A new method based on adaptive treelets transform for fault diagnosis of rolling bearing

Features extracted from rolling bearing signal usually contain a great deal of noise and other redundant information which may reduce the efficiency of fault diagnosis. In this paper, a new method based on adaptive treelets transform is proposed for fault diagnosis of rolling bearing, which is proved to be an effective method. Treelets transform is a local multi-scale analysis method, and it is especially well-suited for high-dimensional and unordered signal with noise. Unlike the basic theory of treelets, this paper defines the similarity matrix from the point view of graph-theoretic based on geodesic distance, which can easily capture the intrinsic structure of original vibration signal of bearing. Because of the bearing signal collected by accelerometers cannot fully describe fault information, this paper select initial feature from the phase space. Treeles decomposition is then used for further feature extraction, which can not only reduce the disturbance of noise but also achieve the effect of dimension reduction. Finally, features are input to the model of SVM, the experimental results indicate that this proposed method can accurately identify different fault states of rolling bearing only by one-dimensional feature, and it is very robust to noise.