Roller Bearing Fault Diagnosis Based On Ant Colony Algorithm And Statistical Filtering

The paper carried out adaptive filtering method based on statistical test theory and feature extraction method based on stochastic resonance nonlinear analysis for roller bearing typical fault diagnosis. A variety of feature extraction methods were combined to optimize the ant colony optimization (ACO). The optimized ACO was also used to fault pattern recognition. Both of vibration and acoustic emission signals of cylindrical roller bearing were studied. The diagnosis of single fault and compound faults were taken to vertify the effectiveness of the method. Details are as follows:(1) Discrimination Index was applied to select the characteristic parameters with high sensitivity. The input vector was composed by sensitive parameters. Next, ant clustering system had elite ants was used to identify equipment state. Then, wavelet packet decomposition, empirical mode decomposition method (EMD) and principal component analysis (PCA) were used to extract signal characteristics, which brought out the complementary advantages between different methods to improve clustering performance of ACO.(2) Adaptive statistical filtering method based on statistical hypothesis testing was proposed.Similarity factor was defined to evalute the filtering performance and used as the objective function of genetic algorithms (GA). The significance level was optimized by GA to complete the filtering process automatically. Signal to noise ratio (SNR) in frequency domain was introduced to evalue the noise suppression ability of the filtering method. The results showed that statistical filtering more effectively improved SNR when compared with high-pass filtering.(3) In addition, statistical filtering was combined with stochastic resonance and chaotic oscillator in nonlinear signal analysis methods. The feature extraction methods of weak fault signal were researched. Statistical filtering and stochastic resonance were used to extract signal feature. Frequencies of peaks in the envelope spectrum, which was the output of stochastic resonance, were detected by chaotic oscillator. The result indicates the nonlinear signal analysis methods have advantage in weak signal feature extraction.