Application of SHapley Additive ExPlanation machine learning approach and ANNs to analyze friction phenomenon in low-carbon steel/tool steel contacts

Authors

• Tomasz Trzepieciński Rzeszów University of Technology https://orcid.org/0000-0002-4366-0135
• Marwan T. Mezher Universidade de Vigo, Middle Technical University
• Marek Kowalik Casimir Pulaski Radom University Radom
• Sherwan Mohammed Najm Northern Technical University
• Mihaela Oleksik University of Sibiu
• Salah Eddine Laouini University of El Oued
• Salwa O. Mohammed Northern Technical University

DOI:

https://doi.org/10.30678/fjt.173168

Keywords:

coefficient of friction, deep drawing, friction, low-carbon steel, sheet metal forming, steel sheets

Abstract

The friction analysis in sheet metal forming is difficult due to the complex influence of numerous friction process parameters on the coefficient of friction (COF). Artificial intelligence tools are one of the methods supporting the analysis of complex models. This article presents the application of Additive ExPlanation (SHAP) Machine Learning (ML) methods and artificial neural networks to recognize important friction process parameters based on strip drawing test results for low-carbon steel/tool ​​steel contacts. Friction tests were conducted under lubrication conditions with selected vegetable oils, an alternative to petroleum-based lubricants consistent with sustainable manufacturing. Experimental results showed that rape-seed oil demonstrated the best lubrication performance among the analysed oils with higher and lower viscosity categories. Cumulative SHAP plots and Shapley values indicated that the average surface roughness of counter samples demonstrated the most significant relative importance (35.02%) of a given variable on the COF. The second most significant factor affecting COF was contact pressure, with relative importance of 23.98%-29.33%, depending on the ML algorithm used. The Gradient Boosting ML algorithm demonstrated the best predictive ability over all other tested algorithms (Decision Tree, linear regression, Random Forest, Ridge, and Lasso).  Based on the obtained results, it can be concluded that ML models enable the prediction of the COF under various friction conditions, allowing the evaluation of new lubricant formulations already at the sheet metal forming process design stage, thereby reducing the number of laboratory tests.