Page 84 - 《应用声学》2025年第1期
P. 84

第 44 卷 第 1 期                                                                       Vol. 44, No. 1
             2025 年 1 月                          Journal of Applied Acoustics                   January, 2025

             ⋄ 研究论文 ⋄



              高压电缆终端铅封缺陷超声图像卷积神经网络识别                                                                     ∗





                   方春华     1†   周 固    1   邵 斌     1   胡冻三     1    夏 荣    2   欧阳本红       2   普子恒      1


                                             (1 三峡大学电气与新能源学院        宜昌  443002)
                                         (2 中国电力科学研究院有限公司武汉分院           武汉  430074)

                摘要：高压电缆终端铅封因安装工艺不当以及在外力作用下会出现孔洞、脱粘或裂缝等缺陷，严重影响输电线
                路稳定运行。为解决传统超声检测铅封缺陷是因通过人工观察超声图像而存在的效率和准确率偏低的问题，
                该文提出了一种基于卷积神经网络的高压电缆终端铅封缺陷超声图像识别方法，可以自动从铅封缺陷超声图
                像中学习特征并完成缺陷分类识别。建立了 4 种典型铅封缺陷超声图像样本库，搭建了铅封缺陷超声图像识
                别模型，采用经过规范化处理的超声图像数据对模型进行训练和测试。结果表明：通过调整卷积神经网络试验
                参数，能够快速准确地识别出铅封不同类型缺陷，准确率可以达到 100%，表明该方法具有良好的鲁棒性，抗干
                扰能力强，对铅封缺陷具有良好的检测性能，在实际的终端铅封缺陷检测中具有很好的应用前景。
                关键词：电缆终端；铅封；超声图像识别；卷积神经网络；缺陷检测
                中图法分类号: TB552           文献标识码: A          文章编号: 1000-310X(2025)01-0080-08
                DOI: 10.11684/j.issn.1000-310X.2025.01.007



               The convolutional neural network recognition of ultrasonic images of lead seal

                                      defects in high-voltage cable terminals


                                           1
                                                                                1
                            FANG Chunhua , ZHOU Gu , SHAO Bin , HU Dongsan , XIA Rong ,
                                                                                            2
                                                                   1
                                                       1
                                                              2
                                           OUYANG Benhong and PU Ziheng      1
                        (1 College of Electricity and New Energy, China Three Gorges University, Yichang 443002, China)
                          (2 China Electric Power Research Institute Co., Ltd., Wuhan Branch, Wuhan 430074, China)
                 Abstract: Due to improper installation process and external forces, defects such as holes, delamination, or
                 cracks may occur in the lead seals of high-voltage cable terminals, seriously affecting the stable operation of
                 transmission lines. To solve the problem of low efficiency and accuracy in traditional ultrasonic inspection of
                 lead seal defects caused by manual observation of ultrasonic images, this paper proposes a high-voltage cable
                 terminal lead seal defect ultrasonic image recognition method based on convolutional neural network, which
                 can automatically learn features from lead seal defect ultrasonic images and complete defect classification
                 recognition. Four typical lead seal defect ultrasound image sample libraries were established, and a lead
                 seal defect ultrasound image recognition model was constructed. The model was trained and tested using
                 standardized ultrasound image data. The results show that by adjusting the experimental parameters of
                 the convolutional neural network, different types of defects in lead seals can be quickly and accurately identified,


             2023-10-07 收稿; 2024-03-07 定稿
             电网环境保护国家重点实验室 2022 年度实验室开放基金项目 (GYW5120221415)
             ∗
             作者简介: 方春华 (1980– ), 男, 湖北武汉人, 博士, 副教授, 研究方向: 高电压试验技术、绝缘子和电力设备状态检测。
             † 通信作者 E-mail: fang107531@163.com
   79   80   81   82   83   84   85   86   87   88   89