Bearing Intelligent Fault Diagnosis Robust To Fault Severity And Operating Regime

Rolling bearings are one of the most critical components that determine the machinery health and widely used in rotating machinery systems. Rolling bearings are vulnerable to fault appear with a great incidence due to its complexity and poor working conditions which in turn make fault diagnosis difficult. In addition as machinery is becoming increasingly large and automatic, the measured data are too many to be analyzed manually. In order to reduce manual intervention and human subjectivity, signals are automatically analyzed so as to implement intelligent diagnosis. Therefore intelligent diagnosis of rolling bearing is of great importance to the rotating machinery.The intelligent fault diagnosis of machinery mainly consists of three steps: the acquisition of vibration signals, feature extraction, and pattern classification. The collection of vibration signals is the cornerstones of intelligent diagnosis in the aforementioned steps. On the other hand, feature extraction is critical for intelligent diagnosis which will directly affect the accuracy and reliability of pattern classification while pattern classification gives the diagnosis results. To classify fault pattern is the result of intelligent diagnosis. Taking rolling bearings as the diagnostic object, this research investigate following intelligent diagnosis philosophy:(1) In order to describe the complexity of non-linear and non-stationary vibration signals emanated from rolling bearings in a multiple scale framework, the Tsallis entropy and SampEn are combined with wavelet packet transform(WPT) and lifting wavelet packet transform(LWPT) respecitively. Probabilistc neural networks(PNN) and RBF neural networks are used to construct two kinds of of intelligent fault diagnosis methods in order to simultaneously determine bearing fault type and severity level. Experimental data were collected from QPZZ-Ⅱ fault simulation test rig involving ten different states of rolling bearing. The results verify the effectiveness of the proposed two kinds of intelligent fault diagnosis methods based on the paradigm of entropy.(2) Bearing intelligent diagnosis is approached by means of multiscale entropy(MSE) and binary tree structure based classifier ensemble. A multi-classifier fusion algorithm is presented using the form of binary tree, which employs three different neural networks and majority voting scheme(MVS) fusion to transform the multi-classification problem into a series of binary classification problems at the tree nodes. Experimental results demonstrate the proposed paradigm can effectively improve the recognition accuracy and stability of rolling bearing fault diagnosis in comparison with the diagnosis method based on a single classifier.(3) Intelligent diagnosis methods which are robust to fault severity and operating regime are presented. In real engineering applications, although testing and training samples are of the same fault type(like pitting on rolling elements), they are usually different in terms of fault size and operating regime. This difference is likely to deteriorate the performance of intelligent fault diagnosis methods available in the sense that the feature extraction and pattern classification will become invalid. In order to tackle such a problem, the vibration signals of rolling bearings were firstly transformed into time-frequency space by the S transform, and then the singular value decomposition(SVD) was utilized to extract feature vectors from transformed matrices. The feature vectors were fed to support vector machines(SVMs) to simultaneously judge bearing fault type. The results verify the effectiveness of the proposed approach and this method is better than the results of traditional methods for the practical problems to fault severity and operating regime robust.