Machine learning insights for predicting density and hardness in centrifugal SHS synthesized ceramic coatings

Ceramic coatings are essential for steel as they protect against corrosion and high-temperature abrasion, significantly improving steel pipes' durability and wear performance. The ceramic coatings fabricated through the Centrifugal Self-propagating High-temperature Synthesis (C-SHS) technique are ideal for application on the inner surface of pipes. Various parameters, including rotating speed, powder particle size, type, and wt.% of additives, influence the ceramic layer's properties. In the present work, several Machine Learning (ML) regressors, such as Categorical Boosting (CatBoost), Decision Tree (DT), Polynomial Regression (PR), Stacking Regression (SR), Extreme Gradient Boosting Regression (XGBoost), and Bagging Regression (BR), were employed to effectively predict the density and hardness of ceramic coating based on the influencing parameters. Feature engineering analysis revealed that rotating speed and the wt.% of additives highly influenced the density and hardness of the coating. Gaussian noise augmentation is applied to generate noise data with the same standard deviation to improve the prediction accuracy of the models through an increase in the database volume. Among the models, the Bagging Regression (BR) model demonstrated the best prediction performance. The BR model achieved a Coefficient of Determination (R²) of 0.91, a Root Mean Square Error (RMSE) of 0.10, and a Mean Absolute Error (MAE) of 0.06 for density predictions. For hardness, the BR model recorded with R2 of 0.93, RMSE of 52.03 and MAE of 40.06. Experimental results confirmed the accuracy of the BR model, with prediction errors of only 3.38 % for density and 4.97 % for hardness.