 - - - (3+, + (a3 + a4 A6T)) 2 + a(6 + AbP) 2 (B.37) Now let us define a6 =a 2 -~ a5 a 7 = a+ a 4 _ (STaO (B.38) a8 = 2a6,, + 2a4(bNn + a 3 ) + 2a6 a K (6i,. + a3 ) 2 - (asa6 )2 . ag = a 6in2+ such that (B.37) may be written as a quadratic equation a7 A6o 2 + a8 A6PT + a9 = 0, (B.39) and the tangential plastic displacement jump increment ART is simply found through the roots of (B.39) as a8 L Va--- 4 a7 ag. 2a7 The normal plastic displacement jump ARN (B.40) is then solved using (B.36). In order for the plastic displacement jump increments computed to be a valid solution to the problem, they must satisfy the following conditions APN > 0, (B.41) -a5(a3 - a4 KTAb6) > 0. The first two conditions simply state that the plastic displacement jump increments must be positive, as required in the theory. The third condition is equivalent to the requirement that the damage variable D., 1 > 0. If the third condition (B.29) 3 is violated, it means that there is no damage during this step, in which case the normal and tangential displacement jumps are simply computed from N = ~l- SNO KN A6" 1T< - ST O T 7 KT 131 (B.42) B.2 Computing the Element-Level Stiffness The time-integration procedure discussed in Section B.1 gives an algorithmic incremental constitutive function for the traction t,+ 1 at the end of the step of the form t,+1 = tn+1 (Pni 6,p bn+). or equivalently tn+1 = tn+l(iQe,trd) (B.43) Within an increment, the internal variables (Pn, 6%) appearing in the traction argument are fixed, and in globally implicit finite element programs only the guess for the total strain bn+i (associated with the guess for the displacement field un+1) changes during a global NewtonRaphson equilibrium iteration. The incremental form of the term A (For simplicity, we will henceforth refer to 6 as being the displacement jump at the integration point) is the consistent tangent modulus or Jacobian required for such equilibrium and is given by D Equivalently, since .5e'' - Oin+1(, 6. Obn+1 (B.44) 6n+1) = 6n+1 - 6P, we also have that D = (B.45) 8af+ 1( (P,6ety.sW) ar 6 etrial For the specific two-dimensional cohesive elements with only displacement degrees of freedom presented in the previous section, the consistent tangent is given by DN [n+l DTT D ONT,n+1 DTN 5 tr OTN,n+1 (B.4i 6N jDNT DNN tr-al Let us recall our incremental relationships for the yield condition KNAN - It 1a I - - sN,O (1 KTA6TP - sTo - Dn+1 ) =0,(B.47) (1 - Dn+l) = 0, as well as the constitutive update for the traction t N,n+1 = KN tTn+1 = KT -tria 654* I+ KN r6t ] A - (B.48) LX661 Substituting the incremental expression for the damage (B.4) into (B.47) , yields t*|"' - KNA6N |t*"a I- KTA6T T sN,O - - + s (B A sTo + 132 - 0, sT'(B - 6 ) =0, A (B.49) where +A6 )2 ( = (B.50) . B First, we will compute the DNN term in eqn. (B.46). Taking the derivative of eqn. (B.48)1 with respect to 6g"' yields N 1 = KN (B.51) . 'id Next, differentiating (B.48) 1 with respect to J,tar yields the DNT term =N,+1 = -KN AN (B.52) c9ttraial Similarly, differentiating (B.48) 2 with respect to Ktr" and ff:tr'al gives, respectfully, the terms DTN and DT &tT,n+1 ( (B.53) -KT - Offirial (B.54) abetrial We take the derivative of (B.49) 1 with respect to 5 . We now need to compute the partial derivatives involving the plastic increments appearing in eqns. (B.51), (B.52), (B.53) and (B.54). After some manipulations we get D OA6T KN (A3k (B.55) where D SNOQ5 ,n + APN) AB - C=KN -SN,o(bTn + AbT) AB 133 (B.56) Next, we take the derivative of (B.49)1 with respect to rial (B.57) C Betrial. are trial Similarly, we take the derivative of (B.49) 2 with respect to KTi(tri) OA6; aj~e: tria F +ARN sgnt = 6e*''". + T This leads to (B.58) ,tr 1 where asTo(3 ,.n + APT) E = K- AB F -T'O(b"n (B.59) + AbNP) AB Then, taking the derivative of (B.49) 2 with respect to tra'a yields F 3O"bP ~9Eria (B.60) Since equation (B.55) and (B.60) are linearly coupled, we can substitute (B.60) into (B.55) and straightforwardly solve them. DFy-1 CE]' KN (i 8A6__ r C a- (B.61) DF\~1 FKN CE aA6T CE Similarly, substituting eqn. (B.57) and (B.58) and solving leads to OA6Tp Bage~a KT KE ( / IF)'sntti E DF 7. E1 sign(tr), (B.62) DaA 6 DKT E ( DF\ ~raL-CE -1.,. sign(t ). Hence, eqns. (B.51), (B.52), (B.53) and (B.54), along with (B.61) and (B.62) allows us to fully characterize the consistent tangent D given by eqn. (B.46) and consequently compute the element-level stiffness. 