JOURNAL OF THEORETICAL AND APPLIED MECHANICS 44, 3, pp. 533-552, Warsaw 2006 MICROSTRUCTURAL EFFECTS ON DAMAGE IN COMPOSITES – COMPUTATIONAL ANALYSIS Leon Mishnaevsky Jr. University of Stuttgart, IMWF and Darmstadt University of Technology, Institute of Mechanics, Germany e-mail: mishnaevsky@web.de In this paper,microstructural effects on thedamage resistance of compo- site materials are studied numerically using methods of computational mesomechanics of materials and virtual experiments. Several methods and programs for automatic generation of 3Dmicrostructuralmodels of composites based on the geometrical description of microstructures as well as on the voxel array data have been developed and tested. 3D FE (Finite Element) simulations of the deformation and damage evolution in particle reinforced composites are carried out for differentmicrostruc- tures of the composites. Some recommendations for the improvement of the damage resistance of lightweight metal matrix composites with ceramic reinforcements are obtained. Key words: damage, composite, finite elements, micromechanics, nume- rical simulations 1. Introduction The optimal design of particle-reinforced materials on the basis of computa- tional simulations of their behaviour has attracted growing interest of resear- chers over the last two decades (Mishnaevsky and Schmauder, 2001). One of the ways to determine the optimalmicrostructures is to use numerical (micro- and mesomechanical) simulations of deformation and failure processes in the materials, and to carry out the ”virtual testing” of different microstructures of materials. According to Mishnaevsky (1998), Mishnaevsky and Schmauder (2001), Mishnaevsky et al. (2003a, 2004b), a possible scheme of the optimiza- 534 L. Mishnaevsky Jr. tion/design of materials on the basis of numerical experiments should include the following steps: • Problem definition: definition of necessary properties to be improved on the basis of the analysis of service conditions and the analysis of the available means of the microstructure control; analysis of the effects of the material manufacturing and processing on the microstructures of materials (examples: effect of duration and temperature of sintering on grain sizes and contiguity of hard alloys; effect of hot working on the type of structures in high speed steel) • Analysis of realmicrostructures of consideredmaterials (digitizingmicro- graphs from cuts of materials, image analysis of the material structure, search for regularities or periodicity in the microstructures) (Mishna- evsky et al., 2003a; Wulf, 1995; Wulf et al., 1993) • Choice of the appropriate simulation approach: unit cell approach (for regularmicrostructures) or real structure simulation, cohesivemodels of fracture, element eliminationmethod, etc. (Mishnaevsky, 1998; Mishna- evsky and Schmauder, 2001; Mishnaevsky et al., 2003a, 2004b) • Experimental determination of mechanical properties, damage mecha- nisms (debonding, particle failure, etc.) and failure conditions for con- stituents of the material to be optimized (Mishnaevsky et al., 1999b, 2003b) • Development of numerical models of the material with real microstruc- tures and verification of the model by comparing the calculated and experimental results (Wulf, 1995; Wulf et al., 1993) • Virtual (computational) testing of ideal artificial microstructures (Mi- shnaevsky, 2004, 2005; et al. Mishnaevsky et al., 1999a, 2004a); opti- mization of microstructures; comparison and recommendations for the improvement of themicrostructures ofmaterials. By testing some typical idealized microstructures of a considered material in numerical experi- ments, one determines the directions of the material optimization and preferable microstructures of materials under given service conditions. Such simulations should be carried out for the same loading conditions and material as the real structure simulations, which proved to reflect adequately the material behaviour • Realization of the recommended microstructures in cooperation with industry,usingpowdermetallurgy technology, etc.; verification of results. The main subject of this work is the development of numerical tools for the computationalmesomechanical testing ofmaterials andcarryingoutnumerical Microstructural effects on damage in composites... 535 experiments, which should lead to the development of recommendations for the improvement of material structures. 