Magnetic bearings now exist in a variety of industrial applications. However, there are still concerns over the control integrity of rotor/magnetic bearing systems and the ability of control systems to cope with possible faults that can occur during operation. Unless control systems can be developed that have the ability to maintain safe operation when the system is in a degraded or faulty state, then many, otherwise viable, magnetic bearing applications will remain unfulfilled. In this paper, a method is proposed for the design of a fault-tolerant control system that can detect and identify both incipient and sudden faults as and when they occur. A multivariable H∞ controller is reconfigured on occurrence of a fault so that stability and performance is maintained. A neural network is trained to identify faults associated with the system position transducer measurements so that the output from the neural network can be used as the decision tool for reconfiguring control. In this way, satisfactory control of the system can be maintained during failure of a control input. The method requires no knowledge of the system dynamics or system disturbances, and the network can be trained on-line. The validity of this method is demonstrated experimentally for various modes of sensor failure.
Output-feedback ℋ2/ℋ∞ control of a class of networked fault tolerant control systems