基于振动信号样本熵和相关向量机的万能式断路器分合闸故障诊断

Diagnosis on the Switching Fault of Conventional Circuit Breaker Based on Vibration Signal Sample Entropy and RVM

为实现对万能式断路器分合闸故障的非侵入式监测和诊断,以分合闸过程中所产生的包含丰富机械特性信息的振动作为信号来源,提出一种基于振动信号互补总体平均经验模态分解(CEEMD)-样本熵和相关向量机(RVM)相结合的万能式断路器故障诊断方法。该方法首先将振动信号通过改进的小波包阈值去噪算法处理;其次采用CEEMD提取若干个反映断路器状态信息的固有模态函数(IMF)分量,依据各IMF分量的能量分布特点,选择其中前7阶进行处理,计算其样本熵形成有效的特征样本;最后通过计算不同故障类型的样本间欧氏距离来定量评价类间样本平均距离,建立基于RVM的二叉树多分类器,诊断得出万能式断路器故障类型。基于所设计的分合闸典型故障模型进行实验。与其他方法的对比实验表明,所提方法可利用相对较少的故障数据样本实现对万能式断路器故障类型的识别并具有较高的识别率;同时实验表明,辅以同一故障类型的样本间欧氏距离,可实现对分合闸故障中三相不同期故障严重程度的初步评估。

In order to realize the non-invasive monitoring and diagnosis of switching fault for conventional circuit breaker,a method that combines complementary ensemble empirical mode decomposition（ CEEMD）-sample entropy and relevance vector machine（ RVM） is proposed for fault diagnosis by using vibration signals during the switching process. Firstly,the vibration signal is denoised by improved threshold wavelet packet denoising algorithm. Secondly,several intrinsic mode function（ IMF） components which reflect main state information are extracted by CEEMD,and the top seven based on energy distribution characteristic are chosen to calculate the sample entropy as the effective feature sample. Finally,the euclidean distance of different fault samples is calculated to evaluate average sample distance between classes. The binary tree classifier based on RVM is established to diagnose the fault type. By results of contrast experiments based on designed typical switching fault models,the method realizes the conventional circuit breaker fault type identification accurately with relatively less fault data samples. And the experiments also show that it can realize the preliminary evaluation of three-phase non-synchronous fault degree based on the euclidean distance of fault samples.