Sliding bearings are critical to rotating machinery with lubrication performance governing friction and lifespan. Traditional numerical methods face efficiency-accuracy trade-offs dependent on mesh density and iteration. The emerging paradigm of tribo-informatics offers a promising alternative by integrating data-driven modeling with physical knowledge. This study develops a tribo-informatics framework to simultaneously predict the coupled pressure and film thickness distributions in sliding bearings, enabling rapid and accurate evaluation of lubrication regimes and frictional characteristics across a range of operating speeds and loads. A physics-informed neural network (PINN) is leveraged with the film geometry and pressure boundary conditions as hard constraints, while its loss function integrates two fundamental physical laws: a local differential constraint derived from the average Reynolds equation, and a global integral constraint based on the load-balance equation that incorporates both hydrodynamic and asperity contact contributions. To enhance the generalization capability for varied operating conditions, a mini-batch cumulative training strategy is implemented for multi-condition learning. A water-lubricated bearing is employed as a case study to verify the consistence of the proposed method and the finite difference method. Results demonstrate that the proposed method can simulate hydrodynamic and mixed lubrication with high precision. Specifically, the bearing’s eccentricity ratio and coefficient of friction (COF) are predicted with nearly zero error, and other tribological metrics are predicted with an average error of no more than 3.48 %. This study provides an efficient and scalable computational tool for bearing performance analysis and contributes to the data-model co-driven design and optimization of tribological systems.
扫码关注我们
求助内容:
应助结果提醒方式:
