Physics-informed identification of dynamic propeller excitations via EM-based Kalman filtering

Accurate identification of propeller-induced hydrodynamic loads is crucial for vibration control and structural reliability in propeller propulsion systems. However, direct measurement of propeller excitation is infeasible due to the harsh operating environment and inherent structural constraints, which necessitates indirect identification approaches such as Kalman filtering. While these methods enable online estimation external inputs, the conventional random-walk assumption fails to represent the colored nature of hydrodynamic excitations, leading to spectrum mismatch and potential instability. To address this limitation, this study proposes an adaptive identification framework that integrates a scaled spherical simplex filter (S3F) with a physics-informed latent force model. The latent model combines an Ornstein–Uhlenbeck (OU) process with a single-degree-of-freedom (SDOF) system to simultaneously capture the broadband attenuation and hump-resonance features of propeller loads. Furthermore, to mitigate the difficulty of filter parameter tuning, an adaptive expectation–maximization (EM) strategy is incorporated to update noise covariances directly from structural response measurements, thereby enhancing robustness against uncertain prior settings. Numerical simulations and experimental validations confirm that, compared with the random-walk assumption, the proposed method significantly suppresses low-frequency drift, improves estimation accuracy, and maintains stability under varying noise conditions. These results highlight the method’s potential as a reliable strategy for indirect hydrodynamic load identification in propeller propulsion systems.