电工技术学报2026，Vol.41Issue(1)：315-328,14.DOI:10.19595/j.cnki.1000-6753.tces.242238
基于GA-BP神经网络的碳纤维复合芯导线压接缺陷识别方法
Identification Method for Crimping Defects of Carbon Fiber Composite Core Wires Based on GA-BP Neural Network
摘要
Abstract
Carbon fiber composite core wire has a good application prospect in the capacity improvement of the line because of its low carbon and energy saving characteristics,but the carbon fiber core rod is very fragile,the technology is immature,and the wire breakage accidents caused by poor crimping often occur,which restricts the promotion of this technology.In recent years,X-ray,acoustic imaging and infrared imaging are used in most studies,but these methods have some problems,such as unclear imaging and great influence by external environmental factors.Therefore,aiming at the two serious crimping defects of carbon fiber core fracture and carbon fiber core undercompression,this paper proposes a magnetic preparation method of carbon fiber composite core wire based on magnetic flux leakage detection technology,and designs a signal feature extraction method of MFL detection of crimping defects.Genetic algorithm optimized back propagation(GA-BP)neural network was built to recognize the defect characteristic value,which effectively improved the accuracy of magnetic flux leakage(MFL)testing detection method to identify the defect degree of carbon fiber core wires. Firstly,the magnetic properties of carbon fiber core wires are prepared by using the film coating method.The defects of 108 groups of carbon fiber core wires were prepared by using pure iron film coated on the surface of carbon core for crimping.The fiber core undercompression defects covered by the samples ranged from 2～60 mm and the carbon fiber core fracture defects covered by the samples ranged from 0.5～5 mm.Secondly,the magnetic leakage detection system of carbon fiber core wire is established to measure 613 groups of magnetic leakage detection data.The MFL detection system can collect MFL signal and velocity data and process and display them.According to the peak value of MFL detection signal,the defects of carbon fiber core wires are divided into 16 categories according to different degrees.Thirdly,this paper optimizes the extraction method of signal characteristic values through experiments,and takes the amplitude of 7 peak points,21 relative position information and 7 waveform type information in the magnetic flux leakage detection signal data as the characteristic values of defect judgment,effectively improving the accuracy of identification of defect type and degree.Finally,the GA-BP neural network was built and trained by the extracted eigenvalue data,and the sensitivity of key indicators such as population number and learning rate was analyzed to optimize the accuracy and operation efficiency of the model. The experimental results show that the inductance sensor signal increases with the increase of defect degree.The fiber core undercompression defect signal of carbon fiber core increases from 3 000 at 20 mm defect to 4 000 at 40 mm defect,and the carbon fiber core fracture defect signal of carbon fiber core increases from 4 000 at 2 mm defect to 6 000 at 4 mm defect.On this basis,the amplitude,relative position and waveform type of the peak point are statistically analyzed.The relative position parameters of the normal carbon fiber core sample are higher at 1～6 positions,and the peak type parameters at 2～3 positions will show a small amount of-1.The relative position parameters of fiber core undercompression defect samples are negative at positions 7～11,and the peak type parameters at positions 2～3 have more-1 values than normal defects.The relative position parameters of carbon fiber core fracture defect samples at positions 7～11 are positive,and the peak type parameters at positions 1 are-1,and the peak type parameters at positions 3～7 are-1.By inputting 35 characteristic values of MFL detection signals into GA-BP neural network,the accuracy of defect degree recognition can reach 94.31%. The following conclusions can be drawn from the above experimental phenomena:(1)The preparation method of carbon fiber core wire based on the film coating method can well adapt to the magnetic flux leakage detection technology and identify hidden defects.(2)The feature data extraction method of the magnetic flux leakage detection signal of carbon fiber core wire designed in this paper,through 7 groups of peak point amplitude feature data,21 groups of peak point relative position feature data and 7 groups of peak point type feature data,The description of the key signal characteristics of MFL detection is concise and clear.(3)The GA-BP neural network trained in this paper uses 613 sets of actual magnetic flux leakage detection and measurement data to train the test model,and performs parameter optimization according to the prediction results of the model,making the prediction accuracy of the model for the defect degree of carbon fiber composite core wires reach 94.31%.关键词
碳纤维复合芯导线/缺陷识别/磁性制备/漏磁检测/遗传算法/BP神经网络Key words
Carbon fiber composite core wire/defect identification/magnetic preparation/magnetic leakage detection/genetic algorithm/BP neural network分类
信息技术与安全科学引用本文复制引用
Du Zhiye,Huang Ziren,Feng Bo,Yue Guohua,Liao Yongli..基于GA-BP神经网络的碳纤维复合芯导线压接缺陷识别方法[J].电工技术学报,2026,41(1):315-328,14.基金项目
南方电网公司科技项目(GXKJXM20220077)和国家自然科学基金项目(52377015)资助. （GXKJXM20220077）
