基于EMD-ICA与遗传算法的轴承故障诊断方法

Bearing Fault Diagnosis Method based on EMD-ICA and Genetic Algorithms

针对振动信号中轴承故障特征信号微弱难以识别的问题,对通过试验采集到的内环故障、外环故障以及滚动体故障振动信号进行处理。采用最小二乘法和指数平滑法对振动信号进行预处理,利用EMD分离振动信号的局部特征,并根据IMF分量的信息熵增益比实现重构;采用ICA对混叠的振动信号进行分离,并对分离后的振动信号进行特征提取;采用遗传算法对多维振动特征参量进行降维,筛选出最优特征参量;采用遗传算法优化的极限学习机对轴承故障振动特征集进行识别,将常见的SVM、BP等诊断模型作为对比算法。试验结果表明:采用ICA能将混叠信号有效分离,实现故障信号的提取;遗传算法不仅能够实现最优特征的选择,同时能够对极限学习机算法进行有效优化,提升算法的诊断效果。优化的算法相比其它诊断识别方法性能较佳,使3种故障的平均诊断效果达到90%以上。

In view of the problem that the bearing fault characteristic signal in the vibration signal is weak and difficult to identify,this paper deals with the inner loop fault,the outer loop fault and the rolling element fault vibration signal collected by the laboratory.The vibration signal is preprocessed by least squares method and exponential smoothing method.And local characteristics of the vibration signal are separated by EMD,and the reconstruction is realized according to the information entropy gain ratio of the IMF component.The hybrid vibration signal is separated by ICA,and the feature extraction of the separated vibration signal is carried out.The genetic algorithm is used to reduce the dimension of the vibration characteristic parameters,and the optimal characteristic parameters are selected.Finally,the bearing fault vibration feature set is identified by the extreme learning machine optimized by genetic algorithm,and the common SVM and BP are used as comparison algorithms.The experimental results show that ICA can effectively separate the mixed signals,so as to realize the extraction of more faulty features.The genetic algorithm can not only achieve the optimal feature selection,but also extreme learning machine optimized by genetic algorithm compared with other diagnostic methods,which can make the average diagnosis effect of three kinds of faults reach more than 90%.