Can machine learning predict friction from third body morphology?

Abstract

This work proposes to lay the foundations for building an efficient database that is representative of the morphology of the third body, with the aim of understanding whether it is possible to predict the local friction coefficient from it, using Machine Learning (ML). Five different databases (including morphological properties of ejected wear particles and textural properties of on-track third-body) are constructed, and a Random Forest (RF) machine learning algorithm is implemented. Results show that an algorithm trained on third body morphological features can provide a fairly accurate prediction of the local value of the friction coefficient in a variety of tribological situations, with an average error close to 0.14 for measured values ranging between 0.1 and 1.2.