Machine learning assisted composition effective design for precipitation strengthened copper alloys

Optimizing the composition and improving the conflicting mechanical and electrical properties of multiple complex alloys has always been difficult by traditional trial-and-error methods. Here we propose a machine learning strategy to design alloys with remarkable properties by screening key alloy factors through correlation screening, recursive elimination and exhaustive screening, and then designing composition iteratively through Bayesian optimization. Taking the precipitation strengthened copper alloys as an example, 5 kinds of key alloy factors affecting hardness (HV) and 6 kinds of key alloy factors affecting electrical conductivity (EC) were obtained by screening alloy factors. HV - key alloy factors model with error less than 7% and the EC - key alloy factors model with error less than 9% were established, respectively. Then, new copper alloys were effectively designed utilizing Bayesian optimization and iterative optimization experiments. Designed Cu-1.3Ni-1.4Co-0.56Si-0.03Mg alloy has excellent combined mechanical and electrical properties with the measured ultimate tensile strength (UTS) of 858 MPa and EC of 47.6%IACS. The property results are superior to the reported precipitation strengthened copper alloys, which realize the simultaneous improvement of the conflicting mechanical and electrical properties. (C) 2021 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.

Automated summary

A machine learning strategy was proposed to design alloys with remarkable properties by screening key alloy factors and utilizing Bayesian optimization, successfully designing new copper alloys with improved hardness and electrical conductivity, achieving simultaneous improvement of conflicting properties.