Persistent-Homology-Based Microstructural Optimization of Materials Using t-Distributed Stochastic Neighbor Embedding

摘要

Microstructure optimization is a core issue to maximize the performance of materials. Due to the increasing demand for highly efficient materials, traditional trial-and-error-based experimental methods have become insufficient for designing novel materials with useful properties. Based on the fact that materials with similar microstructural features exhibit similar properties, this work proposes a persistent-homology-based microstructure optimization approach performed with a machine learning algorithm of t-distributed stochastic neighbor embedding to find optimal microstructures for specific properties. The method is applied to dual-phase steels, where a microstructure with high-fraction martensite is identified for achieving a maximum stress. The method proposed here is expected to provide new basis to understand the materials paradigm and thus accelerate the materials discovery process.