| 引用本文: | 郭蕾,曹伟东,张靖康,白龙雷,邢立勐,项恩新,周利军.基于多尺度纹理特征的EPR电缆终端故障诊断方法[J].电力自动化设备,2020,40(11): |
| GUO Lei,CAO Weidong,ZHANG Jingkang,BAI Longlei,XING Limeng,XIANG Enxin,ZHOU Lijun.Fault diagnosis method of EPR cable terminal based on multi-scale texture features[J].Electric Power Automation Equipment,2020,40(11): |
|
| 摘要: |
| 高速铁路列车中乙丙橡胶(EPR)电缆终端常因制作过程中操作不当而导致出现多种缺陷，造成终端局部放电甚至击穿的现象，对检测到的局部放电信号进行准确分类仍然是亟待解决的难题。分别制作了含尖端、环切划伤、金属微粒、气隙4种典型缺陷的终端试样，通过试验记录了各类试样的放电谱图信息。基于试验得到的局部放电谱图库，提出通过图像金字塔理论，构建多尺度局部放电谱图空间，并从中提取1阶纹理统计量、2阶纹理统计量及高阶纹理统计量作为缺陷类型识别的特征参量。同时，结合随机森林算法，基于gini指数完成了特征空间寻优工作，实现了对缺陷类型的正确分类。结果表明：通过随机森林算法，多尺度纹理特征的模型误差率和分类准确率明显优于单尺度纹理特征，能较好地对终端典型缺陷进行分类；同时，由于气隙缺陷和划伤缺陷的放电机理存在相似性，二者缺陷处的局部放电特征同样存在相似现象，因此气隙缺陷和划伤缺陷的分类较易出现误差，导致其识别率偏低，还需针对其图像特征参数和识别算法继续开展研究。 |
| 关键词: 乙丙橡胶电缆 典型缺陷 纹理特征 图像金字塔 随机森林算法 故障诊断 |
| DOI：10.16081/j.epae.202008018 |
| 分类号:TM247 |
| 基金项目:四川省科技计划资助项目(2020JDTD0009) |
|
| Fault diagnosis method of EPR cable terminal based on multi-scale texture features |
|
GUO Lei1, CAO Weidong1, ZHANG Jingkang1, BAI Longlei1, XING Limeng1, XIANG Enxin2, ZHOU Lijun1
|
|
1.College of Electrical Engineering, Southwest Jiaotong University, Chengdu 610031, China;2.Electric Power Research Institute, Yunnan Power Grid Co.,Ltd.,Kunming 650217, China
|
| Abstract: |
| In high-speed railway train, EPR(Ethylene Propylene Rubber) cable terminals often have many defects due to improper operation in the manufacturing process, resulting in PD(Partial Discharge) or even breakdown of the terminals. It is still a problem to be solved for the accurate classification of detected partial discharge signals. The terminal samples with four typical defects of tip, scratch, metal particle and air gap are made, and the discharge spectrum information of each sample is recorded by experiment. Based on the PD spectrum library obtained from the experiment, a new method is proposed to construct multi-scale PD spectrum space by image pyramid theory, from which the first-order texture statistics, the second-order texture statistics and the high-order texture statistics are extracted as the feature parameters of defect type recognition. At the same time, the random forest algorithm is introduced, and the feature space optimization is completed based on gini index, which realizes the correct classification of defect types. The results show that the model error rate and classification accuracy rate of multi-scale texture feature are significantly better than that of single scale texture feature, which can better classify the terminal typical defects, at the same time, because the discharge mechanism of air gap defect and scratch defect is similar, the partial discharge characteristics of both defects are similar, so the classification of air gap defect and scratch defect is prone to errors, resulting in low recognition rate. Further research on image feature parameters and recognition algorithm is needed. |
| Key words: EPR cable typical defects texture features image pyramid random forest algorithm fault diagnosis |
