Mechanistically Enhanced Machine Learning for Materials Design and Process Optimization

Data analytics and machine learning tools enable accelerated materials discovery and process optimization in vast multi-dimensional spaces. Two emerging fields in recent years can benefit from the application of these tools: high entropy alloys (HEAs) and additive manufacturing (AM). Both fields possess such a large design/optimization space that trial-and-error approaches become almost impossible while informative guidance is highly desired to accelerate materials development. Current challenges from machine learning tasks are (1) complexity of available data and (2) lack of information-rich features. For example, a HEA dataset may contain nominal composition, phase formation, and strength, making it a great challenge to build models connecting the basic data to complex material performance. Instead, if we can construct extra features based on mechanistic models, such as CALPHAD phase data or mechanistic strength predictions, which add further linkage between the huge gap between the basic data and desired properties, the success likelihood of machine learning model will be significantly improved. In this proposed Phase I program, QuesTek, partnered with Texas A&M University will develop a framework of mechanistically enhanced machine learning for HEAs and AM by merging the power of machine learning algorithms and ICME models and applying it to data from disparate sources.