134 In summary, the components of the consistent tangent are as follows KN Otz~n+1 84tial KN CE (1 ), (B.63) DF -KPFKN( 1 KJCE (~CE] =KT (1 DFV KT I - CE CEj E sign(t'"). . 0O DF)-~ra sign(t CE DKT &tN,n+l atT,n+1 CEE C (Ket*il 967:-__ DFV' 1 ' &N,n+( It is worth mentioning that in the case of an elastic step for both normal and tangential mechanism, the consistent tangent simplifies to the elastic tangent D =.T (B.64) 0 KN On the Failure of an Interface B.3 At the end of the time-integration procedure, one must check whether the plastic displacement jump increments Ab obtained during this step causes the cohesive element to fail. In order to check this we compare the equivalent plastic displacement jump at the end of the with the failure displacement jump measure Ff. In summary increment 64 if Sp,1 if S + <S ;> then, the cohesive element has failed completely, then, compute the jacobian and step forward in time. If any integration point within the cohesive element has failed, according to the criteria (B.65), we set the traction t and the jacobian matrix D for all integrationpoints within the cohesive element to zero D = 0 0 . (B.66) Then, according to eqns. (4.31) and (4.33), fu = 0 and Ku = 0. It is worth mentioning that there are multiple ways to fail a cohesive element. For instance, consider a cohesive element with two integration points. Then, one can choose to fail the whole element solely based on whether or not a single integration point has reached the failure value. Depending on the geometry and the interpolation functions, this may increase instability along the crack. Alternatively, one can decide to fail each integration point individually. For instance, a cohesive element with two integration points submitted to a peel test might have an integration point that has failed, while the other has not. Hence, it is clear that for a refined 135 enough mesh along the cohesive zone, that is for cohesive elements whose width is small enough, those two approaches for the treatment of failure should lead to identical results. In our numerical implementation, we have opted for the former method, which consists in verifying whether or not a single integration point has reached the failure value. B.4 Numerical Implementation in a UEL **** ** ***** *** * **** **** *** * In this section, we will discuss the main aspects of writing a user-element subroutine for implementing the previously discussed constitutive model for cohesive elements. Note that this procedure is general in a sense that almost any constitutive model could have been chosen. A complete user-element subroutine for two-dimensional cohesive elements using the interface constitutive model presented earlier is given. The user-element subroutine for cohesive elements is mainly divided as follows * ******** ** ***** **** **** ** * Dummy Mesh Output Section ** *** **** *** ** * ****** ***** * -module global -subroutine UVARM ** ** ******** **** ** ** ****** * Main UEL subroutine Section -Initialization -Begin loop over integration point -do time-integration procedure -update state-variables -compute element-level residuals and stiffness -End loop over integration point -Assembling the global residual and stiffness -Defining the global variables to be sent to the UVARM subroutine * *** ** *** ** ***** ** *** ********* ***** **** ** *** * Secondary subroutine for calculations Section -Any subroutines used in the calculations Here, 1) Dummy Mesh Output The first part of the UEL is used to * Define global variables that will be shared among the different subroutines within the script. The line parameter(numElem= , nIntPt= , nShift=) discussed in Section B.5 must be modified. * Store the desired variables into uvar which can be visualized on the dummy mesh. This is done through the UVARM subroutine and is called for every user-elements. 136 2) Initialization The second part of the UEL is where we * Define all the variables in the script. & Define the elements to be used along with their shape functions and integration points. e Read the material properties. 9 Read the state variables. * Read the nodal quantities for the element and compute, for instance, the displacement jumps. 