Optimizing Boride Coating Thickness on Steel Surfaces Through Machine Learning: Development, Validation, and Experimental Insights

Abstract

In this study, a comprehensive machine learning (ML) model was developed to predict and optimize boride coating thickness on steel surfaces based on boriding parameters such as temperature, time, boriding media, method, and alloy composition. In a dataset of 375 published experimental results, 19 features were applied as inputs to predict the boride layer thickness in various steel alloys. ML algorithms were evaluated using performance metrics like Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and R2. Among the ML algorithms tested, XGBoost exhibited the highest accuracy. XGBoost achieved an R2 of 0.9152, RMSE of 29.57, and MAE of 18.44. Incorporating feature selection and categorical variables enhanced model precision. Additionally, a deep neural network (DNN) architecture demonstrated robust predictive performance, achieving an R2 of 0.93. Experimental validation was conducted using 316L stainless steel (SS), borided at 900 degrees C and 950 degrees C for 2 h and 4 h. The DNN model effectively predicted the boride thickness under these conditions, aligning closely with the observed values and confirming the models' reliability. The findings underscore the potential of ML to optimize boriding processes, offering valuable insights into the relationships between boriding parameters and coating outcomes, thereby advancing surface modification technologies.