Accelerated Composition Optimization Of Hard High Entropy Alloys By Combining High-throughput Experiments And Machine Learning Methods

Developing multicomponent alloy by conventional empirical trial-and-error experiment approaches make the development process slow and expensive due to the enormous potential composition combination.To tackle this challenge,we developed a machine learning(ML)guided high-throughput experiment(HTE)approach,accelerating the development of non-equimolar hard high entropy alloy(HEA)Co_xCr_y Ti_zMo_uW_v.We designed and developed the all-process HTE facilities covering from multi-tube ingredient assignment to multi-station electronic arc smelting and specimen preparation for bulk alloy samples with discrete compositions.Instead of huge combinatorial composition search,only the~1/28 fractional HTE were conducted at two stages guided by the ML prediction.First we developed and evaluated 12 ML methods combining various ML algorithms and descriptors based on the111 experiments initially designed by varying the critical Mo and W compositions as well as phase descriptors and built 120 models.The selected preliminary ML models were used to design 27 compositions with various hardness at the second experiment stage.The final ML models,trained using the all 138 experiment data,predicted the hardness with the mean relative errors of 5.3%,6.3%,and 15.4%at the high(HV>800),medium(HV=600-800),and low(HV<600)hardness ranges,respectively,compared with the experiment uncertainty1.69%?1.88%?1.87%.Moreover,the multiple ML models predicted the hardness of hypothetical 3876 alloys covering the whole Co_xCr_yTi_zMo_uW_v composition range,revealing the consistent component effects based on the complete composition-hardness and descriptor-hardness correlations.The hardening mechanisms were elaborated by analyzing the microstructures of CoCrTiMoW.This work demonstrated that the machine learning guided high-throughput experiment(ML-HTE)approach provides an effective strategy for multicomponent alloy development with possible hundred times overall acceleration at lower cost.