Researches On Rotating Machinery Fault Classification And Recognition Based On Compressive Sensing Theory

With the improvement of industrial automation and equipment manufacturing, rotating machinery plays an important role in modern industrial whose structure and function have become more and more complicated. Once a fault generates in the rotating machinery, the system may have to be shutdown, which will result in huge breakdown loss. Therefore, it is of great significance to adopt modern information technology and signal processing methods to monitor the operation status of rotating machinery, by detecting faults and taking response actions. Current fault diagnosis methods are often faced with massive monitoring data and fault feature extraction difficulties. In this thesis, on the basis of compressive sensing theory, researches on online monitoring, off-line analysis and sampling signal compression for bearings and gear system have been carried out to establish fault classification and recognition methods.The compressive sensing classification method based on redundant dictionary for the bearing fault was established for online monitoring in this thesis. Because of the poor sparsity in the traditional orthogonal transform basis for the rotating machinery vibration signal, a sparse representation method based on redundant dictionary is proposed and bearing fault sparse classification has been achieved via random dimension reduction and sparse solving. The influence of the bearing system rotating speed on fault classification accuracy has been discussed, the recognition performance against early weak faults of bearings has been analyzed, and the classification results comparison between the sparse representation classification method and the traditional pattern recognition methods has been explored.A fault classification method based on sparse dimension reduction and wavelet energy feature is established in this thesis to solve data storage and transmission problems in offline analysis. The classification of bearing fault has been achieved by wavelet energy feature extracted from sparse wavelet modulus maxima, unnecessary to reconstruct the original vibration signal. The classification accuracy of this method has been investigated with bearings and gear vibration signals under different operating conditions.Besides, researches on compressive sampling method of rotating machinery vibration signal has also been carried out and a block sparse Bayesian learning based compressive sensing method was proposed. Accurate reconstruction of the compressed signal can be acquired via Bayesian probability estimation, without considering the sparsity of the vibration signal. The performance of data compression and reconstruction has been investigated with the feature extraction of bearing signals under different conditions, verifying the effectiveness of this method.Based on the compressive sensing theory, the methods proposed in this thesis can realize rotating machine fault classification and vibration signals compression, supporting the online monitoring, off-line analysis and sampling signal compression of rotating machinery effectively.