3) Integration point calculation For each integration point, we compute the time-integration procedure for the constitutive model. * Computations are done for each integration point of the user-element using Gauss quadratures. * State variables are updated and saved for the next increment. * We check if criteria for failure are met. * Element-level residuals and stiffness are defined. 4) Assembling the global residual and stiffness In this part, we assemble the global residual(RHS) and stiffness(AMATRX) for the Newton-Raphson minimization algorithm of Abaqus/Standard. We also define what variables to be sent to the UVARM subroutine for visualization. e Computes RHS and AMATRX. e Defines the global variables to be sent to the UVARM subroutine for visualization. 5) Secondary subroutines This last section of the script includes all the necessary subroutines created by the user that are used throughout the main script. For instance, the integration procedure is computed in a subroutine called INTEG. B.5 Guide to Creating an Input File for Use with a User-Element Subroutine This section explains how to create an input file for use with a user-element subroutine (UEL) for cohesive elements in Abaqus/Standard. Emphasis is put on the specific requirements regarding cohesive element modeling, and the steps for creating an input file in Abaqus/CAE is not discussed in detail. 137 Creating the cohesive elements in Abaqus/CAE * Draw the part with a finite thickness layer of cohesive element at the interface. We will further discuss this in Step 5). * Insert part into Assembly e Mesh the cohesive elements (COH2D4) using quadrilateraland sweep making sure the layer is only one element thick. * Mesh the rest of the model and create the necessary element and node sets. In particular, create two distinct node sets for both the top and bottom nodes of the cohesive elements. * Right click on your Model name => Edit Attributes. - Check the box that reads "Do not use parts and assemblies in input files". e Create job 9 Right click on your Job name => Write Input. The input file will be written to your Abaqus working directory. Modifications of input file 1) Adding the User Elements. Before *Element, type=COH2D4 insert the following two lines: *User Element, Nodes=4, Type=U1, Iproperties=1, Properties=10, Coordinates=2, variables=18, Unsymm 1,2 In the first line: e Nodes= is the number of nodes for the user element. * Type= assigns a name to the user element. e Iproperties= is the number of Integer properties passed into the UEL. & Propeties= is the number of FloatingPoint properties passed into the UEL. * Coordinate= refers to either 2D or 3D simulation * Variables= is the total number of internal variables per element. This is equal to the number of internal variables multiplied by the number of integration points. e Unsymm specifies that the tangent matrix defined in the UEL need not be symmetric. 138 The second line determines the active degrees of freedom of the UEL. In this case, 1 and 2 refer to the x-displacement (D.O.F. 1) and the y-displacement (D.O.F. 2), respectively. Furthermore, in the line containing *Element, type=C0H2D4, change type to U1. 2) Making the Dummy Mesh. Since Abaqus 6.10 does not support user-element visualization, we create a "Dummy Mesh" for the cohesive elements. These elements are connected to the same nodes as the UEL elements but have different numberings and properties. 9 Copy *Element, type=U1 including all the data lines underneath. * Paste this after the last *Nset or *Elset that you have created. * Change the type to type=COH2D4, the Abaqus built-in cohesive elements. * Increase all element numbers by a fixed factor such that there is no repeated element numbers between the dummy mesh and any other elements in the model. Make sure to create an element set for all these dummy elements For instance, if you have 200 cohesive elements (and consequently, 200 Dummy cohesive elements) and 300 regular continuum elements in your simulation, you would increase the element number for the Dummy cohesive elements by a factor of at least 301. Choosing that number to be 1000, the result would look like *Element, type=COH2D4, elset=DummyCobElem 1001 ,1 ,61 ,96 ,4 1002 ,61 ,62 ,95 ,96 1003 ,62 ,63 ,94 ,95 1004 ,63 ,64 ,93 94 1005 64 ,65 ,92 ,93 1200 276 ,277 ,314 ,315 3) Material Properties. We now need to define the material properties for