Prediction of hot corrosion behavior of Inconel 617 via machine learning

Abstract

Purpose This paper aims to study the feasibility of using machine learning in hot corrosion prediction of Inconel 617 alloy. Design/methodology/approach By examination of the experimental studies on hot corrosion of Inconel 617, a data set was built for machine learning models. Apart from the alloy composition, this paper included the condition of hot corrosion like time and temperature, and the composition of the saline medium as independent features, while the specific mass change is set as the target feature. In this paper, linear regression, random forest and XGBoost are used to predict the specific mass gain of Inconel 617. Findings XGBoost yields the coefficient of determination (R2) of 0.98, which was highest among models. Also, this model recorded the lowest value of mean absolute error (0.20). XGBoost had the best performance in predicting specific mass gain of the alloy in different times at temperature of 900°C. In sum, XGBoost shows highest accuracy in predicting specific mass gain for Inconel 617. Originality/value Using machine learning to predict hot corrosion in Inconel 617 marks a substantial progress in this domain and holds promise for simplifying the development and evaluation of novel materials featuring enhanced hot corrosion resilience.