2. Automatic generation of microstructural 3D models of composite materials The concept of the optimal design of materials on the basis of numerical te- sting of microstructures can be realized if big series of numerical experiments for different materials and microstructures can be carried out quickly, in a systematic way, automatically. This can be done if the labour costs of the numerical experiments, a significant part of which are the efforts of the gene- ration of micromechanical models, are kept very low. To achieve this, a series of programs was developed, which should automate the step of generation of 3D microstructural models of materials. After a 3D microstructural model of a material with a complex microstructure is generated, the numerical testing of the microstructure is carried out with the use of commercial finite element software. In this Section, we present several newly developed programs for the automatic generation of 3D microstructural models of heterogeneous mate- rials. 2.1. Programm ”Meso3D” for automatic geometry-based generation of 3D FE microstructural models To simplify and automate the generation of 3D FE microstructural mo- dels of materials, the program ”Meso3D” was developed (Mishnaevsky, 2004; Mishnaevsky et al., 2004a). The programdefines geometry, mesh parameters and boundary conditions of different multiparticle unit cell models of materials, and generates a com- mand file (session file) for the commercial FE Pre- and Post-Processing so- ftware MSC/PATRAN, which creates automatically a multiparticle unit cell model of a representative volume of a composite material. During the model generation controlled by the command file, the geometry of the cell is created as a box containing a given amount of round or ellipsoidal particles of dif- ferent sizes. Both embedded and non-embedded unit cells can be produced. Then, the designed microstructures (matrix and particles) are meshed with tetrahedral elements using the free meshing technique (Mishnaevsky, 2004). After that, themesh is automatically improved, and finally the boundary con- ditions andmaterial properties are defined.Themodel can be further changed 536 L. Mishnaevsky Jr. or run at different commercial or non-commercial FEprograms (as ABAQUS, NASTRAN, etc.). The geometry and parameters of the unit cell are defined during a short interactive session, in which the parameters are introduced into the program either directly or bymultiple choice. Themicrostructures to be generated are defined by the sizes of the considered cell, shape, volume content and amount of inclusions, kindof the inclusiondistribution (random,pre-defined, clustered, graded, etc.), probability distribution of the inclusion sizes, etc. Themodel is definedby the fineness of themeshing, availability or non-availability and sizes of embedding, boundary conditions (uniaxial tension or triaxial loading). Due to the fact that the models are geometry-based, only rather simple shapes of the inclusions (round and ellipsoidal) can be taken into account in thismodel. The radii, form and positions of the inclusions can be read from the input text file (for the cases of pre-defined or regular particle arrangements), or generated with the use of a random number generator. In the second case, there are options of the random, clustered, gradient arrangements or dense packing of particles. FE models of both artificial and real microstructures can be generated with this program. In the case of real microstructures, either experimentally determined coordinates and radii of inclusions are given in the input file, or experimentally determined probability distributions of these values can be used to generate quasi-real microstructures. For generation of different artificial microstructures and particle arran- gements as well as for statistical analysis of the generated microstructures, several subroutines were used. In the case of generation of random particles arrangement using a uniform random number generator, each coordinate is produced independently, with another randomnumber seed. After the coordi- nates of the first particle are defined, the coordinates of each new particle are determined both by using the random number generator and from the condi- tion that the distance between the new particle and all available particles is no less than 0.1 of the given particle radius. If the condition is not met, the seed of the random number generator is changed, and the coordinates of the new particle are determined anew. In order to avoid the boundary effects, the distance between a particle and the borders of the box is set to be no less than 0.05 of the particle radius. In order to generate localized particle arrangements, like clustered, layered and gradient particle arrangements, the coordinates of particle centers were calculated as random values distributed by the Gauss law. The mean values of the corresponding normal distribution of the coordinates of particle centers Microstructural effects on damage in composites... 