both the UEL elements as well as the dummy elements. This input should look like ****** ** ***** ** ****** *** UEL COHESIVE ** *uel property, elset=RightCohElem 10.OdO , 9.dO , 0.01dO ,0.01d0 1.75d0 .75d0 1.OdO 9 * **** *** ***** ** *** *** *** , 5.5d0 , 1.D3 , 1.D3 ** **DUMMMY COHELEM ** ** * *** ****** * ****** ** *** *Cohesive section, elset=DummyCohElem, material=adhesive, response=traction separation, thickness=specified 1.0,1.0 *Material, name=adhesive *Elastic, type=traction 1.0E-10,1.0E-10,1.0E-10 139 *User output variables 9 Some things to note: 9 In Step 1), we have specified 10 FloatingPoint properties and 1 Integer property. Notice how we are giving those material properties. * The Integer properties specifies the number of global output variables. These variables can be used for displaying different quantities via the dummy mesh, computed at the integration points of the UEL elements. That number should also appear after the *User output variables command line in the dummy mesh definition. e The properties given for the dummy cohesive elements are taken so that these elements do not affect the overall response of the system. Note that we do not specify any criteria for softening/failure. 4) Defining the Step. The UEL outputs quantities are visualized via the Dummy cohesive elements. The following command must be included in the *Step definition *Output, field *Element Output, elset=DummyCohElem uvarm Here, * The command *Output, f ield requests output that can be visualized, these are called field outputs, i.e. used in contour plots. 5) Creating zero thickness cohesive elements In Step 1), we have created cohesive elements with a finite thickness. In order to make them zero thickness, we want to project the top nodes of the cohesive elements onto the bottom nodes. This is the reason why we created node sets for the top and bottom nodes of the cohesive elements. Here is a small Matlab code that can do it. fid = fopen('MyinputFile.inp'); while feof(fid)~=1 tline = fgetl(fid); %Get all the nodes if isempty(strfind(tline,'*Node,nset=nall')) == 0; Nodes=(fscanf(fid,'Xg, %g, %g",[3 inf]))'; end %Get the nset with the nodes to be projected if isempty(strfind(tline,'*Nset=Top')) == 0 data = fscanf(fid,'%i, %i, %i, %i, Xi, Xi, Xi, %i, Xi, Xi, Xi, Xi1,[16 inf)'; data=reshape(data',1,size(data,1)*size(data,2)); data = data(data~=0); 140 Xi, %i, %i, Xi, NSETTOP = data'; end XGet the nset with the nodes to be projected ON if isempty(strfind(tline,'nset-Bottom')) = 0 data = fscanf(fid,'%i, Xi, Xi, Xi, Xi, Xi, Xi, Xi, %i, Xi, Xi, Xi, Xi, Xi, %i, Xip,[16 inf])'; data=reshape (data',1, size(data, 1)*size (data,2)); data - data(data=0); NSETBOT = data'; end end fclose(fid); %Get the nodes to project and their coordinates TopNodes = Nodes(NSETTOP,:); BotNodes = Nodes(NSETBOT,:); TopNodes(:,2) = BotNodes(:,2); TopNodes(:,3) = BotNodes(:,3); Nodes(NSET,:) = TopNodes; File = fullfile(cd,'NeuNodes.inp'); fid=fopen(File, 'w'); for i = 1:length(Nodes) fprintf(fid,'%d, %.5f, %.5f \n',Nodes(i,1), Nodes(i,2), Nodes(i,3)); end fclose(fid) Some remarks, * This will generate a file with a new list of nodes. In the input fie, replace all the data lines below *Node,nset=nall with the data lines from the generated file. Modification of the UEL subroutine The input file is now ready to be used. Before running, however, in the UEL subroutine, you must modify the line containing parameter (numElem= ,nIntPt= ,nShift= ): e numElem= equal to a number greater or equal to the total number of elements in your real mesh. e nIntPt= equal to the number of integration points per dummy element. * nShift= equal to the difference in numbering between the dummy elements and the real elements defined in Step 2). Since we have increased the element numbering by 1000 in Step 2), that number should be 1000. 141 Running the simulation The job can now be run by typing the following command in the directory containing the UEL subroutine and the input file: abaqus double inp=MyInputFile.inp user=MyUelSubroutine.for job=MyJobname B.5.1 Input file for UEL This input file creates a single cohesive element along with its dummy counterpart. The load is applied directly on the node of the cohesive element. Educational example on user-elements for cohesive zone modeling ** ** By Jacques Luk-Cyr, June 2013 ** ** *HEADING One element ** PARAMETER DEFINITION ** ** ** MODEL DEFINITION ** *Node, nset=nall 0., 1, 2, 0.100000001, 3, 0., 4, 0.100000001, 0. 