Rolling Bearing Fault Diagnosis System Based On Wavelet Transform And Neural Network

Rolling Bearing Fault Diagnosis has gradually developed into a discipline in order to meet the needs of industrial production, and it is a practical technology to maintain mechanical device running smoothly. Wavelet transform and neural network’s theoretical system increasingly mature and improve, and they are also widely used as an effective diagnostic tool in the rolling bearing fault diagnosis engineering fields. In this paper, use the wavelet transform and neural network as the basis to design and study of the rolling bearing fault diagnosis system. The main contents are as follows:1. Study the wavelet transform de-noising method of rolling bearing vibration signals. Learn the basic principles and algorithm of modulus maximum de-noising, correlation de-noising and threshold de-noising. Denoise the Bumps test signal containing white noise to verify the effectiveness of these methods. By comparing and analyzing their advantages and disadvantages, decided to use the wavelet soft-threshold to denoise the rolling bearing fault signals. The Matlab simulation results show that this method has a good de-noising effect.2. Study extraction method of the wavelet energy spectrum fault feature of vibration signals for rolling bearing. Firstly, wavelet decompose the signals, and reconstruct wavelet coefficients of each layer respectively. Then calculate the time-domain energy of signals on each band to constitute the energy spectrum. In this process, the wavelet decomposition and reconstruction algorithm makes single’s frequency on different bands aliased, so use Fourier transform to eliminate the redundant components after signals through wavelet filters. If the wavelet energy spectrum’s dimension is too much, it will cause system’s identification rate to slow, so propose the new idea that utilize PCA technique to reduct dimension. Matlab simulations show the effectiveness of the two methods.3. Study the fault feature identification method based on Probabilistic Neural Network(PNN) for rolling bearing. Firstly, use feature vectors of known fault type to consist of the training set, establish the PNN’s model. Then apply the PNN to identify failure site according to the extracted fault feature. In order to improve the network identification accuracy, use learning vector quantization(LVQ2) algorithm to train PNN’s output competition layer neuron weights. Matlab simulations verify the effectiveness of the proposed method.