Searching for medium entropy alloys with desired mechanical property by adaptive design combined with experiments

To overcome the problems caused by the trial and error method in the vast element composition space and to rapidly design and discover new medium entropy alloys (MEAs) with required mechanical properties, an efficient data-driven machine learning (ML) strategy based on random forest regression (RFR) algorithm was proposed, utilizing existing data. The alloys were represented by chemical composition and additional features composed of material descriptors. Through correlation analysis, recursive elimination, and exhaustive screening, three key additional features from 20 material descriptors were identified, including six power of work function (W6), valence electron concentration (VEC), and mean Yong's modulus (E). Furthermore, an Efficient Global Opti-mization (EGO) algorithm was adopted to search for the optimal alloys to guide synthesis. After evaluating >448,000 kinds of alloy systems, the most strongly recommended alloy Al52Co24Cr18Ni4Mn2 with the highest predicted hardness value of 802.5 HV was determined. Interestingly, experimental validation of the synthesized alloy revealed mainly BCC and ordered B2 structure, with a hardness of 815.4 HV, which is approximately 5.2% higher than the highest value in the original dataset. Given the independence of the problem, the proposed approach could be extended to other potential multi-component regression problems.

Automated summary

In order to rapidly design and discover new medium entropy alloys (MEAs) with required mechanical properties, a data-driven machine learning (ML) strategy based on random forest regression (RFR) algorithm was proposed. Through correlation analysis and iterative screening, key additional features from material descriptors were identified. An Efficient Global Optimization (EGO) algorithm was used to search for optimal alloys and the alloy Al52Co24Cr18Ni4Mn2 was determined to have the highest predicted hardness value. Experimental validation revealed a higher hardness than the highest value in the original dataset. This approach can be extended to other multi-component regression problems.