Processing Optimization and Property Predictions of Hot-Extruded Bi–Te–Se Thermoelectric Materials via Machine Learning

Abstract

Traditional experiment-based materials research is becoming increasingly insufficient to thoroughly understand materials’ characteristics and thus it is becoming difficult to develop novel materials with better performance. Machine learning is applied to hot-extruded CuxBi2Te2.85+ySe0.15 thermoelectric materials, and the relationships between the processing, microstructure, and properties are further investigated via a data-driven approach. A properties-to-microstructure-to-processing inverse analysis is proposed and performed by a genetic algorithm. The analysis results indicate that hot-extruded materials have a potential best figure of merit (ZT) value of 1.15, which is 1.32 times larger than their best experimental value (0.87). To obtain this optimal property, processing variables such as higher extrusion temperature and larger Cu content and microstructure with a larger average grain size and higher density are required. The proposed data-driven approach is expected to provide a new avenue for designing high-performance Bi–Te–Se thermoelectric materials and thus to accelerate discoveries of novel materials.