Application of a Balancing Filter for Model-Based Fault Diagnosis on a Centrifugal Pump in Active Magnetic Bearings

This paper discusses the suitability of a special discrete filter, called balancing filter, to improve the performance of model-based fault detection and fault diagnosis on a centrifugal pump in active magnetic bearings. The focus in this subject lies on the extraction of better symptoms for the fault diagnosis. The application of the balancing filter sets up on a multi-model approach which uses a model of the system for the reference state and every fault that is to be detected. These models are stimulated with the same test signals as the ones applied to the process while it is running. To compare the simulation results of the models with the process response the output error is calculated. After this the remaining residuals are used as symptoms for the fault detection. The balancing filter is used to remove the large differences within the amplitude responses of the models caused by the lowpass characteristics of the mechanical part of the system. Hence the influence of the smaller differences caused by the examined faults is weighted equally at all interesting frequencies. This leads to new residuals which are separated more clearly. This approach is used to detect common faults appearing on centrifugal pumps as dry run, incorrect installation and worn out balance pistons. The test rig used to examine the suitability of the proposed filter is a one-level centrifugal pump in magnetic bearings. The rotor of the pump is driven by an asynchronous motor at rotation speeds up to 3000 rpm. The first flexible mode of the rotor is located at 280 Hz. In the seal gap fluid-structure-interaction is appearing. The forces on the rotor are calculated based on the current applied to the bearings, while its displacement is measured by eddy current sensors integrated into the bearings. The first two natural frequencies of the system are located at about 200Hz and 500 Hz. These frequencies are shifted when a fault is occuring. In the models for the fault states this behaviour is represented. Hence the model matching the current state of the pump leads to the lowest residual. The advantage of the balancing filter is that the detection of faults becomes more reliable. Below the examined faults, the model-based concept and the design of the balancing filter are described in detail. Results from experiments on the test rig are given to show the advantages of the balancing filter.