0. 0. 0. ** * ***** ** **** **** ** ** ** ** ** **** * *** ***** * ** Cohesive Elements ** *User element, type=U1, node=4, coordinates=2, properties-10, Iproperties=1, variables=18, unsymm 1,2 *Element, type=Ul, elset=CohElem 1, 1, 2, 4, 3 ** ** **** *** *** *** **** * ******** Dummy Cohesive Elements ** ***** **** **** **** ** * ** ** ** *Element, type=COH2D4, elset=Dummy-CobElem 10001, 1, 2, 4, 3 ** *Nset, nset=Left 142 1,3 *Nset, nset=Right 2,4 *Nset, nset=Bottom 1,2 *Nset, nset=Top 3,4 *Nset, nset=BottomRight 2 *Nset, nset=TopRight 4 *Node 999,0.05,0.05 *Nset, nset=RefPt 999 MATERIAL DEFINITION ** ** * *** ** ***** *** * ******** ** DUMMY COHESIVE ELEMENTS ** ** *** ***** *** *** * ***** ** *Cohesive section, elset=DummyCohElem, materia]L=adhesive, response=traction separation, thickness=specified 1.0,1.0 *Material, name=adhesive *Elastic, type=traction 1.0E-3,1.0E-3,1.0E-3 *User output variables 9 ** ** ***** **** ***** *** * *** COHESIVE ELEMENTS ******** **** *** ** ** **** ** *** * ** - UEL ** *** ******** ** *uel property, elset=CohElem 10.OdO , 8.OdO , 0.01dO ,0.01dO 1.0d0 , 1.OdO , 9 , 0.25d0 , 0.30d0 ** ** BOUNDARY CONDITIONS ** *Boundary Bottom, 1,2 *Equation 2 Top,2,1.0,RefPt,2,-1.0 *Equation 2 Top,1,1.0,RefPt,1,-1.0 ** 143 , 1.E3 , 1.E3 ** STEP DEFINITION ** *Step, Name=Loadl, nlgeou=yes, inc=10000 *Static 0.00001, 1.0, 1E-8, 0.01 ** Loading *Boundary RefPt,2,2,0.35 RefPt,1,1 ** ** Output *Output, field *Node output, nset=nall u,rf *Element output ,elset=DummyCohElem uvarm ** *Output, history *Node output, nset=RefPt ul,u2,rf2, rf1 *Element output , elset=DummyCohElem uvarm *End Step ** 144 Part V Appendices related to Part II 145 146 Appendix C Numerical implementation of the coupled fluid permeation and large deformation theory C.1 Numerical Methodology We present here the numerical methodology for solving the theory presented in Chapter 8. In this numerical implementation, the force balance (8.36)and mass balance (8.38) axe treated as balance laws to be satisfied in a weak form within the body. Furthermore, we choose to solve numerically in the current configuration (cf. [11],[121). The strong form of our theory can be summarized as follows: momentum balance divT + b = 0 in B, u= u on SU, Tn =t on St, (C.1) divj iO = 0, {JQ0~2d mass balance in B, on -j- _=j on on Sj, Then with w, w, and w 2 denoting two test fields which vanish on S, and S,, respectively, the corresponding weak forms are fB(w -divT+w -b) dv =0, )(C.2) Wi ivj dv = 0. Using the identity Div(aa) = Va - a + aDiv(a), divergence theorem, and the boundary 147 conditions on St and S, we may simplify the weak form (C.2) to f(-grad f w1 j w - . da = 0, w: T + w - b) dv + + grad w1 - j dv + Jwj da = 0. The body is approximated using finite elements, B = U Be, and the trial solutions for the chemical potential and displacements are interpolated inside each element by U=EUAN A(C.4) A S= A A with the index A = 1 2, ... , M denoting the nodes of the element, pA denoting the nodal chemical potential and NA the shape functions. We employ a standard Galerkin approach, in that the test fields are interpolated by the shame shape functions, that is W= WAN, A (C.5) w1 = E wNA. A Using (C.4) and (C.5) in (C.3) yields the following element-level equations J Be (-wA - (TgradNA +bNA)) NAwA -i da= 0, d + St wNA da=0. + grad N A- j) dv + LeW (N (C.6) The system of equations (C.6) is solved using a Newton procedure. Thus, since w and w 1 are arbitrary, we define the element-level residuals for the displacement and concentration as (bNA-TgradNA) de+ R$= NAda=0, f~te j NA +gradNA-j dv+ (C.7) NAjda=O. In addition to the residuals (C.7), the following tangents are also required for the Newton 148 procedure - KAB = UiU /1KU" = ORA KAB (C.8) A KAB B KI= Using (C.7) the tangents (C.8) may be evaluated as K KB (ONA ax, =iu fBe ONA J jk)O (A 02 JS:NANB 6da, &rUk 4 dv, NB (N 4 N K B- dv XI~k)O x r -208 LO - 2 5A_ jN A BO 3da, I -dNA N NOl ^NB pA - I- a- - a) dv (C.9) N K = -Be f K AB = (+,- NA (m(/-it t5~ Ox (Xk NB OB dv O dx .)