Model Based Fault-Detection and -Diagnosis using Active Magnetic Bearings

This paper shows how model based fault detection and diagnosis can be integrated into the active magnetic bearing system. It describes two appropriate fault-detection methods on the example of centrifugal pumps in magnetic bearings and shows how typical fault states occurring on these pumps can be detected and diagnosed. Prior to the fault detection the modeling of the magnetic bearing system is described. The investigated multi model method contains a bank of models representing the systems transfer behavior for the different fault. With this method the error between the outputs of the models and the output of the plant are provided as features for the fault diagnosis. Abalancing filter is described helping to separate the different features more clearly. Instead of computing the complete frequency behavior of the plant the transfer factor method uses the Goertzel algorithm to compute only significant discrete frequency points and provides the complex transfer factor as feature. It is shown that fault detection and diagnosis could be integrated within AMB systems. Both methods are well suited to provide reliable information about the system state.