Modeling and optimization of friction materials based on genetic programming and experimental frictional data

Abstract

A genetic programming technique was employed to develop empirical models for predicting specific wear rate and coefficient of friction on brake friction materials. The models consist of independent variables representing the volume fraction of the ingredients. The average absolute relative error for specific wear rate and coefficient of friction were found to be 1.93% and 1.92%, respectively. The models were also verified by the experimental data used for further control of the formulation. Utilizing Tornado plots, the models were found to be able to properly demonstrate the role of the ingredients on overall tribological performance of the brake friction materials. Moreover, a non-dominated sorting genetic algorithm II method was utilized to optimize the formulation of the brake friction materials in terms of braking performance and final cost of the materials.