Accurately predicting friction and wear in lubricated tribosystems is essential for improving mechanical performance and durability. However, models that simultaneously predict the COF and wear remain scarce due to the complex, non-linear nature of these phenomena and the absence of a direct correlation between them. Tribology has traditionally relied on experimental methods to quantify these properties. In this study, a data-driven machine learning model is developed to predict the COF and WSD over time using data from four-ball tester experiments with lubricants containing various additives. Input parameters include lubricant properties, additive properties, and operating conditions. The data-driven model effectively captures the complex interactions among these variables, demonstrating high predictive accuracy for both the COF and WSD. These findings underscore the potential of deep learning in partnership with large data sets, overcoming modeling limitations in tribology, offering a valuable tool for predictive maintenance and lubricant optimization in industrial applications.
Granja et al. (2026) studied this question.