Bayesian Framework for Rapid Exploration and Establishment of High-dimensional Microstructure-Property Relations

We present a machine learning framework, leveraging several tools from Materials Informatics and microstructure-explicit simulations, to rapidly explore these high-dimensional spaces and establish structure-property relationships for material systems. The framework allows for a rapid establishment of a learned structure-property relationship for enabling materials design and engineering, while minimizing the number of expensive simulations necessary. We achieve this objective by integrating three tool sets: (1) quantification and feature engineering using two-point correlation functions for topologically compact statistical microstructure description, along with Principal Component Analysis for dimensionality reduction of correlation maps; (2) expansion of the microstructure space using synthetic microstructure generation for rapid identification of possible candidate structures; and (3) active learning on Multi-Output Gaussian Process Regression model for rapid establishment of structure-property relations to minimize the number of necessary simulations.
The engineering material system considered is an Al2O3-TiB2 composite system, where we study the effects of microstructure attributes on its failure mechanisms and energy dissipation capacity. We consider microstructure attributes such as volume fraction, size and clustering or agglomeration of reinforcement particles to primarily study the effects of interfaces on the failure mechanisms. Through the proposed framework we show a delineation and summary of the primary microstructure attributes that dominate the failure mechanism and energy dissipation capacity of the material system, and effectively explore the design space considered.