Predicting wear damage in moving mechanical contacts: Comparative analysis of regression algorithms and feature...

Area of Science:

• Materials Science
• Mechanical Engineering
• Data Science

Background:

• Accurate wear prediction is vital for reducing operational risks, minimizing downtime, and extending component lifespan in manufacturing, transportation, and power generation.
• Pin-on-disc systems are widely used to study tribological behavior and material wear under controlled conditions.

Purpose of the Study:

• To develop and evaluate a machine learning-based predictive model for estimating wear volume in pin-on-disc systems.
• To identify the most influential parameters affecting wear and compare the performance of different regression algorithms for wear prediction.

Main Methods:

• Utilized experimental data including friction coefficient, tangential force, penetration depth, sliding distance, sound pressure, and load.
• Employed feature selection techniques (wrapping and embedding) to identify key predictive parameters.
• Determined optimal sample size to balance model complexity and prevent overfitting.
• Applied and compared regression models: linear regression, support vector machines (SVMs), and random forests (RFs).
• Evaluated model performance using Mean Absolute Error (MAE), Mean Bias Error (MBE), Root Mean Square Error (RMSE), and Coefficient of Determination (R²).

Main Results:

• Feature selection successfully identified the most relevant parameters for wear prediction, enhancing model accuracy.
• Optimizing sample size improved model generalization, mitigating underfitting and overfitting.
• Comparative analysis demonstrated the varying predictive capabilities of different regression algorithms for wear volume estimation.
• The developed models provided quantitative assessments of wear volume with defined error metrics.

Conclusions:

• The study presents a systematic machine learning approach for accurate wear volume prediction in pin-on-disc systems.
• The findings offer valuable insights into parameter importance and algorithm performance for wear analysis.
• The research provides a robust framework for practitioners and researchers aiming to improve wear prediction accuracy and reduce industrial operational costs.