Data-Driven Modeling of Friction in Drawbead Test Through Advanced Machine Learning.

Area of Science:

• Materials Science
• Mechanical Engineering
• Computational Science

Background:

• Friction in metal forming drawbeads critically influences product quality.
• Experimental determination of friction's complex influence on the coefficient of friction (CoF) is challenging.

Purpose of the Study:

• To apply machine learning (ML) to analyze drawbead simulator test data.
• To identify the most effective ML algorithm for predicting CoF.
• To detail feature importance and explainability of ML models.

Main Methods:

• Utilized experimental data from drawbead simulator tests on DC04 steel.
• Compared three ML algorithms: support vector machine (SVM), regression trees, and ensemble trees.
• Evaluated model performance using root mean square error (RMSE) and correlation coefficient (R²).
• Analyzed feature importance, permutation importance, and Shapley values for model interpretability.

Main Results:

• The SVM algorithm with a cubic kernel achieved the best performance (RMSE=0.0085, R²=0.9657).
• Key predictors influencing CoF, in descending order, were friction conditions, drawbead height, sample width, roughness (Sa), and orientation.
• Shapley analysis indicated low drawbead height strongly increases CoF, while low values of other parameters decrease CoF.

Conclusions:

• SVM is a highly effective ML method for modeling drawbead friction.
• Drawbead height is the most influential parameter affecting the coefficient of friction.
• Understanding parameter effects via ML enhances prediction accuracy and process optimization in metal forming.