Study On Rotor Faults Diagnosis Base On Matching Pursuit And Information Entropy

As an important part of mechanical equipment of industrial system in our country, rotor machinery has important economic and social benefit. This paper makes a deep research on several common faults of rotor machinery from the view of information entropy.It includes the following several aspects:Firstly, according to the operation characteristics of the rotor system, the rotor machinery several common fault of rotor machinery are simulated in the laboratory, for example imbalance, misalignment, touch grinding, oil film whirl, to analysis the mechanism of fault and make preparation for the further research.Secondly, based on the characteristics of the rotor vibration, as complex signal and redundant information, matching pursuit algorithm is introduced for the decomposition and denoise of vibration signal. The use of genetic algorithm and differential evolution algorithm to improve the matching tracking algorithm in computation and low efficiency, and achieved good effect. At the same time, Gabor atoms libraries are better adapted to the impact characteristics of rotor vibration signal. It is proved that the experiment achieves the result of a better signal denoise and information feature of signal is improved.Then, according to the theory of information entropy fusion, singular spectrum entropy and power spectrum entropy are selected to reflect in the characters of signal in both time domain and frequency domain. Information fusion entropy difference matrix theory is proposed based on it. It uses matrix obtain mean and method to realize the fault diagnosis respectively in steady speed mode and process mode. Finally, the experiment proved the method effectively.Finally, according to the extraction of two kinds of information entropy as a feature vector, support vector machine method is used for fault diagnosis, at the same time, two key parameters in support vector machine(SVM) is better optimized by particle swarm algorithm: penalty factor and kernel function parameter. the final experimental result reflects the superiority of the algorithm.The two fault diagnosis methods in this paper both achieve the final fault diagnosis, the fusion entropy matrix theory is easy to understand with simple and intuitive process and high reliability. while, support vector machine(SVM) method with high efficiency and less time consuming is suitable for a large number of computation and application system.