537 were assumed to be the coordinates of a center of a cluster (for the cluste- red structure), or the Y - or Z-coordinate of the border of the box (for the gradient microstructures). The standard deviations of the distribution can be varied, which allows one to generate different particle arrangements, from hi- ghly clustered or highly gradient arrangements (very small deviation) to fast uniformly randomparticle arrangements (a deviation comparablewith thebox size). Another procedure is used to create multiparticle unit cells with a high volume content of particles. In this case, a dense packing algorithm is used. First, the average distance between particle centers is determined from the required volume content of particles and their amount. Then, the unit cell is filled by the particles ”layer after layer”. With this procedure, the volume content of particles of about 40% can be achieved. Fig. 1. Schema of the program”Meso3D” and examples of 3Dmicrostrucuralmodels of representative volumes ofmaterials generatedwith themodel (Mishnaevsky, 2004) 2.2. Program ”Voxel2FEM” for the voxel array based generation of 3D microstructural models The geometry-based approach to the description of inclusion shapes, reali- zed in theprogram”Meso3D”,workswell only for relatively simplegeometrical forms of microstructural elements in a composite (like round or ellipsoidal in- clusions). This can be considered as a general drawback of many methods of 3D modelling of microstructures: both shapes of inclusions and their spatial distribution are often oversimplified. To automate the generation and meshing of 3D FE Models of materials with complexmicrostructures, theprogram”Voxel2FEM”,whichallows one to avoid this drawback,was developed (Mishnaevsky, 2005). The programprodu- ces interactively a command file, which generates 3D FEmicrostructural mo- 538 L. Mishnaevsky Jr. dels of representative volumes of amaterial. In the framework of the program, the information about the spatial distribution of phases in the representative volume is presented as a voxel array. The material volume is presented as an N×N×N array of points (voxels), each of them either black (0) or white (1) (for a two-phase material). A generalization of this approach for amultiphase material can be simply done. The voxel array data can be either read from an input file (real microstructure case) or generated in the program by using the random number generator. The program can generate also voxel arrays for different arrangements of round particles in a matrix (multiparticle unit cell). On the basis of voxel arrays, the program ”Voxel2FEM” generates 3D microstructural models of the representative material volume. Figure 2 shows examples of the generated models (Mishnaevsky, 2005). Fig. 2. 3D FEmicrostructural models generated with the use of the program ”Voxel2FEM”: a porousmaterial, multiparticle unit cell model of a composite, a cell with the inclined fiber 2.3. Routine for damage simulation In this Section, a newly developed ABAQUS subroutine for the damage modeling inAl/SiC composites is presented. Themicromechanisms of damage evolution inmost Al/SiC composites undermechanical loading can be descri- bed as follows: first, some particles become damaged and fail (in the case of relatively large particles) or debond from the matrix (for smaller particles); after that cavities and voids nucleate in the matrix (initially, near the broken particle), grow and coalesce, and that leads to the failure of the matrix liga- ments betweenparticles andfinally to formation of amacrocrack in thevolume Mummery and Derby (1993), Derrien et al. (1999). According to Mummery and Derby (1993), the interface debonding becomes one of the main damage mechanisms in the case of relatively small particles (∼< 10µm), but does not play the leading role in the case of bigger particles. Microstructural effects on damage in composites... 539 Wulf (1995) studied experimentally and numerically the damage growth and fracture in real microstructures of Al/SiC composites. According toWulf (1995), finite-element simulations with the damage parameter based on the model of spherical void growth in a plastic material in a general remote stress fieldwithhigh stress triaxiality, developedbyRice andTracey (1969) produced excellent results forAl/SiC composites: both crack paths in a realmicrostruc- ture of a material and force-displacement curves were practically identical in experiments and simulations. The damage parameter, considered by Wulf, is determined by integrating the incremental damage indicator, which is defined as an increment of the plastic strain divided by the reference failure strain (which is determined as a value inversely