Prediction of surface roughness of tempered steel AISI 1060 under effective cooling using super learner machine learning.

Ziyad, Firi; Alemayehu, Habtamu; Wogaso, Desalegn; Dadi, Firomsa

Surface roughness is a critical factor in evaluating the quality of a product's surface. To predict surface roughness, researchers have employed statistical and empirical methodologies, both of which often lack generalizability when applied to unseen data. To overcome the limitations of existing models, scholars have turned to machine learning and artificial intelligence approaches. Machine learning can accurately predict the surface roughness of machined parts and demonstrates strong generalization ability when applied to new, unseen data. For instance, this research develops a super-learner machine learning model designed to predict surface roughness by leveraging a diverse array of techniques, including kernel ridge regression (KRR), support vector machine (SVM), K-nearest neighbors (KNN), decision trees (DT), random forests (RF), adaptive boosting (ADB), gradient boosting (GB), and extreme gradient boosting (XGB). The optimization of these models was achieved through grid search hyperparameter tuning and K-fold cross-validation. The predictive efficacy of the proposed super-learner model was compared to that of all alternative models. With a coefficient of determination (R2) of 99.8% between the experimental and predicted values for surface roughness on the test dataset, the super-learner model demonstrated superior predictive capabilities. It emerged as the most accurate model, distinguished by the highest R2, the lowest mean absolute error (1.92%), the lowest mean absolute percentage error (1.76%), and the lowest root mean square error (2.29%). Additionally, the model's predictions were further interpreted using the Shapley additive explanations (SHAP) technique, which provided valuable insight into the significant variables influencing the surface roughness of tempered steel AISI 1060.

STANDARD deviations; SURFACE roughness; RANDOM forest algorithms; ARTIFICIAL intelligence; DECISION trees

International Journal of Advanced Manufacturing Technology, 2025, Vol 136, Issue 3, p1421

0268-3768

Academic Journal

10.1007/s00170-024-14952-3