Physics-Informed Neural Network Framework for Predicting Friction and Wear Evolution in Lubricated Tribological Contacts under Variable Operating Conditions

Reliable prediction of friction and wear in lubricated sliding contacts is essential for improving the durability and efficiency of mechanical systems. This study develops a coupled Physics-Informed Neural Network (PINN) framework for predicting friction coefficient and wear evolution in an AISI 52100 bearing-steel tribopair lubricated with polyalphaolefin (PAO) oil under variable operating conditions. The computational investigation considers contact pressures of 0.5–1.5 GPa, sliding velocities of 0.05–1.0 m/s, PAO lubricant viscosities of 20–100 mPa·s, surface roughness values of 0.05–0.50 μm, and material hardness values of 600–800 HV. The proposed framework treats friction and wear as coupled tribological responses rather than independent prediction tasks. Hertzian contact mechanics, Archard’s wear law, and lubricant film-thickness relationships are incorporated into the PINN training objective to impose physically meaningful constraints on the predicted response. The model is trained and evaluated across systematically varied combinations of contact pressure, sliding velocity, lubricant viscosity, surface roughness, and hardness, with particular emphasis on generalization to operating conditions that are sparsely represented in the training data. A sensitivity analysis is performed to quantify the relative contribution of the investigated parameters to frictional response and wear progression and to identify conditions associated with increased wear and deterioration of lubrication performance. The physics-informed formulation is designed to improve physical consistency and reduce dependence on large datasets compared with purely data-driven approaches. By integrating contact mechanics, lubricant-film behaviour, and wear evolution within a unified learning framework, the proposed methodology provides an interpretable computational approach for coupled friction–wear prediction. The framework offers potential applications in lubricant selection, wear-rate estimation, tribological condition assessment, and reliability-oriented maintenance of lubricated mechanical systems.