A Novel Model for Prediction of Wear and Coefficient of Friction Characteristics of Glass Fibre Reinforced Polytetrafluoroethylene Composites

Abstract

In this paper, two linear (weighted and simple averages) and two non-linear (neural network and support vector) ensemble models were built by integrating the responses of the two nonlinear models namely feed forward neural network (FFNN) and support vector regression (SVR) and one multilinear regression (MLR) model to improve the efficiency of single models in estimating the wear and coefficient of friction (COF) characteristics of glass fibre reinforced Polytetrafluoroethylene composites. The nonlinear models were found to increase the efficiency of single MLR model to at least 25% and 35% for wear and COF, respectively. It was found that SVR ensemble model was most robust by increasing the wear prediction efficiency of MLR, SVR, and FFNN single models to 45.19%, 27.23% and 9.77%, respectively. However, for COF it was seen that FFNN ensemble model was superior to other single models with increase of 39.00% for MLR, 4.05% for SVR and 5.05% for FFNN. These novel models hold potentials of saving cost and time in the study of wear and COF characteristics of FRPCs