Non-Linear Friction Force Estimation for Ball and Beam Mechanism Using R-PINN

• Published in: 2025 IEEE Symposium on Computational Intelligence on Engineering/Cyber Physical Systems (CIES) pp. 1 - 6
• Main Authors: Kaya, Ozan, Ertugrul, Seniz, Abedinifar, Masoud, Egeland, Olav
• Format: Conference Proceeding
• Language: English
• Published: IEEE 17.03.2025
• Subjects: Computational modeling; Data models; Dynamics; Friction; Friction models; Mathematical models; Noise measurement; Physics-Informed Neural Network; Predictive models; Recurrent Neural Network; Recurrent neural networks; System performance; Training data
• DOI: 10.1109/CIES64955.2025.11007630

Different friction forces or torques are affecting the system's performance and control. Friction forces occur due to bearings, gearboxes, or any other contacts in the system. Researchers have reported different forms of friction, such as stiction, viscous and Stribeck effects, pre-sliding displacement, stick-slip effects, hysteresis (or frictional lag), etc. Developing a mathematical model to describe the underlying dynamics of a complex system may become necessary to design either a modelbased controller or at least compensate for the non-linear effects of friction forces. For this reason, either test set-ups or datadriven techniques might be used. In this study, the RecurrentPhysics Informed Neural Network is studied to determine the friction forces and model the Ball and beam system. While PINN provides faster results to model non-linear systems with noisy and small data sizes, Recurrent Neural Network architecture is fruitful for modeling time-dependent systems. Thus, R-PINN is trained with noisy signals for system response and friction model of the ball and beam system. Despite noisy signals and nonlinearity in the system, R-PINN is promising in modeling the system response and estimating the friction model.