The depth-dependent strain partitioning across the interfaces in the growth direction of the NiAl/Cr(Mo) nanocomposite between the Cr and NiAl lamellae was directly measured experimentally and simulated using a finite element method (FEM). Depth-resolved X-ray microdiffraction demonstrated that in the as-grown state both Cr and NiAl lamellae grow along the direction with the formation of as-grown distinct residual ~0.16% compressive strains for Cr lamellae and ~0.05% tensile strains for NiAl lamellae. Three-dimensional simulations were carried out using an implicit FEM. First simulation was designed to study residual strains in the composite due to cooling resulting in formation of crystals. Strains in the growth direction were computed and compared to those obtained from the microdiffraction experiments. Second simulation was conducted to understand the combined strains resulting from cooling and mechanical indentation of the composite. Numerical results in the growth direction of crystal were compared to experimental results confirming the experimentally observed trends.

Parallel linear equation solvers are one of the most important components determining the scalability and efficiency of many supercomputing applications. Several groups and companies are leading the development of linear system solver libraries for HPC applications. In this paper, we present an objective performance test study for the solvers available on a Cray XE6/XK7 supercomputer, named Blue Waters, at National Center for Supercomputing Applications (NCSA). A series of non-symmetric matrices are created through mesh refinements of a CFD problem. PETSc, MUMPS, SuperLU, Cray LibSci, Intel PARDISO, IBM WSMP, ACML, GSL, NVIDIA cuSOLVER and AmgX solver are employed for the performance test. CPU-compatible libraries are tested on XE6 nodes while GPU-compatible libraries are tested on XK7 nodes. We present scalability test results of each library on Blue Waters, and how far and fast the employed libraries can solve the series of matrices. Keywords-parallel linear equation solver; dense direct solver; sparse direct solver; sparse iterative solver; CPU-compatible library; GPU-compatible library

In this study, a planar spring lattice model is used to study the evolution of damage variable dL in disordered media. An elastoplastic softening damage constitutive law is implemented which introduces a cohesive length scale in addition to the disorder-induced one. The cohesive length scale affects the macroscopic response of the lattice with the limiting cases of perfectly brittle and perfectly plastic responses. The cohesive length scale is shown to affect the strength-size scaling such that the strength increases with increasing cohesive length scale for a given size. The formation and interaction of the microcracks is easily captured by the inherent discrete nature of the model and governs the evolution of dL . The proposed method provides a way to extract a mesoscale dependent damage evolution rule that is linked directly to the microstructural disorder.

In this study 2d two phase microstructures closely resembling the experimentally captured micrographs of the interpenetrating phase composites are generated using a Gaussian correlation function based method. The scale dependent bounds on the effective thermal conductivity of such microstructures are then studied using Hill-Mandel boundary conditions. A scaling function is formulated to describe the transition from statistical volume element (SVE) to representative volume element (RVE), as a function of the mesoscale δ, the correlation length of the Gaussian correlation function λ, the volume fraction v, and the contrast k between the phases. The scaling function is determined through fitting the data from extensive simulations conducted over the parameter space. The scaling function shows that SVE approaches RVE as (δ/λ)−1.16. A material scaling diagram allows estimation of the RVE size, to within a chosen accuracy, of a given microstructure characterized by the correlation length of the Gaussian correla...

A Nonlinear elastic visco-plastic thermo-mechanical steel microstructure model is coupled with a Gurson-Tvergaard-Needleman (GTN) model to predict local damage and failure in the columnar solidification zone of a steel casting. The new model operates on two scales a unit cell grain model of micro-scale in the columnar zone as well as a macro-model tensile specimen to map observations with experiments. The model aims to investigate inter-granular embrittlement at intermediate temperatures during solidification processes. This embrittlement occurs due to pro-eutectoid ferrite film formation and precipitation of inclusions at the prioraustenite grain boundaries. This behavior of the unit cell is then mapped onto macro-model tensile specimens to measure reduction in ductility. The effect of ferrite film and temperature are studied by calculating the micro-strains, macro-strains and void fractions at which cracks begin to form during the continuous casting process. INTRODUCTION: Steel is an important engineering material which is subject to defects and embrittlement at elevated temperatures. Although thermo-mechanical analysis using computational models has been applied as a powerful tool to predict stress and deformation during steel processing, little previous work has been done to predict steel ductility. PROCEDURES, RESULTS AND DISCUSSION: The model developed in this work to predict steel ductility has two different scales – a micro-scale grain model capturing the microstructural behavior and a macro-model capturing necking behavior during a tensile test [Cardoso 1995]. The grain model represents the columnar microstructure as a honeycomb of regular hexagons, as shown in Fig.1. The unit cell modeled includes three grain boundaries intersecting at the triple point and parts of the three regular hexagonal grains, which gives a rectangular shape to the unit cell. The preliminary isothermal model presented here solves the standard mechanical equilibrium equations, with total strain decomposed into elastic and inelastic (plastic) components [Koric 2006]. The dimensions of the grain model are based upon grain size measurements from experiments and equating the area of the grain to that of a circular grain of the same diameter as the grain size measurement. To account for the effect of ferrite films formation, the grain boundary region is assumed to be composed entirely of pro-eutectoid alpha ferrite while the grain matrix region consists of prior austenite and alpha-ferrite. The phase fractions are obtained from the Fe-C phase diagram at the temperature of interest. The temperature dependent mechanical constitutive model for the austenite is based on the formulations by Kozlowski [Kozlowski 1992], while the mechanical model of alpha ferrite is based on the power law formulation [Zhu 1993] An experiment based strain rate is used as input to both these models to get the plastic behavior relationship of the two phases as shown in Fig 2. The behavior of the ferrite film is the same as

Modern numerical algorithms for computational electromagnetics lead to many large sparse systems of linear equations. Their solution takes up to 90% of the total computational time in the geophysical inversion process. This paper provides evaluation and comparison of several state-of-the-art direct solvers in a massively parallel environment. We determine the largest complex systems that can be solved today with these methods and evaluate their performance and scalability on one of the world’s most powerful supercomputers. Small sensitivity of direct methods to the number of sources, modeling frequency and conductivity distribution in the subsurface is confirmed. The results show the potentials and limitations of different parallel implementations on a petascale high-performance computing system.