/At. Finally, Also, the term in (C.7) 2 is computed using the approximation q terms such as 4/Op appearing in (C.9) 3 are computed numerically using a finite difference scheme. To complete the evaluation of the tangents (C.9) we have Aijkl = J-'FjmFn(Aa)imkn, (C.10) where O(TR)ij (AR)ijkl O9FkI "G Fi + Goo(F-1)j(F-1)jk GoC5ik3,l + F ki + J 89K I{(F')jj(F 1 )k - (InJe)(F- l)i(F- )jk} (C.11) - with OG 4 / 9 G5F 2 2-A 149 F, (C.12) aT.j T-o 0# ao C.2 K J2 )1 >1-I d R t ++2X _ (1 - In Je) - -, - (C.13) - # and Description of the Element A linear, reduced-integration, two-dimensional, four-noded element was implemented; it is shown in Fig. C-1. The corresponding shape functions in natural coordinate Ce, ?e are given by 1 N1 = -1 -- e )(1 4 77e), + ( N2 = -('+e )(1 -77e), (C.14) Ce )(1+re), N3=j+1 N4 = (1 Ce )(1 +7). There are two types of elements: (1) Plane strain; (2) Axisymmetry. Details for these cases are given in what follows. C.2.1 Plane strain We take the 3-direction to be the out-of-plane direction. Then, in the present case, the following restrictions apply to the deformation gradient F: F3 3=1, C.2.2 F1 3 =F23 =F3 1 =F32 =0. (C.15) Axisymmetry For this case, we take the radial, axial, and circumferential directions to be the 1-direction, 2-direction and 3-direction, respectively. We again have F13 = F23 =F31 =F32 = 0, 150 (C.16) and, denoting the referential and current radial positions of a material point by R and r, respectively, for F we have that (C.17) F=R These radial positions can be calculated as R = NAX, (C.18) r = NA1, with Xj and x- the referential and current positions of node A i the 1-direction (radial), respectively. Further, we denote the referential and current 2-direction (axial) position of a material point by Z and z, respectively. With this nomenclature, volume integrals for the present case are evaluated as follows: J() dv, = f J ()2rRdRdZ, )dv = J()2,rr dr dz. 3 40 e 1 2 Figure C-1: Schematic of linear finite element with natural coordinates 151 (C.19) 152 Bibliography [1] D. Li ad H. Wang. Recent developments in reverse osmosis desalination membranes. Journalof Materials Chemistry, 2009. [2] L. Anand. A constitutive model for compressible elastomeric solids. Mechanics, 1996. Computational [3] E.M. Arruda and M.C. Boyce. A three-dimensional constitutive model for the large stretch behavior of rubber elastic materials. Journal of the Mechanics and Physics of Solids, 1993. [4] G.I. Barenblatt. The formation of equilibrium cracks during brittle fracture: general ideas and hypotheses, axially symmetric cracks. Applied Mathematics and Mechanics, 1959. [5] M. Bialas. Finite element analysis of stress distribution in thermal barrier coatings. Surface and Coatings Technology, 2008. [6] M. Caliez, J.L. Chaboche, F. Feyel, and S. Kruch. Numerical simulation of ebpvd thermal barrier coatings spallation. Acta Materialia,2003. [7] G.T. Camacho and M. Ortiz. Computational modeling of impact damage in brittle materials. Internationaljournal of solids and structures, 1996. [8] P.P. Camanho and C.G. Davila. Mixed-mode decohesion finite elements for the simulation of delamination in composite materials. NASA/TM, 2002. 211737. [9] E.P. Chan, A.P. Young, J.H. Lee, and C.M. Stafford. Swelling of ultrathin molecular layer-by-layer polyamide water desalination membranes. Journal of Polymer Science, PartB Polymer Physics, 2013. [10] C.A. Chester and L. Anand. A coupled theory of fluid permeation and large deformations for elastomeric materials. Journal of the Mechanics and Physics of Solids, 2010. [11] C.A. Chester and L. Anand. A thermo-mechanically coupled theory for fluid permeation in elastomeric materials: Application to thermally responsive gels. Journal of the Mechanics and Physics of Solids, 2011. [12] C.A. Chester, C.V. Di Leo, and L. Anand. A finite element implementation for diffusiondeformation in elastomeric gels. In writing, 2014. 153 [13] D.R. Clarke and C.G. Levi. Materials design for the next generation thermal barrier coatings. Annual Review of Materials Research, 33, 2003. 