Fault Feature Extraction Based On Nonlinear Analysis And Recognition Methods Research

To effectively extract the fault features, and realize fault diagnosis and recognition of the equipment, the nonlinear fault feature extraction methods based on Stochastic Resonance and Chaotic Oscillator for rolling bearing and gear were carried out in this paper. Furthermore, the fault pattern recognition method based on the improved Locality Preserving Projections was performed. Details are as follows:(1) The nonlinear fault feature extraction method based on Stochastic Resonance was studied. To solve the problem that Stochastic Resonance is only suitable for low frequency signal components, Frequency-shifted and Re-scaling Stochastic Resonance was introduced, by which the characteristic frequency component was transformed into low frequency component to meet the requirement of Stochastic Resonance. In case of the problem that the structural parameters of the system are difficult to be determined, Frequency-shifted and Re-scaling Stochastic Resonance method optimized by Particle Swarm Optimization was proposed. By using the energy ratio as the fitness function, the structural parameters of Stochastic Resonance were optimized by Particle Swarm Optimization algorithm, thus the optimal output signal was obtained. The proposed method was applied in the feature extraction of the typical fault signal of rolling bearing. The results show that the signal-to-noise ratio of the signal is significantly improved when using the proposed method.(2) The nonlinear fault signal detection method based on Chaotic Oscillator was studied. A method for determining the threshold of Chaotic Oscillator driving force was proposed, through which the numbers of the points near the origin in the output phase diagram under different driving force were calculated. Then the mutation point was obtained, and the driving force value at the mutation point was confirmed to be the threshold of the driving force. The effectiveness of the proposed method was verified by the simulation signal. The Chaotic Oscillator method based on the proposed driving force threshold determination method was applied to detect the rolling bearing failures in the outer race, inner race and rolling element. The results show that the proposed method can detect the bearing fault signal effectively.(3) The fault pattern recognition method of based on Locality Preserving Projections was studied. Traditional Locality Preserving Projections doesn't use the sample class information effectively, which causes poor clustering results. To solve this problem, an improved Locality Preserving Projections method based on the class information was proposed, which improved the clustering results. Based on the improved Locality Preserving Projections, a fault pattern recognition method was proposed. Firstly time domain parameters and wavelet packet energy parameters were selected as the feature parameters of the signal. And then the improved Locality Preserving Projections was used to reduce the dimension of the high dimension feature parameters. Finally the state of each signal was confirmed by Euclidean distance. The proposed method was applied in the fault pattern recognition of rolling bearing and gear. The results show that the proposed method can identify the fault states effectively.