Vibration fault detection and diagnosis in aircraft power plant using model-based technique

To assure successful operation over a long period of time, the aeroengine requires a certain degree of maintenance. To achieve this object, the automated condition monitoring system which can early detect potentially catastrophic faults is needed. Therefore, the vibration signal analysis and fault pattern recognition become important issues. A novel approach combining wavelet transform with fuzzy theory is proposed to complete feature extraction fault mode recognition. The wavelet transform uses a rich library of redundant bases with arbitrary time-frequency resolution which enables the features extraction from aeroengine vibration signal. The neural-fuzzy network is used for fault recognition purposes. The improved algorithm is used to complete the network parameters determination and the robustness of neural network is discussed. By means of network training phase, each fault mode of training set is represented by a certain number of codewords and the trained wavelet-fuzzy network is utilized to detect and classify vibration fault of aeroengine. Finally, the fault pattern recognition is accompanied by a belief degree that is introduced as estimations to the recognition accuracy. The proposed solution has been validated through experiment and diagnosis result.