The Machine Learning–Based Prediction and Experimental Validation of the Friction Performance of Recycled Brake Materials

Abstract The increasing emphasis on sustainability in the transportation sector has driven interest in friction brake materials containing high fractions of recycled constituents. However, designing such materials while maintaining stable and reliable friction performance remains challenging due to the complex, nonlinear nature of tribological behavior. This study presents a machine learning–based framework for predicting the coefficient of friction (CoF) of recycled friction brake materials using data generated from scaled-down Federal Motor Vehicle Safety Standard (FMVSS 135) laboratory-scale friction testing. Brake pad samples containing up to 60 wt.% recycled friction material were developed using a Taguchi L8 Design of Experiments (DOE) to systematically vary material composition. The experimental data were used to train and validate Artificial Neural Network (ANN) and Random Forest (RF) models for CoF prediction. The predictive capability of the models was subsequently assessed by manufacturing new brake pad formulations and subjecting them to identical FMVSS 135 test conditions. Both ANN and RF models demonstrated strong predictive accuracy, with predictions closely matching experimentally measured CoF values. The results confirm the robustness and reliability of the proposed framework and demonstrate the feasibility of sustainable brake materials while reducing experimental effort and material consumption.