Data Augmented Machine Learning Approach for Predicting Wear Behavior of 3D Printed Recycled and Virgin Polymers

Polylactic Acid (PLA) and Polyethylene Terephthalate Glycol (PETG) are widely used in additive manufacturing for engineering applications. However, their tribological performance, particularly in recycled forms, remains underexplored. This study evaluates the wear behavior of virgin and recycled PLA and PETG under dry and lubricated conditions using a pin-on-disk tribometer. Key tribological parameters, including wear rate, friction coefficients, energy dissipation, and temperature rise, were measured across 336 samples. Machine learning models were employed to predict wear rate and classify material-lubrication performance. Gradient Boosting Regression achieved the highest prediction accuracy (R² = 0.698, RMSE = 0.1336), with energy dissipation emerging as the most influential factor. Classification models distinguished between high and low-performing conditions, with Logistic Regression achieving an accuracy of 88%. Data augmentation using a Gaussian Mixture Model-based approach enhanced model robustness by expanding the dataset from 168 to 336 samples. Experimental results indicate that recycled materials exhibit higher wear rates, but lubrication significantly reduces material loss. These insights are crucial for manufacturing, biomedical, and automotive applications, where selecting appropriate materials and lubrication strategies can enhance overall durability and efficiency. This study demonstrates the integration of tribological testing with machine learning, providing a data-driven approach for wear prediction and material optimization.