Predicting static friction coefficients under heavy loads using machine learning algorithms

Fastening structures in vehicles that endure repetitive shear loads must maintain sufficient clamping forces to prevent loosening caused by joint slippage. The minimum clamping force required for controlling slippage is calculated using analytical and theoretical methods and is closely related to the static friction coefficient between the joint materials. In this study, we introduce a novel test apparatus designed to measure the static friction coefficient between two materials under high load conditions, with its experimental suitability confirmed through reliability verification. We experimentally analyzed the effects of the normal load, surface roughness, and mechanical properties on the static friction coefficient for materials commonly used in vehicle joints, including coated steel, steel, and aluminum alloys. Four machine learning algorithms—Gaussian process regression (GPR), ensemble, artificial neural network (ANN), and support vector regression (SVR)—were evaluated to develop a prediction model for the static friction coefficient. The predictive performance of each model was assessed using various evaluation metrics, and the results revealed that the GPR model achieved higher accuracy in predicting the static friction coefficient than did the other models.