Development and Evaluation of Machine Learning Based Predictive Models for Tribological Properties of Blended Coatings at Elevated Temperature

The current research is undertaken to evaluate the Tribological properties like wear and Coefficient of Friction (CoF) of three popular blended coatings on a mild steel substrate at elevated temperature. The scope of the research also includes predicting the tribological properties by employing three Machine Learning (ML) based predictive models viz. Elastic Net, k-NN and Random Forest regressions. The regressions are fit and tested at different proportions of Training and Testing data to find the best proportion. Random Forest regression is observed to be the best fit based on the acceptable values of MSE and R-Squared. Random Forest regression model of wear yielded MSE and R-Squared values as 22.01 and 0.95 for Coating 1, 5.75 and 1 for Coating 2, and 14.13 and 1 for Coating 3, respectively. Likewise, Random Forest regression model of CoF yielded MSE and R-Squared values as 0.01 and 0.99 for Coating 1, 0 and 1 for Coating 2, and 0 and 1 for Coating 3, respectively. The deviation between the experimental and predicted results (tested data: experimental runs 3, 14, and 29) in wear using the Random Forest algorithm for Coating 1, Coating 2, and Coating 3 is found to be 21.18%, − 2.72%, and 0.42%; − 4.54%, − 13.87, and 2.57%; 11.85%, 1.69%, and 1.89%, respectively. The deviation for CoF is found to be 6.29%, 1.56%, and 2.93%; − 0.86%, − 0.56%, and 0.20%; 0.85%, − 0.19%, and 0.17%, respectively. The variance between the actual experimental and predicted results from Random Forest regression is observed to be relatively acceptable.