Real-Time Prediction of Pressure and Film Height Distribution in Plain Bearings Using Physics-Informed Neural Networks (PINNs)

The increasing application of plain bearings in various industries, especially under challenging conditions like thin lubricating films and high temperatures, necessitates effective monitoring to prevent failures and ensure reliable performance. While sensor-based monitoring incurs significant costs and complex installation due to physical sensors and data acquisition systems, model-based tracking offers a more cost-effective alternative. Model-based monitoring relies on mathematical or physics-based models to estimate system behaviour, reducing the need for extensive sensor data. However, reliable results depend on real-time capable and precise simulation models. Conventional real-time modelling techniques, including analytical calculations, empirical formulas, and data-driven methods, exhibit significant limitations in real-world applications. Analytical methods often have a restricted range of applicability and do not match the accuracy of numerical methods. Meanwhile, data-driven approaches rely heavily on the quality and quantity of training data and are inherently constrained to their training domain. Recently, Physics-Informed Neural Networks (PINNs) have emerged as a promising solution for model-based monitoring to capture complex system behaviour. This approach combines physical modelling with data-driven learning, allowing for better generalisation beyond the training domain while reducing reliance on extensive data. Thus, this study presents an approach for load monitoring in radial plain bearings using PINNs. It extends the application of PINNs by relying solely on simple sensor inputs, such as radial load and rotational speed, to predict the hydrodynamic pressure and oil film thickness distribution under varying stationary conditions. The real-time model is trained, validated, and evaluated within and beyond the training domain using elastohydrodynamic simulation results. The developed real-time model enables load monitoring in plain bearings by identifying critical hydrodynamic pressure and oil film thickness values using readily available speed and load sensor data under varying stationary conditions.