Fault Diagnosis Method Based Semi-supervised Manifold Learning and Transductive SVM

Abstract

A fault diagnosis method was proposed based on Semi-supervised manifold learning and Transductive support vector machine (TSVM), to overcome scarcity of labeled training samples. Firstly, wavelet packet decomposition (WPD) was used to decompose vibration signals into several sub-bands. The fault features were extracted from the sub-bands to construct a high-dimensional fault feature set, and the improved kernel space distance evaluation method was used to extract the sensitive fault features. Then, orthogonal semi-supervised linear local tangent space alignment (OSSLLTSA), which is a semi-supervised manifold learning method, was proposed to reduce dimension of the fault feature set, and extract fusion features with low dimension and high clustering performance. OSSLLTSA is able to extract the structural information in both labeled and unlabelled samples, and so OSSLLTSA overcomes over-learning problem of supervised manifold learning. Meanwhile, OSSLLTSA also avoids the drawback of weak pertinence of unsupervised manifold learning. Finally, the extracted low dimensional feature set was input into TSVM for fault diagnosis. TSVM is able to completely utilize the fault information contained in unlabelled samples to modify the model, and as a result the trained fault diagnosis model has better generalization ability. Through a gearbox fault instance, the effectiveness of the proposed method was verified. Experimental results showed that the proposed method is able to achieve very high fault diagnosis accuracy even when labeled samples are insufficient.