基于VMD散布熵与改进灰狼优化SVDD的轴承半监督故障诊断研究

Semi-supervised fault diagnosis of bearings based on the VMD dispersion entropy and improved SVDD with modified grey wolf optimizer

为充分挖掘未标记样本所蕴含的有效信息,进而提升诊断精度,研究提出一种基于变分模态分解(VMD)散布熵与改进灰狼优化支持向量数据描述(SVDD)的轴承半监督故障诊断方法。采用中心频率观察法确定VMD分解模态参数K,进而将原始信号分解为一系列本征模态函数并计算各分量的散布熵值,构成测试样本和部分标记的训练样本;再由半监督模糊C均值(SSFCM)聚类对训练样本进行聚类分析,从而对所得聚类簇进行SVDD建模,同时采用k近邻准则进行决策优化,并由所提自适应变异灰狼算法优化SVDD模型参数;将基于最优参数训练的改进决策SVDD模型用于测试样本的故障模式识别。试验分析和对比结果表明,所提方法具有较好的诊断性能。

To fully explore the effective information contained in unlabeled samples which can contribute to improve the diagnostic accuracy,a semi-supervised fault diagnosis method for bearings was proposed based on the variational mode decomposition(VMD)dispersion entropy and improved support vector data description(SVDD)with modified grey wolf optimizer.The central frequency observation method was applied to determine the mode number K of VMD,thus the original signal was decomposed into a set of intrinsic mode functions.The dispersion entropy values of all components were calculated to construct testing samples and partly labeled training samples.Then,the training samples were clustered by semi-supervised fuzzy C means clustering(SSFCM),after which the obtained clusters were modeled by SVDD.Meanwhile,the k nearest neighbor criterion was employed to improve the decision strategy,and the proposed adaptive mutation grey wolf optimizer was applied to optimize the parameters of SVDD.Finally,SVDD models based on optimal parameters were trained and utilized to recognize the fault pattern of testing samples.The experimental analysis and comparison illustrate that the proposed method achieves superior diagnostic performance.