A Property‐Driven Stepwise Design Strategy for Multiple Low‐Melting Alloys via Machine Learning. Issue 12 (12th September 2021)

• Record Type: Journal Article
• Authors: Chen, Huimin; Shang, Zhongwen; Lu, Wencong; Li, Minjie; Tan, Fuping
• Is Part Of: Advanced engineering materials. Volume 23:Issue 12(2021)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2021-09-12
• Subjects: low-melting alloys -- machine learning -- melting point -- property-driven -- stepwise design strategy; Materials -- Periodicals; 620.11
• DOI: 10.1002/adem.202100612
• Languages: English
• ISSNs: 1438-1656

Abstract

Abstract : Low‐melting alloys (LMAs) have an extensive application prospect due to their extremely low melting points for further research of other properties. However, it is difficult to design new multiple alloys with required melting point based on experiments in vast chemical space. Herein, a property‐driven stepwise design strategy for multiple alloys design based on complete machine learning process is developed. The R‐X‐S integrated model based on Ridge Regression (RR), eXtreme Gradient Boosting (XGBoost), and Support Vector Regression (SVR) performs well in melting point prediction with the root mean squared error (RMSE) and correlation coefficient ( R ) on the validation set of 4.578 and 0.988, respectively. After model construction, the stepwise strategy is used to design potential LMAs with smaller estimation error according to the prediction error function combining variance and bias. The candidates with melting point of 90 °C provide the possibility for solder applications with melting points below 100 °C, and the low‐cost candidates with melting point of 16 °C may be used to replace the expensive 75Ga–25In alloy. The stepwise strategy with different step length can improve the search efficiency and map the relationship between LMAs compositions and melting points, which may also be applied to explore other functional materials with high performance. Abstract : A property‐driven stepwise design strategy is proposed for multiple low‐melting alloys based on machine learning. After the stable R‐X‐S model construction, potential low‐melting alloys with small estimation error are designed according to prediction error function combining variance and bias. The stepwise strategy with different step sizes can improve search efficiency.