Predicting Friction under Vastly Different Lubrication Scenarios

Abstract

Friction is ubiquitous in daily life, from nanoscale machines to large engineering components. By probing the intricate interplay between system parameters and frictional behavior, scientists seek to unveil the underlying mechanisms that enable prediction and control of friction -- an essential step toward carbon neutrality. Yet, reproducing frictional behavior in experiments is notoriously difficult. Here, we show that this challenge stems from the extreme sensitivity of tribological systems to tiny variations, e.g. in surface topography, typically presumed well-controlled. Even after meticulous surface preparation to semiconductor-industry standards and curtailing misalignment-induced oscillations, subtle variations remain and interact. In turn, such minute initial differences lead to statistically significant variations in friction and wear, giving rise to system-level chaotic behavior. Yet, by leveraging mid-scale features of surface topography and misalignment-induced oscillations -- information often filtered out or overlooked -- we established a model that accurately predicts high-friction regions under vastly different lubrication scenarios, with its performance further enhanced by machine learning.