proportional to the void growth ve- locity) (Mishnaevsky et al., 1999a; Wulf et al., 1993). In our simulations, the Rice-Tracey damage indicator was used as a parameter of the void growth in the Al matrix. Tomodel the damage and local failure of SiC particle, the criterion of cri- ticalmaximumprincipal stress in the particlematerial was used.According to Derrien et al. (1999), the SiC particles inAl/SiC composites become damaged and ultimately fail when the critical maximum principal stress in a particle exceeds 1500MPa. This value was used in our simulations as a criterion of damage of SiC particles as well. An the ABAQUS Subroutine USDFLD, which calculates the Rice-Tracey damage indicator in thematrix and themaximumprincipal stress in particles, and allows one to visualize the damage (microcrack and void) distribution in the material was developed. The damage was modeled as a local weakening of finite elements in which the damage criterion exceeded the critical value (Mishnaevsky et al., 2003a). After an element failed, the Young modulus of this element was set to a very low value (50Pa, i.e., about 0.00001% of the initial value). 2.4. Comparison of the geometry-based and the voxel array based me- thods of generation of 3D FE models The program ”Voxel2FEM” and the voxel array based method of recon- struction of 3D microstructures were tested by carrying out FE simulation of deformation and damage in composite unit cells with an identical ideal 3D microstructure, generated with the use of the programs ”Meso3D” and ”Voxel2FEM” and comparing the results of simulations. Themultiparticle unit cells (10×10×10mm)with 5 round particles was considered in both cases. The consideredmaterial wasAlmatrix reinforced by SiC particles (volume content 5%). 540 L. Mishnaevsky Jr. The SiC particles behaved as elastic isotropic damageable solids, characte- rized by Young’s modulus EP =485GPa, Poisson’s ratio 0.165 and the local damage criterion, discussed below.TheAlmatrixwasmodeled as an isotropic elasto-plastic solid, with Young’s modulus EM =73GPa, and Poisson’s ratio 0.345. The experimental stress-strain curve for the Al matrix was taken from Mishnaevsky (2004), Mishnaevsky et al. (2004a). When approximating the experimental stress-strain curve by the deformation theory, tjhe flow relation (Ludwik hardening law) is σy =σyn+hε n pl, where σy denotes the actual flow stress, σyn – initial yield stress, and εpl – accumulated equivalent plastic stra- in, h and n – hardening coefficient and hardening exponent. The parameters of the curve are as follows: σyn =205MPa, h=457MPa, n=0.20. As output parameters of the numerical testing of microstructures, the ef- fective response of materials and the amount of failed particles NF versus the far-field strain curves were considered. Totally, the geometry-based model contained about 30000 elements, and the voxel basedmodel 8000 brick elements. Each particle contained about 400 finite elements in the geometry-based model, and 80 elements in the voxel- basedmodel. The nodes at the upper surface of the boxwere connected, and the displa- cement was applied to one node only. The model was subject to the uniaxial tensile displacement loading, 2.0mm. The uniaxial tensile response of each microstructure was computed by the finite element method. The simulations were done with the ABAQUS/Standard. Figure 3b shows the considered cells as well as the stress-strain curves and the fraction of failed elements versus applied strain curves obtained numeri- cally. One can see that the obtained results are very close: the stress-strain curves differ only by 4%.The functions of the fraction of failed particles versus the applied strain, obtained in the simulations, are very close as well. 3. Numerical experiments: effect of microstruxcuture of composites on damage resistance 3.1. Effect of particle arrangement of damage resistance In this part of the work, the effect of particles arrangement on the de- formation and damage evolution in the composite were considered, using the program ”Meso3D”. The mechanical behaviour and damage evolution in the materials with different (artificially designed)microstructures were simulated, Microstructural effects on damage in composites... 541 Fig. 3. 