Application Of Wavelet Neural Network To Gearbox Bearings Fault Diagnosis

Rolling bearings are one of widely used mechanical parts in rotating machines and vulnerable to damage. Many faults of rotating mechanism are related to rolling bearings. The performance of rolling bearings directly affect the performance of axis,gear and whole equipment. The defectiveness of rolling bearings can result in abnormal vibration and noise of equipment, even seriously damage to the equipment and catastrophic accident. Thus, developing fault diagnosis of rolling bearings has great practical significance.In recent years, the technique of mechanical fault diagnosis, which plays an important role in national manufacture, is quickly developed in domestic and oversea, and is deeply implied in many fields. On the basis of former achievements, wavelet neural network method is used in gearbox rolling bearings, and valid results are obtained in this paper. The major works of this thesis are listed as follows:This thesis focuses on gearbox rolling bearings, according to gearbox rolling bearings' structures and vibration, the cause of gearbox rolling bearings' fault and characteristic frequency are analyzed. After studying the application method of the wavelet analysis and the neural network in the gearbox rolling bearing fault diagnosis, Signal processing and fault diagnosis system that is with the help of wavelet packet to single de-noising and with the assistance of RBF algorithm to be the pattern recognition is built.Then, the wavelet packet de-noising of vibration signal of rolling bearing is studied in the two modes of normal bearings and outer ring fault; Extract signal characteristics by using the method of "wavelet packet-energy" as neural network input vector. Correct rate of pattern recognition is analyzed by using RBF neural network learning algorithm in the two modes of normal rolling bearings and rolling bearings outer fault and the web that's of training and pattern recognition is identified.Through simulation and experimental results show that: after de-noising of wavelet packet analysis to weak fault signal of gearbox rolling bearings ,can significantly improve the signal-noise ratio; The wavelet analysis in combination with the RBF neural network for fault diagnosis of rolling bearings, can accurately identify anomalies of the work of running of gearbox rolling bearings; Using decomposition signal of frequency domain of wavelet packet that structures energy characteristic vector, accurately reflects frequency domain energy of the gearbox rolling bearing vibration signal which changes with the change of the status information; Using energy characteristic vector, wavelet neural network can achieve non - linear mapping with the accuracy of more than 90% from rolling bearing vibration signals space to the rolling bearing fault status space.