Accelerated development of hard high-entropy alloys with data-driven high-throughput experiments

Abstract

The development of multicomponent alloys with target properties poses a significant challenge, owing to the enormous number of potential component combinations, high costs and the inefficiency of conventional empirical trial-and-error experimental approaches. To tackle this challenge, we develop a machine learning (ML)-guided high-throughput experimental (HTE) approach to accelerate the development of non-equimolar hard CoxCryTizMouWv high-entropy alloys (HEAs). We first develop a set of all-process HTE facilities ranging from multi-tube ingredient assignment to multi-station electrical arc smelting and specimen preparation for bulk alloy samples with discrete compositions. Instead of random or combinatorial composition searching, HEAs with only ~1/28 of all the potential compositions are synthesized in two stages guided by the ML prediction. The final ML models, trained using 138 experimental data, predict the alloy hardness with mean relative errors of 5.3%, 6.3% and 15.4% at high (HV > 800), medium (HV = 600-800) and low (HV < 600) hardness ranges, respectively. In total, 14 superhard HEAs with HV > 900 are discovered by our ML-guided HTE approach. Moreover, the multiple ML models predict the hardness of 3876 hypothetical alloys covering the whole composition range, thereby revealing the systematic component effects based on the complete composition-hardness and descriptor-hardness correlations. The hardening mechanisms are elaborated by analyzing the microstructures of CoCrTiMoW. Furthermore, physical insights can be gained by transitioning from “machine learning” to “learning from machine”. This work demonstrates that our ML-guided HTE approach provides an effective strategy for multicomponent alloy development with a potential hundred-fold overall increase in efficiency at a fraction of the cost compared to conventional methods.