基于ANHGA-VMD和耦合隐马尔可夫模型的有载分接开关机械故障诊断

Mechanical Fault Diagnosis of On-load Tap-changer Based on ANHGA-VMD and Coupled Hidden Markov Model

机械故障是有载分接开关(on-loadtap-changer,OLTC)的主要故障类型。为解决变分模态分解(variational mode decomposition,VMD)参数设置对分解质量的影响难以确定的问题,并进一步提高OLTC机械故障诊断准确度,提出了一种基于多通道振动信号的有载分接开关机械故障诊断方法。首先,采用具有稳定寻优能力的自适应小生境递阶遗传算法(adaptive niched hierarchical genetic algorithm,ANHGA)并将品质因数作为衡量信号分解质量的标准对VMD进行参数寻优,然后利用优化后的VMD方法对预处理过的多通道振动信号进行分解,提取VMD能量熵和模糊熵作为特征值,最后建立耦合隐马尔可夫模型(coupled hidden Markov model,CHMM)进行多通道数据的故障诊断。实例验证表明:采用经优化后的VMD分解能够有效提高VMD信号分解质量;基于CHMM对OLTC的7种典型状态诊断准确率达100%,其故障诊断准确率高于隐马尔可夫(hidden Markov model,HMM)方法和支持向量机(support vector machine,SVM)方法。所提方法为由采样困难造成样本稀少的复杂机电设备的诊断问题提供了新思路。

Mechanical fault is the main fault type of on-load tap-changer(OLTC).To solve the problem that it is difficult to determine the influence of variational mode decomposition(VMD)parameter settings on the decomposition quality and to improve the mechanical fault diagnosis accuracy of OLTC,a mechanical fault diagnosis method of OLTC based on multichannel vibration signal is presented.Firstly,the adaptive niched hierarchical genetic algorithm(ANHGA)with stable optimization capability is adopted to optimize the VMD parameters,in which the quality factor are used as the criterion to measure the signal decomposition quality.After that,the improved VMD method is used to decompose the pretreated multichannel vibration signals,and the energy entropy and fuzzy entropy of VMD are extracted as eigenvalues.Finally,a coupled hidden Markov model(CHMM)is established for fault diagnosis of multichannel data.The case study results show that the optimized VMD can effectively improve the decomposition quality of VMD signals.The diagnostic accuracy of the seven typical states of OLTC based on the CHMM is 100%,which is higher than the correct rate of fault diagnosis methods of hidden Markov model(HMM)and support vector machine(SVM).A new idea for the diagnosis of complex electromechanical equipment with few samples,which is difficult to sample,is provided by the proposed method.