3D unit cells (a) and comparison of simulations in the framework of geometry-based and voxel array based approaches (b) (Mishnaevsky, 2005) and the amount of failed particles and tensile stress-strain curves for each microstructures determined. Figure 1 gives examples of random, regular, clustered, and highly gradient arrangements of particles, considered in the simulations.Two types of gradient particle arrangements were considered: an arrangement of a particle with the vector of gradient (from low to high particle concentration region) coinciding with the loading direction (called in the following the ”gradient Y” micro- structure), and amicrostructurewith the gradient vector perpendicular to the loading vector (called in the following the ”gradient Z”microstructure). The standard deviations of the normal distribution of the Y or Z coordinates of particle centers (for the Y and Z gradientmicrostructures, respectively) were taken 2mm, what ensured rather high degree of gradient. The same standard deviations were taken for the clustered particle arrangements. Figure4 showstensile stress-strain curvesandtheamountof failedparticles in the box plotted versus the far-field strain for random, regular, clustered and gradient microstructures (for 15 particles, VC = 10%). Figure 5 gives distributions of equivalent plastic strains on the particle/matrix interfaces and in a vertical section in themicrostructureswith randomparticle arrangements (15 particles, VC =10%). It can be seen in Fig.4 that the particle arrangement hardly influences the effective response of the material in the elastic area or at small plastic deformation. The influence of the type of particle arrangement on the effective 542 L. Mishnaevsky Jr. Fig. 4. Tensile stress-strain curves and the amount of failed particles in the box plotted versus the far-field applied strain for random, regular, clustered and gradient microstructures (for 15 particles, VC =10%) response of the material becomes significant only at the load at which the particles begin to fail. However, after the particle failure begins, the effect of particle arrangement increases with the increasing load. After the first particle fails, the flow stress of the composite increases with varying arrangement particle in the following order: gradient < random < clustered < regular microstructure. One can see in Fig.4 that the rate of damage growth increases in the following order: gradient< random< regular < clustered. Microstructural effects on damage in composites... 543 Fig. 5. Distributions of equivalent plastic strains on particle/matrix interfaces and in a vertical section in microstructures with random particle arrangements (15 particles, VC =10%) The strength and damage resistance of a composite with a gradient mi- crostructure strongly depends on the orientation of the gradient in relation to the direction of loading. In the case of a microstructure with the vertical gra- dient (along the loading vector), the rate of particle failure is very low (about 6.35 particle/mm) and the particle failure begins at a relatively high displa- cement loading, 0.2mm. In the case of a microstructure with the horizontal gradient (normal to the loading vector), the rate of particle failure is the same as for randommicrostructures. 3.2. Damage evolution in graded composites and the effect of the degree of gradient In the previous Section, it was shown that gradedmicrostructures of com- posites ensure the highest damage resistance among all the considered mi- crostructures. In this Section, we analyse the effect of the degree of particle localization in graded composites on their damage resistance. Here, we use a 2D version of the program ”Meso3D” and 2D FE analysis. As discussed above, the gradient degree of particle arrangement is deter- mined by the standard deviation of the normal probability distribution of distances between the Y -coordinates of particle centers and of the upper bo- undary of the cell. Since the X-coordinates of particles are generated from a pre-defined random number seed parameter (idum) (which should ensure reproducibility of simulations), variations of this parameter lead to the gene- ration of new realizations of microstructures with the same gradient. Many gradedmicrostructures with different standard deviations of the distributions of Y -coordinates (which ensureddifferent gradient degrees) andwith different 544 L. Mishnaevsky Jr. Fig. 6. Schema of design of artificial gradientmicrostructures and some examples of generatedmicrostructures with different degrees of gradient random number seed parameters for random X coordinates were generated, meshed and tested. Figure 7 shows some typical tensile stress-strain curves and the fraction of failed elements in the particles plotted versus the far-field applied strain for graded particle arrangements with different degrees of gra- dient. Fig. 7. Tensile stress-strain curves (a) and the fraction of failed elements in the particles plotted versus the far-field strain (b) for graded particle arrangements with different degrees of gradient Figure 8a shows the failure strain (critical applied strain) plotted versus the degree of gradient in the composites. Figure 8b shows the flow stress of the composite (at the far-field strain u= 0.15) as a function of the gradient degree. It is of our interest that the flow stress and stiffness of composites decrease with the increasing degree of gradient. Apparently, the more homogeneous is Microstructural effects on damage in composites... 