Integration of machine learning prediction and optimization for determination of the coefficient of friction of textured UHMWPE surfaces

Abstract

The frictional performance of water-lubricated UHMWPE is influenced by the combination of structural parameters and operating conditions. To improve the efficiency of optimal design of surface texture aimed at improving frictional performance, a novel integration of the Orthogonal Array method (OAM), machine learning (ML) prediction, and Particle Swarm Optimization (PSO) is proposed for predicting and optimizing the coefficient of friction (COF) of copper ball-textured UHMWPE surfaces using a small dataset. In order to reduce manufacturing and testing cost, decrease required training samples for ML algorithm, OAM which could efficiently acquire data set with comprehensive feature information is used to determine the parameters of test samples to generate a small but effective dataset. 25 textured samples based on L16 (4⁴) and L9 (3⁴) are fabricated, with the parameter set determined using OAM. COFs of the samples are tested using RTEC tribo-tester. Trend analysis is conducted to investigate the influence of force, frequency, depth and ellipse axis ratio on COF. Multi-linear Regression (MLR) and Gaussian Process Regression are employed. MLR exhibits better prediction accuracy and is integrated with PSO to minimize COF. The error between the experimental and the theoretical results obtained by the integration method of MLR and PSO is only 1.04%, demonstrating the feasibility of predicting COF and optimizing surface texture using the integrated method with a limited dataset determined by OAM.