383-417. [14] M.H. Cohen and D. Turnbull. Molecular transport in liquids and glasses. The Journal of Chemical Physics, 1959. [15] M.C. Cookson. An elastic-plastic interface constitutive model for combined normal and shear loading: application to adhesively bonded joints. PhD thesis, Massachusetts Institute of Technology, 2004. [16] M. Doi. Gel dynamics. Journal of the Physical Society of Japan, 2009. [17] D.S. Dugdale. Yielding of steel sheets containing slits. Journal of the Mechanics and Physics of Solids, 1960. [18] C. Eberl, X. Wang, D.S. Gianola, T.D. Nguyen, M.Y. He, A.G. Evans, and K.J. Hemker. In situ measurement of the toughness of the interface between a thermal barrier coating and a ni alloy. Journal of the American Ceramic Society, 2011. [19] P.J. Flory. Thermodynamics of high polymer solutions. Physics, 1942. The Journal of Chemical [20] E. Fried F.P. Duda, A.C. Souza. A theory for species migration in finitly strained solid with application to polymer network swelling. Journal of the Mechanics and Physics of Solids, 2010. [21] C. Fritzmann, J. L6wenberg, T. Wintgens, and T. Melin. State of the art review reverse osmosis desalination. Desalination,2007. [22] B.P. Gearing. Constitutive equations andfailure criteriafor amorphouspolymeric solids. PhD thesis, Massachusetts Institute of Technology, 2002. [23] B.D. Freeman G.M. Geise, D.R. Paul. Fundamental of water and salt transport properties of polymeric materials. Progress in Polymer Science, 2014. [24] L.F. Greenlee, D.F. Lawler, B.D. Freeman, B. Marrot, and P. Moulin. Reverse osmosis desalination: Water sources, technology and today's challenges. Water Research, 2009. [25] S. Gudlavalleti. Mechanical testing of solid materials at the micro-scale. Master's thesis, Massachusetts Institute of Technology, 2002. [26] S. Gudlavalleti, B. Gearing, and L. Anand. machines. Experimen- tal Mechanics, 2005. Flexure-based micromechanical testing [27] A.A. Hattiangadi. Thermomechanical cohesive zone models for analysis of composites failure under thermal gradients and transients. PhD thesis, Purdue University, 2004. [28] T.S. Hille, T.J. Nijdam, A.S.J. Suiker, S. Turtletaub, and W.G. Sloof. Damage growth triggered by interface irregularities in thermal barrier coatings. Acta Materialia, 2009. 154 [29] T.S. Hille, S. Turtletaub, and A.S.J. Suiker. Oxide growth and damage evolution in thermal barrier coatings. Engineering FractureMechanics, 2011. [30] I. Hofinger, M. Oechsner, H.A. Bahr, and M.V. Swain. Modified four-point bending specimen for determining the interface fracture energy for thin brittle layer. International Journal of Fracture, 1998. [31] Q. Hongyu, Y. Xiaoguang, and W. Yamei. Interfacial fracture toughness of aps bond coat/substrate under high temperature. InternationalJournal of Fracture,2009. [321 C.Y. Hui, A.S. Jagota, S.J. Bennison, and J.D. Londono. strength of soft elastic solids. Proc. R. Soc. Lond., 2003. Crack blunting and the [33] T. Humplik, J. Lee, S.C. O'Hern, B.A. Fellman, M.A. Baig, S.F. Hassan, M.A. Atieh, F. Rahman, T. Laoui, R. Karnik, and E.N. Wang. Nanostructured materials for water desalination. Nanotechnology, 2011. [34] J.W. Hutchinson and A.G. Evans. Mechanics of materials: top-down approaches to fracture. Acta materialia, 2000. [35] R.G. Hutchinson and J.W. Hutchinson. Life-time assessment for thermal barrier coatings: Tests for measuring mixed mode delamination toughness. Journal of American Ceramic Society, 2011. [36] A. Khaled, K. Loeffel, H. Liu, and L. Anand. Modeling decohesion of a top-coat from a thermally- growing oxide in a thermal barrier coating. Surface and Coatings Technology, 2013. [37] S.S. Kim, Y.F. Liu, and Y. Kagawa. Evaluation of interfacial mechanical properties under shear loading in eb-pvd tbcs by the pushout method. Acta Materialia,2007. [38] Q. Liu, A. Robisson, Y. Lou, and Z. Suo. Kinetics of swelling under constraint. Journal of Applied Physics, 2013. [39] Y.F. Liu, Y. Kagawa, and A.G. Evans. Analysis of a barb test for measuring the mixed-mode delamination toughness of coatings. Acta Materialia, 2008. [40] K. Loeffel. On the oxidation of high-temperaturealloys, and its role in failure of thermal barriercoatings. PhD thesis, Massachusetts Institute of Technology, 2012. [41] K. Loeffel, L. Anand, and Z.M. Gasem. On modeling the oxidation of high-temperature alloys. Acta Materialia,2013. [42] H.K. Lonsdale, U. Merten, and R.L. Riley. Transport properties of cellulose acetate osmostic membranes. Journal of Applied Polymer Science, 1965. [43] L. Malaeb and G.M. Ayoub. Reverse osmosis technology for water treatment: State of the art review. Desalination,2011. 