中心修正投影结合IGWO-SVM的滚动轴承故障分类方法

A rolling bearing fault classification method based on IGWO-SVM combined with center correction projection

针对滚动轴承故障分类中全局信息损失和模态识别精度低的问题,提出一种中心修正投影算法(center modified projection, CMP)结合改进的灰狼算法(improved grey wolf optimization, IGWO)优化支持向量机(support vector machine, SVM)的滚动轴承故障分类方法。首先,融合样本高维空间全局分布信息和样本局部信息提出CMP降维算法,利用CMP的信息保留能力,实现轴承信号特征矩阵降维;其次借助钟形收敛曲线不同阶段下降速度的差异性特点以及前进式搜索和包围式搜索模式优化灰狼算法收敛性能,并利用IGWO实现SVM参数的自主寻优;最后采用优化后的SVM进行轴承故障分类识别。该方法充分结合了CMP的特征信息保留能力和SVM的小样本分类性能,有效减弱特征冗余成分对诊断结果的影响。多组对比试验表明,所提方法能有效的去除冗余成分,较好的保留样本空间分布信息,具有较好的分类性能。

Aiming at the problems of global information loss and low mode recognition accuracy in rolling bearing fault classification,a rolling bearing fault classification method based on a center modified projection(CMP)dimensionality reduction algorithm and support vector machine(SVM)optimized by an improved gray wolf algorithm(IGWO)was proposed.Firstly,a dimension reduction algorithm of CMP was proposed by combining the global distribution information of the high-dimensional space and the local information of the sample.The dimension reduction of bearing signal feature matrix was realized by using the information retention ability of CMP.Secondly,the bell-shaped convergence curve of normal distribution and the forward search and bounding search modes were introduced to optimize the performance of the gray wolf algorithm.The independent optimization of SVM parameters was realized by using the improved gray wolf algorithm.Finally,the optimized SVM was used for bearing fault classification and recognition.This method fully combines the feature information retention ability of CMP with the small sample classification performance of SVM.The influence of multiple sets of comparative experiments show that this proposed method can effectively remove redundant components,better retain sample space distribution information,and has a good classification performance.