Study On Pattern Recognition Method For Fault Of Rotating Machines

With the increasingly large scale, high speed, automation, making the composition and structure of machinery equipment more complex, the links between the various parts more closely. Very high demand is raised to the reliability of machinery equipment and its each component. How to Increase automation and intelligent of fault diagnosis has become the hot research field. Pattern recognition is one of the key element affecting results of fault diagnosis, so the research of fault pattern recognition has theory value and practical significance.Fault pattern recognition based on the feature data extracted is studied in the dissertation. For some problems in practice, such as small fault samples, many fault types and more complex fault features, we research the application of rough set, support vector machines, and neural networks in rotating machinery fault pattern recognition. Main research work is as follows:1) In the pattern recognition methods based on rough set, the continiues feature data need to be discreted. The Self-organizing feature map neural network is one of the most commonly used discretization method, but it requires a prior determination of the number of clusters, which is often difficult because of the complexity of practical engineering objects. In this paper, we propose a norvel self-organizing feature map neural network to optimize the cluster mnumbers for discreting the feature data, by means of which the number of clusters can be determined and optimized from the extracted feature data automatically.2) The classical rough set method has the shortcoming of a lower anti-interference ability and poor generalization ability, that often causes a wrong diagnosis result. We analysis the reasons of the classical rough set approach and propose to use the variable precision rough set based pattern classification method to increase the anti-interference ability and generalization ability. In the proposed method the threshold parameters of the classification model are determined by using the information from the extracted fearture data.3) The least squares support vector machines (LSSVM) is often used in the area of the multi-fault pattern recognition with small sample. However, the selection of the parameters in the model of the LSSVM depends on the user's experience and will greatly affect the accuracy of fault identification. We propose a improved least squares support vector machine method based on parameter optimization to solve this problem. In this method the threshold parameters of the classification model are determined and optimized automatically by using the information from the extracted fearture data. Two cases of fault diagnosis for a gearbox and a element bearing with different fault types demonstrate, the accuracy of the fault classification are improved.4) In fault pattern recognition, the using of excessive number of fault features will produce redundant data, which makes the classifier complexity, and led to the classification results deteriorate. In this paper, least squares support vector machine fault rough pattern recognition is proposed. We choose effective fault features by the reduction ability of the rough set and then improve support vector machine by the parameters optimization for fault recognition. An example demonstrates that the method has a strong anti-interference ability and generalization ability.