Probabilistic Framework for Prediction of Material Property Distributions from Small Microstructural Models (Preprint).

摘要

A probabilistic framework for prediction of material property distributions from small scale (i.e. 2-grain) models is proposed. Monte Carlo Simulation and kernel density estimation are used to estimate the material property distribution of a grain boundary with a 2-grain model. Extreme value and order statistics are then employed to estimate the distribution of larger microstructure models. An example of the methodology is presented for identifying the applied uniaxial stress at which plastic slip initiates in a titanium alloy with a crystal elastic finite element model. The framework was verified by comparing the predicted plastic slip initiation strength distribution with the obtained distribution from Monte Carlo Simulation of larger scale finite element models (i.e. n-grain models, up to approximately 600 grain RVE). The methodology performs well for larger microstructure models but less so for smaller ones, for a much smaller computational cost.