Application of artificial neural networks for the fault detection and diagnosis of active magnetic bearings

Abstract

Active magnetic bearings (AMBs) are the class of advanced mechatronic systems. The stable operation of these depends on the normal operation of its sensors and actuators whose malfunctioning may disturb the stability of the supported rotor. Therefore, online fault detection and diagnosis (FDD) of sensors and actuators of AMB system is essential for safe and reliable operation. Model based FDD requires complex mathematical modelling and has higher chances of subjected to modelling errors. Redundant sensors and actuators based FDD incurs additional cost and also requires additional space for installation. Therefore, in the present work, simulation data driven artificial neural network (ANN) based methodology with statistical analysis is proposed for FDD of AMBs. Faults in single position-sensor or actuator as well as in multiple sensors and actuators are detected and diagnosed simultaneously. Various types of faults such as bias, multiplicative and noise addition are considered for the diagnosis.