Directional Surface Roughness Parameters as Predictors of Polymer Wear Debris Size under Dry and Lubricated Conditions: An Experimental and Machine Learning Approach

This study investigates the influence of counter surface directional roughness parameters on polymer wear debris size under dry and distilled water-lubricated conditions, while integrating machine learning to enhance predictive capability. High-density polyethylene pins were tested against stainless steel plates with varied grit-induced roughness and orientations: unidirectional parallel (UPL), unidirectional perpendicular (UPD), and criss-cross (CC). Debris size, quantified via equivalent circle diameter (ECD) using SEM and image analysis, was correlated with two novel 2D roughness parameter categories measured parallel (Ra‖) and perpendicular (Ra⊥) to the sliding direction. Under dry conditions, debris size showed the strongest correlation with Δa‖ (r = 0.94), whereas under lubrication, Δa⊥ (r = 0.93) was most predictive, highlighting the advantages of directional over conventional 3D parameters. Experimental datasets combining grit size, surface topography, and lubrication were used to train multiple regression models: Linear, Polynomial, Decision Tree, Random Forest, and Support Vector Regression (SVR). Polynomial Regression and Decision Tree consistently delivered the highest accuracy (R² ≈ 1.00), effectively capturing nonlinear dependencies and validating experimental observations of topography lubrication interactions. The combined experimental and machine learning approach provides a robust framework for predicting and controlling wear debris size, offering valuable insights for optimizing implant surface design to minimize biologically detrimental debris formation and extend prosthetic service life.