Showing 1 results for Regression Models.
Ali Azari Beni, Saeed Rastegari,
Volume 22, Issue 3 (9-2025)
Abstract
Aluminide coatings are widely used in high-temperature applications due to their excellent corrosion resistance and thermal stability. However, optimizing their composition and thickness is crucial for enhancing performance under varying operational conditions. This study investigates the optimization of aluminide coatings through a data-driven approach, aiming to predict the coating thickness based on various composition and process parameters. A comparative analysis of six machine learning models was conducted, with the k-nearest neighbors regressor (KNNR) demonstrating the highest predictive accuracy, yielding a coefficient of determination R² of 0.78, a root mean square error (RMSE) of 18.02 µm, and mean absolute error (MAE) of 14.42. The study incorporates SHAP (Shapley Additive Explanations) analysis to identify the most influential factors in coating thickness prediction. The results indicate that aluminum content (Al), ammonium chloride content (NH4Cl), and silicon content (Si) significantly impact the coating thickness, with higher Al and Si concentrations leading to thicker coatings. Zirconia (ZrO2) content was found to decrease thickness due to competitive reactions that hinder Al deposition. Furthermore, the level of activity in the aluminizing process plays a crucial role, with high-activity processes yielding thicker coatings due to faster Al diffusion. The pack cementation method, in particular, produced the thickest coatings, followed by gas-phase and out-of-pack methods. These findings emphasize the importance of optimizing composition and processing conditions to achieve durable, high-performance aluminide coatings for high-temperature applications.