545 Fig. 8. Failure strain of the composite (at the far-field strain u=0.15) plotted versus the degree of gradient in the composites the distribution of hard inclusions in the matrix, the stiffer is the composite. If the particles are localized in one layer in the composite, the regions with low particle density determine the deformation of thematerial, and that leads to a low stiffness. One can see in Fig.7b that all microstructures have rather low damage growth rate at the initial stage of damage evolution. At some far-field strain (called here ”failure strain”), the intensive (almost vertical) damage growth takes place and the falling branch of the stress-strain curve begins. For all the gradedmicrostructures, the failure strain is higher than for the homogeneous microstructures. The failure strain of composites increases with the increasing gradient degree. Figure 9 shows von Mises stress distribution in a highly gradient (grad3) microstructure. One can see that the stresses are lower in the low part of the microstructure (particle-free region) than in the particle-rich regions. If two particles are placed very closely one to another, the stress level in the particles ismuchhigher than in other particles, especially if these particles are arranged along the gradient (vertical) vector. Then, the stress level is rather high in particles which are located in the transition region between the high particle density and particle-free regions. One could expect that these particles begin to fail at later stages of loading, and that was observed in damage simulations indeed. Figure 9b shows damage distribution in the particles and in the matrix (grad3 microstructure, far-field strain 0.29). The fact that the particles begin to fail not in the region of high particle density but rather in the transition regionbetween the particle-rich andparticle-free regions is similar to our obse- rvations for the case of the clustered particle arrangement, where the damage 546 L. Mishnaevsky Jr. Fig. 9. VonMises stress distribution in a highly gradient (grad3) microstructure (a) and damage distribution in the particles and in the matrix (grad3microstructure, far-field strain 0.29) begins in particles which are placed at the outer boundaries of clusters (Mish- naevsky et al., 2004a). One can see in Fig.9b that the damage in the matrix begins near the damaged particles or between particles which are arranged closely in the direction of the gradient vector. 3.3. Effect of sharpness of the transition region in graded composites In this Section, we analyse the effect of sharpness of the transition region between phases in graded composites on strength and stiffness of composites. To do it, we use the program ”Voxel2FEM”, described in Section 2.2, which allows us to vary the sharpness of the transition region. The 3D unit cell models of graded composites with varied sharp- ness/smoothness of the transition region between phases were generated as follows. The unit cells were considered as consisting of voxels (points) of white (matrix) and black (particles) phases. The distribution of black voxels was defined as random distributions in the X and Z directions and a graded di- stribution in the Y direction. The graded distribution of black voxels (e.g., grains of hard phase) along the axis Y follows the formula vc(y)= 2vc0 1+exp ( g− 2gy L ) (3.1) Here vc(y) is the probability that a voxel is black at a given point, vc0 is the volume content of the black phase, L denotes the length of the cell, g is a parameter of the gradient, y is the Y -coordinate. Equation (3.1) allows one to vary the smoothness of the gradient interface of the structures (highly localized arrangements of inclusions and a sharp interface versus a smooth interface), Microstructural effects on damage in composites... 