Model-Based Fault Detection on Active Magnetic Bearings by Means of Online Transfer-Factor Estimation

This paper gives a deeper insight in model-based fault detection based on transfer-factor esti-mation on active magnetic bearings. The method aims at performing fault detection and diagnosis at minimal computational efforts. Hence, only a limited number of transfer-factors of the inves-tigated system are considered. These are used as features for the diagnosis and represented by the magnitude and the phase at frequencies showing strong deviations between the examined states. Previously modeled faults are diagnosed by an online-estimation of the transfer-factors and a comparison to reference values. Three algorithms for this fault detection method are de-scribed and compared in detail: Recursive Least Squares (RLS), Least Mean Squares (LMS) and Goertzel-algorithm. For the comparison of the algorithms, their convergence behavior and updat-ing rate are considered as well as the computational and implementation effort. Investigations on a test rig of a centrifugal pump in active magnetic bearings show that the RLS leads to slightly slower convergence with a higher computational effort than the LMS. The Goertzel-algorithm outperforms both in computational effort, but needs several samples for the computation and thus provides a lower updating rate. Based on their findings, the authors assess suitability of the algorithms for fault detection on active magnetic bearings.