Tribological Analysis of Ti6Al4V Using Machine Learning to Determine Coefficient of Friction and Wear Rate

Abstract Random Forest and AdaBoost, machine learning (ML) models are applied in this paper to predict friction coefficient (COF) and wear rate based on experimental datasets collected from the literature. Suboptimal classification performance for some models is one challenge faced by this study due to limitation in the number of experiments available in the literature. Therefore, this research considered the problem as a binary classification task through employment of threshold value instead of regression problem. Even with these limitations, both Random Forest and AdaBoost demonstrated significant success in classifying data sets with or without reduction dimensionality using Principal Component Analysis (PCA). These models were useful for binary classification, specifically when considering pre-defined thresholds for COF and wear rate. The conclusion suggests that future work could improve ML model performance by comparing their regression capabilities in augmented datasets from real lab experiments. Additionally, it advocates for an inverted application of ML approaches to obtain input parameters corresponding to desired COF and wear rate values.