Modeling the Friction Behavior of Low-Carbon Steel Sheets Using Various Machine Learning Algorithms Based on Strip...

Area of Science:

• Materials Science
• Mechanical Engineering
• Computational Science

Background:

• Sheet metal friction is crucial in manufacturing processes.
• Accurate modeling of friction phenomena is essential for process optimization.
• Experimental data provides a basis for developing predictive friction models.

Purpose of the Study:

• To compare the predictive capabilities of various machine learning (ML) algorithms for sheet metal friction.
• To model the coefficient of friction (CoF) using data from strip drawing tests.
• To identify key parameters influencing friction in sheet metal applications.

Main Methods:

• Utilized strip drawing test data with varying parameters: sheet orientation, load, sample orientation, and sheet drawing quality.
• Trained and evaluated multiple ML algorithms, including a trilayer neural network.
• Assessed model performance using coefficient of determination (R²) and root mean squared error (RMSE).
• Employed SHapley Additive exPlanations (SHAP) for parameter importance analysis.

Main Results:

• A trilayer neural network demonstrated superior predictive performance (R² = 0.986, RMSE = 0.0025).
• Coefficient of friction (CoF) decreased with increased countersample surface roughness and applied load.
• Sample orientation relative to the sheet rolling direction showed a statistically insignificant effect on CoF.
• SHAP analysis and F-test confirmed that countersample roughness and load are the most influential parameters.

Conclusions:

• Machine learning, particularly neural networks, effectively models sheet metal friction.
• Countersample surface roughness and load are critical factors determining the coefficient of friction.
• The developed ML models provide accurate predictions for sheet metal friction under various conditions.