Sensor fault detection for uninterruptible power supply (UPS) control system using fast fuzzy-neural network and immune network

In power electronic systems, many researchers have been investigating troubles caused by a sensor failure. Sensorless vector control of induction motor drives has attracted researchers´ attention for a long time. The paper describes a sensor fault detection method for UPS current and voltage feedback systems. Once a certain sensor fails, then its influence propagates through the whole system and may cause a fatal situation. It is usually difficult to identify a failed sensor by observing other sensors´ outputs. The proposed detection method uses a fast fuzzy neural network and an immune network. The fast fuzzy neural network roughly but very quickly calculates the failure rate of each sensor. The immune network is decomposed into a decision tree structure, which has only the forward passes in parallel. The density of each antibody, called failure origin ratio, is calculated by a nonlinear differential equation driven by stimulation, suppression, failure rate and dispassion. The sensor that shows the highest failure origin ratio is considered as the failed sensor. The proposed method is applicable to fault diagnosis for large-scale and complex systems such as multi-UPSs operated in parallel