基于WPT-CNN的复合绝缘子内部缺陷智能识别研究

Research on Intelligent Identification of Internal Defects in Composite Insulators Based on WPT-CNN

超声波技术常用于复合绝缘子内部缺陷的检测,但缺陷识别过程依赖于试验人员专业经验。为实现复合绝缘子内部缺陷的智能识别,提出了一种基于小波包变换和卷积神经网络的超声波检测信号识别模型。首先,通过小波包变换对超声波检测信号进行时频特征提取,并将一维信息转化为二维特征矩阵;其次,将二维特征矩阵输入卷积神经网络中,实现对信号特征的智能识别;最后,采用试验信号样本集对模型进行训练与测试。结果表明,提出的模型能对缺陷、气孔、裂纹、界面脱粘和夹杂五类复合绝缘子超声波检测信号进行识别,且平均准确率可达98.7%,能为复合绝缘子内部缺陷的智能识别提供很好的工程应用参考。

Ultrasonic technology is commonly used for detecting internal defects in composite insulators,but the defect identification process still depends on the professional experience of the testing personnel.To achieve intelligent identification of internal defects in composite insulators,an ultrasonic detection signal recognition model based on wavelet packet transform and convolutional neural network was proposed.Firstly,time-frequency features of ultrasonic detection signals were extracted through wavelet packet transform,and one-dimensional information was transformed into a two-dimensional feature matrix;secondly,the two-dimensional feature matrix was input into the convolutional neural network to achieve intelligent recognition of signal features;finally,the model was trained and tested using an experimental signal sample set.The results show that the proposed model can identify ultrasonic detection signals of five types of composite insulators:defects,pores,cracks,interface debonding and inclusions,and the average accuracy can reach 98.7%.It can provide good engineering application reference for intelligent identification of internal defects in composite insulators.