lures, and accurately predicting the wear between
friction pairs is essential for equipment health management.
As diesel engines progressively develop toward higher explosion
pressures and power densities, the working environments of key
components such as crankshaft bearings become increasingly
harsh, making them more prone to wear issues that com
promise engine reliability. Most existing data-driven methods
primarily focus on wear modeling using sparse data, yet such
approaches exhibit limitations in interpretability. To address
this challenge, this paper proposes a physics-informed neural
network approach by integrating the Archard wear law and
typical wear processes as physical constraints into the neural
network framework, thereby enhancing model interpretability.
Experimental results demonstrate that incorporating prior
physical knowledge improves the overall performance of the
neural network. Furthermore, parameter sensitivity analysis
reveals that the model maintains behavior consistent with
typical wear processes even when applied to unseen data.