Hardness Predicting of Additively Manufactured High-Entropy Alloys Based on Fabrication Parameter-Dependent Machine Learning

High-entropy alloys (HEAs) have received much attention since presented in 2004. Machine learning (ML) can accelerate the research of new HEAs. At present, among the ML research methods used to predict the properties of HEAs, alloys are manufactured mainly by the melt-casting method. The existing ML methods do not use the process parameters of the manufacturing process as input features. Unlike the melt-casting method, additive manufacturing (AM) has promising applications with its ability to prototype and manufacture complex-shaped parts rapidly. The AM process parameters can significantly affect the performance of HEAs. The process parameters are a critical factor that must be considered for ML. Therefore, an ML method dependent on AM process parameters is proposed to predict the hardness of HEAs. The prediction results of six commonly used ML models are compared. The dependence of ML on process parameters is investigated. Four new HEAs are manufactured based on AM to verify the reliability of ML prediction results. The experimental results show that adding process parameters to ML improves the prediction accuracy by 4%. The prediction accuracy of ML reaches 89%, and the average prediction error for new HEAs is 3.83%.

Automated summary

High-entropy alloys (HEAs) have gained significant attention since their introduction in 2004. Machine learning (ML) is proposed as a tool to accelerate the research on new HEAs. Unlike the traditional melt-casting method, additive manufacturing (AM) has the potential for rapid prototyping and manufacturing of complex-shaped parts. The ML method proposed in this study takes into account the AM process parameters to predict the hardness of HEAs. Experimental results show that incorporating process parameters into ML improves the prediction accuracy by 4%, with an overall accuracy of 89% and an average prediction error of 3.83% for new HEAs.