Integrated Fault Diagnosis Method Of Gear Transmission System Based On Hibrid Intelligent

Along with the development of modern equipment, the mechanical equipment is being developed toward high efficiency, high intensity, high performance, large scale and automation. Gear transmission system which transmits movement and power, plays an important role in large-scale equipment. However, it is vulnerable to be damaged and appear faults because of the complexity of load conditions, poor working conditions and other reasons, and these faults are so likely to induce the breakdown of machine as to bring about incalculable economic losses and social influence. So, research on the theories and technologies of faults diagnosis to gear transmission system will have the theoretical and practical significance.Although the existing methods of feature extraction and fault recognition technology have gained many achievements, there are still many imperfections: Power spectral analysis is applicable to analysis for steady signals. But it's not suitable to the vibration signals acquired by gear transmission system owing to the existence of non-steady signals; the non-stationary signal processing algorithms, such as short-time Fourier transform, wavelet transform, EMD decomposition and so on, can adapt to non-stationary characteristics of signals of fault diagnosis in gear. However, it contains so many frequency-band features in extracting characteristics as to increase the difficulty of in fault diagnosis. It is urgent how to reduce the feature dimensions and ensure the effect of fault diagnosis. The pattern recognition method based on feature extraction is an important aspect of fault diagnosis, BP neural network, with its ability of strongly self-learning and non-linear identification is researched widely in fault identification, but it's critical that this method is apt to fall into local optimum. At the same time, it is an imperative topic in research on fault diagnosis how to apply the achievement in feature extraction algorithm to improve the recognition rate of faults.The study methods of combining the physical simulation and laboratory testing with theoretical analysis and simulation are applied in the paper. From the signal preprocessing and feature extraction to pattern recognition and fault information fusion, the intelligent methods, such as the rough set theory, biological evolution and neural network, are integrated organically and play their respective algorithm advantages so as to form a route of the integrated fault diagnosis technology which is a rough set - immune genetic - to improved BP neural network in gear system. In accordance with the merits and demerits of feature extraction techniques, the characteristics reduction methods of the power spectral analysis, short-time Fourier transform, wavelet packet decomposition algorithm and improved EMD are proposed; In the study of fault pattern recognition, according to the characteristics of BP algorithm, the method of hybrid search algorithm and change on error improved the recognition rate of faults using the immune genetic algorithm of forward multi-layer neural network by comparison with 4 second function value-- BP algorithm weight. The three parameters: quality factor parameters, fault recognition rate and uncertainties, are obtained by the failure decision-making information fusion as the basis for gear fault identification have improve the accuracy of fault diagnosis identification in gear.After the cause mechanism and respective characteristics of four faults, such as rolling bearing inner ring failure, the outer ring failure, gear wear and broken teeth, their causes, mechanism and respective characteristics were analyzed, the designed physical simulation system laid the foundation for the implementation of this program as a result of optimizing the sensor configurations. The failure data is preprocessed for de-noising of signal by wavelet analysis. The results show that wavelet analysis can be used to solve the de-noising problem of gear fault signal, and provide the follow-up analysis with facilitate.In the light of edge effects of EMD, the envelope continuation method based on gray prediction to solve the problem of edge effects in empirical mode decomposition is put forward in the paper. A number of typical characteristics of gear failure are get applying the improved EMD. The simulation and experimental data analysis results showed that the method is effective in solving the problem of the edge effects.After spectral analysis, short-time Fourier transform, wavelet packet decomposition and improved intelligence EMD decomposition to denoising fault vibration signals of gear transmission system , features of all kinds of faults have been acquired by four extraction methods. Research and analysis show that spectral analysis is applied to stationary signals and that it may lead to wrong conclusions to determine which faults belong to by characteristics parameters based on the frequency spectrum considering that gear vibration signals of gear transmission system are complicated and contain non-stationary signals. There is uncertainty of the broken tooth fault feature extraction using short-time Fourier transform; there are the obvious characteristics of the fault in the gear wear as well as broken gear tooth using wavelet packet analysis. However, it is easily to lead to wrong results of fault identification by wavelet packet analysis on account of less difference in energy distribution of faults between the inner and outer ring. Right conclusion could be drawn in feature extraction of gear wear, broken teeth, and inner bearing fault using intelligent EMD feature extraction method. Unfortunately, we are difficult to draw accurate conclusions in bearing fault feature extraction of outer ring with regard to confused characteristics. Studies have shown that single fault feature extraction methods are effective for certain faults and difficult to provide accurate fault characteristics for others. So, this will have an impact on accuracy of fault diagnosis conclusion.Studies of fault Diagnosis of transmission system have been made adopting Rough Sets - Immune Genetic - Improved BP Neural Network Algorithm. Firstly, the fault characteristics reduction of gear transmission system has been done to remove redundancy of frequency-domain characteristics and to reduce the calculation workload of fault identification using attribute back reduction algorithm of rough set. Secondly, in order to effectively improve the overall identification capacity of fault diagnosis network, neural network weight optimization of fault diagnosis has been carried out respectively adopting immune genetic algorithm and Improved BP algorithm according to different judge criterions based on 4 times function value. Consequently, identification results of bearing faults have showed that the algorithm is better applying the fault characteristics acquired by spectral analysis, short-time Fourier transform, wavelet packet decomposition and improved EMD decomposition.Judging from the results using intelligent integrated hybrid algorithm, the four types of gear faults could be basically separated .But differences in the credibility of different fault type identification are not obvious. Once there exists noise interference according to the index value of credibility applying the algorithm for fault diagnosis, it is possible to lead to misjudge to faults.The method integration applying DS evidence theory to fault characteristics extraction has been put forward in the paper. Furthermore, fault identification rate supported by quality factor parameters and integrated fault diagnosis effect measured by uncertainty have been proposed. Results showed that the fault recognition rate was raised and the uncertainty of fault identification was reduced caused by increase of quality factor owing to adopting extraction technology integration for fault characteristics. This could overcome the defect using extraction methods of single fault feature and effectively improve accuracy and reliability of fault diagnosis conclusions.