机器学习在复合绝缘子缺陷超声检测中的应用与展望  

Application and prospect of machine learning in ultrasonic testing of composite insulator defects

作　　者：孙洪宇 彭丽莎[1] 屈凯峰 王珅[1] 赵伟[1] 黄松岭[1] SUN Hongyu;PENG Lisha;QU Kaifeng;WANG Shen;ZHAO Wei;HUANG Songling(State Key Laboratory of Power System and Generation Equipment Control and Simulation,Tsinghua University,Beijing 100084,China;Beijing Yundao Zhizao Technology Co.,Ltd.,Beijing 100190,China)

机构地区：[1]清华大学电力系统及发电设备控制和仿真国家重点实验室,北京100084 [2]北京云道智造科技有限公司,北京100190

出　　处：《无损检测》2021年第5期58-63,共6页Nondestructive Testing

基　　金：国家自然科学基金项目(52077110);国家自然科学基金青年项目(52007088)。

摘　　要：在综述不同无损检测方法的基础上,重点介绍了3种复合绝缘子缺陷的超声检测方法,分析了各自的优势和不足,并讨论了基于机器学习原理的超声检测缺陷识别方法,总结了深度学习在超声缺陷识别预测中的优势,探讨了基于深度学习的复合绝缘子超声检测中的3大难题,给出了可行的解决策略,可为该领域的进一步研究提供技术参考。Based on a review of different nondestructive testing methods, we focused on three ultrasonic testing methods for composite insulator defects and analyzed their respective advantages and disadvantages. We also discussed the ultrasonic defect identification method based on the machine learning principle, summarized the advantages of deep learning in ultrasonic defect recognition and prediction, and discussed the three major problems in the ultrasonic testing of composite insulators based on deep learning. Finally, a feasible solution strategy is given, which can provide a technical reference for further research in this field.

关 键 词：绝缘设备 复合绝缘子 超声检测 机器学习 深度学习 

分 类 号：O347[理学—固体力学] TG115.28[理学—力学]