Orbit-based identification of unbalance in overhung rotors using physics-informed machine learning

This paper presents a physics-informed orbit-tracking approach in which synthetic steady-state data from a reduced-order rotor model are used to train supervised algorithms to estimate the unbalance magnitude and angular position of overhung rotors under steady-state conditions. The model represents the rotor as a rigid body supported by flexible, anisotropic, and cross-coupled bearings, and is validated against an independent Simulink/Simscape model. The analysis explores the effects of stiffness anisotropy, cross-coupling, damping, and unbalance magnitude on orbit geometry. Results demonstrate the reconstruction of reference orbits with low amplitude and phase errors, confirming the effectiveness of the proposed orbit-based physics-informed machine-learning approach and supporting its future extension to simulation-to-real implementation using experimental data and dynamic unbalance conditions.