Rolling Bearing Faults Diagnosis Based On The Improved Wavelet Neural Network

Rolling bearings are important parts in mechanical equipment, is one of the most easily damaged components. When the rolling bearing fault is easy to make mechanical equipment generating noise and vibration abnormality, further damage the equipment. For the early detection of equipment failure to prevent production line shutdown, and avoid serious accidents, development has very practical significance for the fault diagnosis of rolling bearing. This paper mainly studies on the application of wavelet neural network in fault diagnosis of rolling bearing. The thesis mainly study the following three aspects:Part1.Discusses the analysis and extraction method of bearing fault mechanism and fault characteristic parameters. For the non-stationary characteristics of the nonlinear system of rolling bearing surface vibration signals, the introduction of wavelet analysis method and the wavelet analysis is easy to produce frequency aliasing puts forward an improved wavelet packet algorithm. Improved wavelet analysis for frequency aliasing phenomenon has a good performance, to overcome the traditional wavelet packet algorithm in high and low frequency overlapping indistinguishable and analysis technique using wavelet frequency band, and separation of the noise signal containing the fault signal.Part2.Combined with the advantage of wavelet and neural network are the structural model of the improved wavelet neural network, the learning algorithm of wavelet neural networks, the convergence speed of traditional BP algorithm is slow and easy to fall into local mini problems, from two aspects of the learning rate and connection weights to improve algorithm.Part3.Application of improved wavelet network to typical faults diagnosis of rolling bearing. N205type rolling bearing on the test bench. The experiment data of the training network, using vibration signal input vector for the network, results are given for training. Through the simulation example, we can see the improved wavelet neural network can well classify faults, the convergence speed of BP network is obviously faster than the same condition of wavelet neural network and improved, effectively realized the fault diagnosis of rolling bearing.