Feature Extraction Based on Adaptive Multiwavelets and LTSA for Rotating Machinery Fault Diagnosis

Feature extraction is a key procedure in the fault diagnosis of rotating machinery. To obtain fault features with lower dimensionality and higher sensitivity, a feature extraction method based on adaptive multiwavelets transform (AMWT) and local tangent space alignment (LTSA) is proposed in this paper. AMWT is first used to obtain multiple features from the vibration signals of the machine under test to form a high-dimensional feature set. Then, in order to avoid the adverse effect of the irrelevant features in this high-dimensional feature set on the fault diagnosis result, a detection index (DI) is investigated to evaluate the sensitivity of the features and those with lower sensitivity are removed. After that, LTSA is applied for feature fusion to reduce the redundant features in the high-dimensional feature set. To validate the proposed method, performance of four feature extraction schemes based on (i) wavelet and LTSA, (ii) Geronimo, Hardin, and Massopust (GHM) multiwavelets and LTSA, (iii) AMWT and principal component analysis (PCA), and (iv) AMWT and multidimensional scaling (MDS) is compared with the proposed method. The feature extraction results by these methods are then fed into K-medoids classifier to discriminate the faults. The results show that the proposed method can improve the sensitivity of the extracted features and obtain higher fault recognition rate.