547 keeping the volume content of the inclusions constant. If g < 3, the transition between the regions of high content of black or white phases is rather smooth, and if g > 10, the transition between the regions is rather sharp. Figure 10 gives shapes of this curve for different g. Figure 11 gives examples of such microstructures. Fig. 10. Shapes of the curve which describes transition between the regions of high content of different phases for different parameters g (g=5, 10, 100, vc=50%) (Mishnaevsky, 2005); y denotes Y -coordinate, L is the linear size of the cell Fig. 11. Examples of the considered gradedmicrostructures of the material: g=3, g=6, g=100 Figure 12 shows stress-strain curves of composites with a volume content of the hard phase 10% and 20%, and with a varied gradient parameter g, Eq. (3.1). Observing the curves, one can see that the critical strain, at which damage growth begins, does not depend on the parameter of the volume frac- tion gradient g.Whether the transition between the region of high content of the hard phase to the region of low content is sharp or smooth, the critical applied strain remains constant. However, the stiffness of the composite and 548 L. Mishnaevsky Jr. the peak stress of the stress-strain curve increase with increasing sharpness of the transition between the regions. A reduction of the value g from 20 to 1 can lead to a decrease in the peak stress by 6%. Fig. 12. Typical tensile stress-strain curves for different sharpness of transition zones of graded composites: (a) VC =10%, (b) VC =20% (Mishnaevsky, 2005) Figure 13 shows the peak stress of the stress-strain curve plotted versus the parameter g of the transition of sharpness between the regions of high and low content of the hard phase. Comparing this conclusion with the results from Sections 3.1 and 3.2, one may summarize that Al/SiC graded composites with a high gradient degree and smooth transition between the regionwith a high content of the SiCphase and the reinforcement-free region can ensure both highest damage resistance and a relatively high stiffness. Fig. 13. Peak stresses of the stress-strain curve plotted versus sharpness of the transition zone for graded composites with different volume content of the hard phase (10% and 20%) (Mishnaevsky, 2005) Microstructural effects on damage in composites... 549 3.4. Mesomechanical simulation of wear of diamond grinding wheels In this Section, the methods of computational mesomechanics were used to analysemechanical wear of diamond grindingwheels in grinding.We consi- der here straight grinding wheels with synthetic diamonds and bronze copper bond (which are used for grinding ceramics, for instance). 3D FE models of cutouts of the work surface of a grinding wheel have been generated auto- matically using the program ”Meso3D” (Section 2.1). On the contact surface 10× 10mm, 15 diamond grains were randomly placed. The radius of grains was 1.16mm (therefore, 63% of the contact surface was taken by the diamond grains). The material properties were as follows: diamond: Young’s modulus E = 900GPa, Poisson’s ratio ν = 0.2, compressive yield strength 5MPa; bond: E = 93GPa, tensile yield stress 125MPa, yield strain 0.2%, ultima- te tensile strength 255MPa (Bauccio, 1994). The failure stress of the dia- mond grains was assumed to be 20GPa (Mishnaevsky, 1982). The grain tips have been loaded by an inclined force, 70N/grain. The force was oriented at 60◦ angle to the horizontal line. The temperature effect was neglected in the first approximation, following the results by Mishnaevsky (1982), which demonstrated that the local heating up to 200-300◦C (i.e., local tempera- tures observed on grain surfaces at relatively low cutting speeds) does not change the mechanisms and critical parameters of grain destruction. The da- mage in the diamond grains was modeled using the subroutine User Defi- ned Field, presented in Section 3.1, and the element weakening approach. The critical maximum stress was taken as the criterion of the finite element failure. Figure 14 shows the von Mises strain distribution in the diamond grains and in the metal bond in grinding (a) and the fraction of failed elements in the grains plotted versus the applied force (b). One can see in Figure 14 that high strains are localized near the pe- aks of the diamond grains. This corresponds to experimental observations in Mishnaevsky (1982): damage in diamond grains in grinding is observed in a small region near the grain peak. It can be seen in Figure 14b that the damage growth starts at the force 24N and becomes very intensive at about 45N. The results presented in this section demonstrate the applicability of the computational mesomechanics approach and the numerical tools developed above to analysis of wear of diamond grinding wheels. 