A novel feature extraction and optimisation method for fault diagnosis by combining WPT with FCM

作者：Lu, Wenqing^*; Zhang, Laibin; Liang, Wei; Yu, Xuchao

来源：Insight: Non-Destructive Testing and Condition Monitoring , 2016, 58(1): 35-41.

DOI：10.1784/insi.2016.58.1.35

• feature extraction and optimisation
• fault diagnosis
• wavelet packet transform
• fuzzy c-means
• support vector machine

In the process of fault diagnosis, features of the vibration signals have to be extracted and optimised before fault pattern recognition. Wavelet packet transform (WPT) is one of most widely used methods in feature extraction for fault diagnosis. According to a number of previous studies, a vibration signal can be decomposed into several nodes (sub-bands) by WPT, and then Shannon entropies of all the nodes are calculated to compose a vector, which is taken as the feature of the vibration signal in the subsequent processing. However, not all the elements included in the feature vector contain useful information for fault diagnosis. Without optimisation, the computation quantity of fault pattern recognition would be increased and the accuracy of fault diagnosis would be decreased. Therefore, a novel feature extraction and optimisation method called WPT-FCM, which combines WPT with fuzzy c-means (FCM), is proposed to solve the problem in this paper. FCM is employed to identify and remove the useless elements in the feature vector and a new feature vector can be obtained after optimisation. Next, a support vector machine (SVM) is trained by new feature vectors of known fault samples and then unknown faults can be identified accordingly. A case study of rotor fault diagnosis conducted on a rotor model system (RMS) experiment platform has been performed to validate the feasibility and superiority of the proposed method. The result of experimental verification shows that the accuracy of fault diagnosis by WPT-FCM has been improved significantly compared with WPT- and PCA-based methods.

• 出版日期2016-1
• 单位中国石油大学（北京）