155 [44] G.H. Meier and F.S. Pettit. The oxidation behavior of intermetallic compounds. Material Science and Engineering, 1992. [45] A. Needleman. An analysis of decohesion along an imperfect interface. International journal of fracture, 1990. [46] A. Needleman and X.P. Xu. Numerical simulations of fast crack growth in brittle solids. Journal of the Mechanics and Physics of Solids, 1994. [47] M. Okazaki, S. Yamagishi, Y. Yamazaki, K. Ogawa, H. Waki, and M. Arai. Adhesion strength of ceramic top coat in thermal barrier coatings subjected to thermal cycles: Effects of thermal cycle testing method and environment. International Journal of Fatigue, 2013. [48] I. Ozdemir, W.A.M. Brekelmans, and M.G.D Geers. A thermo-mechanical cohesive zone model. ComputationalMechanics, 2010. [49] W. Pan, S.R. Philpot, C. Wan, A. Chernatynskiy, and Z. Qu. Low thermal conductivity oxides. MRS bulletin, 37(10), 2012. 917-922. [50] D.R. Paul. Reformulation of the solution-diffusion theory of reverse osmosis. Journal of Membrane Science, 2004. [51] D.R. Paul and O.M. Ebra-Lima. Pressure-induced diffusion of organic liquids through highly swollen polymer membranes. Journal of Applied Polymer Science, 1970. [52] Y. Louand A. Robisson and S. Cai andZ. Suo. Swellable elastomers under constraint. Journal of Applied Physics, 2012. [53] SIMULIA. Abaqus/Standard, Version 6.10. 2010. [54] C. Su, Y.J. Wei, and L. Anand. An elastic-plastic interface constitutive model: application to adhesive joints. InternationalJournal of Plasticity, 2004. [55] P.Y. Thery, M. Poulain, M. Dupeux, and M. Braccini. Spallation of two thermal barrier coating systems: experimental study of adhesion and energetic approach to lifetime during cyclic oxidation. Journal of Materials Science, 2009. [56] A. Turon, P.P. Camanho C.G. Divila, and J. Costa. An engineering solution for mesh size effects in the simulation of delamination using cohesive zone models. Engineering Fracture Mechanics, 2007. [57] J.S. Vrentas and C.M. Vrentas. Pressure effects in nonporous membranes. Engineering Science, 2013. Chemical [58] X. Wang, C. Wang, and A. Atkinson. Interface fracture toughness in thermal barrier coatings by cross-sectional indentation. Acta Materialia,2012. 156 [59] M. Watanabe, S. Kuroda, K. Yokoyama, T. Inoue, and Y. Gotoh. Modified tensile adhesion test for evaluation of interfacial toughness of hvof sprayed coatings. Surface and Coatings Technology, 2008. [60] Y.J. Wei and L. Anand. Grain-boundary separation and sliding: application to nanocrystalline materials. Journal of the Mechanics and Physics of Solids, 2004. [61] J.G. Wijmans and R.W. Baker. The solution-diffusion model: a review. Journal of Membrane Science, 1995. [62] Z.H. Xu, Y. Yang, P. Huang, and X. Li. Determination of interfacial properties of thermal barrier coatings by shear test and inverse finite element method. Acta Materialia, 2010. [63] Y. Yamakazi, S.I. Kuga, and M. Jayaprakash. Interfacial strength evaluation technique for thermal barrier coated components by using indentation method. Procedia Engineering, 2011. [64] Y. Yamakazi, A. Schmidt, and A. Scholz. The determination of the delamination resistance in thermal barrier coating system by four-point bending tests. Surface and Coatings Technology, 2006. [65] J. Yan, T. Leist, M. Bartsch, and A.M. Karlsson. On cracks and delaminations of thermal barrier coatings due to indentation testing: Experimental investigations. Acta Materialia, 2010. [66] H. Yasuda, C.E. Lamaze, and L.D. Ikenberry. Permeability of solutes through hydrated polymer mebranes. part i. diffusion of sodium chloride. Die Makromolekulare Chemie, 1968. [67] P.F. Zhao, X.D. Li, F.L. Shang, and C.J. Li. Interlamellar cracking of thermal barrier coatings with tgos by non-standard four-point bending tests. Materials Science and Engineering: A, 2011. [68] P.F. Zhao and F.L. Shang. Experimental study on the interfacial delamination in a thermal barrier coating system at elevated temperatures. Journalof Zheijian university, 2010. [69] X. Zhao, J. Liu, D.S. Rickerby, R.J. Jones, and P. Xiao. Evolution of interfacial toughness of a thermal barrier system with a pt-diffused gamma/gamma' bond coat. Acta materialia,2011. [70] Y. Zhao, A. Shinmi, X. Zhao, P.J. Withers, S. Van Boxel, N. Markocsan, P. Nylen, and P. Xiao. Investigation of interfacial properties of atmospheric plasma sprayed thermal barrier coatings with four- point bending and computed tomography technique. Surface and Coatings Technology, 2012. 157