550 L. Mishnaevsky Jr. Fig. 14. VonMises strain distribution in diamond grains and in the metal bond of a grinding wheel (a) and the fraction of failed elements in the diamond grains plotted versus the applied force (b) 4. Conclusions Numerical investigations of the effect of a microstructure, arrangement and volume content of hard damageable inclusions in a plastic matrix on the de- formation and damage growth are presented in the paper. It was shown that the flow stress of a composite increases with the particle arrangement varying in the following order: highly gradient< random< clustered< regularmicro- structure.The rate of damage growth increases in the following order: gradient < random < regular < clustered. The flow stress and stiffness of composites decrease with the increasing gradient degree, whereas the failure strain incre- ases with the increasing gradient degree of the particle arrangement. More localized and highly gradient microstructures have lower stiffnesses and hi- gher failure strains than homogeneousmicrostructures. It was shown that the stiffness of a composite and the peak stress of the stress-strain curve increase Microstructural effects on damage in composites... 551 with increasing smoothness of the transition between the region of a high vo- lume content of the hard phase to the region of a low content of the hard phase. Acknowledgement The author is grateful to the German Research Council (DFG) for its support through the Heisenberg fellowship. References 1. Bauccio M., edit., 1994, ASM Engineered Materials Reference Book, 2nd Ed., ASM International, Materials Park, OH 2. Derrien K., Baptiste D., Guedra-Degeorges D., Foulquier J., 1999, Multiscale modelling of the damaged plastic behaviour of AlSiCp composites, Int. J. Plasticity, 15, 667-685 3. MishnaevskyL. Jr., 1998,Damage and Fracture of HeterogeneousMaterials, Balkema, Rotterdam, 230pp. 4. Mishnaevsky L. Jr., 2004, Three-dimensional numerical testing of micro- structures of particle reinforced composites,Acta Mater., 52, 14, 4177-4188 5. Mishnaevsky L. Jr., 2005, Automatic voxel based generation of 3D micro- structural FEmodels and its application to the damage analysis of composites, Materials Science and Engineering,A407, 1/2, 11-23 6. MishnaevskyL.Jr.,DerrienK.,BaptisteD., 2004a,Effectofmicrostruc- tures of particle reinforced composites on the damage evolution: probabilistic and numerical analysis,Composites Science and Technology, 64, 12, 1805-1818 7. Mishnaevsky L. Jr., Dong M., Hoenle S., Schmauder S., 1999a, Com- putational mesomechanics of particle-reinforced composites, Comput. Mater. Sci., 16, 1/4, 133-143 8. Mishnaevsky L. Jr., Lippmann N., Schmauder S., 2003a, Computatio- nal modeling of crack propagation in real microstructures of steels and virtual testing of artificially designedmaterials, Int. J. Fracture, 120, 4, 581-600 9. Mishnaevsky L. Jr., Lippmann N., Schmauder S., 2003b,Micromechani- sms andmodelling of crack initiation and growth in tool steels: role of primary carbides, Zeitschrift f. Metallkunde, 94, 6, 676-681 10. Mishnaevsky L. Jr., Lippmann N., Schmauder S., Gumbsch P., 1999b, In-situ observations of damage evolution and fracture in AlSi cast alloys,Eng. Fract. Mech., 63, 4, 395-411 552 L. Mishnaevsky Jr. 11. MishnaevskyL. Jr., Schmauder S., 2001,Continuummesomechanical fini- te elementmodeling inmaterialsdevelopment:a state-of-the-art review,Applied Mechanics Reviews, 54, 1, 49-69 12. MishnaevskyL. Jr.,WeberU., SchmauderS., 2004b,Numerical analysis of the effect of microstructures of particle-reinforcedmetallic materials on the crack growth and fracture resistance, Int. J. Fracture, 125, 33-50 13. Mishnaevsky L.L., 1982,Wear of Grinding Wheels, Kiev, NaukovaDumka 14. MummeryP.,DerbyB., 1993, In:Chapter 14, Fundamentals ofMetal-Matrix Composites, S. Suresh, A. Mortensen, A. Needleman (edit.), Elsevier, 251-268 15. Rice J.R.,TraceyD.M., 1969,On theductile enlargementofvoids in triaxial stress fields, Int. J. Mech. Phys. Solids, 17, 201-217 16. Wulf J., 1995, Neue Finite-Elemente-Methode zur Simulation des Duktil- bruchs in Al/SiC, DissertationMPI für Metallforschung, Stuttgart 17. Wulf J., Schmauder S., Fischmeister H.F., 1993, Finite element model- ling of crack propagation in ductile fracture,Comput. Mater. Sci., 1, 297-301 Wpływ mikrostruktury na zniszczenie kompozytu – analiza numeryczna Streszczenie W artykule przedstawiono analizę wpływumikrostrukturymateriałów kompozy- towychna ich odpornośćna zniszczenie, przeprowadzającbadania symulacyjneoparte namodelach obliczeniowych stosowanychwmezo-mechanice oraz na eksperymentach numerycznych. Opisano zaproponowane i przetestowane programy do automatyczne- gogenerowania trójwymiarowychmodelimikkrostrukturkompozytowychbazującena opisie geometrycznymz zastosowaniemtechnikiwokseli (3-wymiarowychodpowiedni- ków pikseli). Przeprowadzono symulacje deformacji i ewolucji zniszczenia elementów skończonych reprezentujących kompozyty zwtrąceniami punktowymi o różnejmikro- strukturze. Sformułowano pewne wytyczne dla poprawy odporności na zniszczenia lekkich kompozytów zbudowanych zmetalicznego lepiszczawzmacnianego ceramiką. Manuscript received December 24, 2005; accepted for print January 11, 2006