Fault detection of high speed turbopump via vibration signal analysis

In this paper, the main causes of occurred faults in the fuel turbopump of a rocket engine are first discussed. According to these faults, the features for fault detection are extracted from the power spectrum density of vibration acceleration signal measured from the turbopump case. Neural networks are used to detect faults with a BP neural network acting as fault detector and an unsupervised clustering neural network being used to obtain patterns required for training the fault detector from the original feature vectors of test data. The validation result of the vibration signal in ground firing test df engine indicates the feasibility and effectiveness of this method.