摘要
为了克服传统电机故障检测方法的准确率低、测量过程为侵入式、严重依赖先验知识的缺点,提出了一种基于卷积神经网络的非侵入式电机故障检测方法。通过将电机与其他设备共同工作时的总电源信号作为检测样本,实现检测过程的非侵入式,并基于残差优化卷积网络结构进行神经网络训练,最终实现电机超载、单相短路及相间短路故障的非侵入式检测与分类。结果表明,本文提出的方法可以使故障识别准确率达到96.79%,能够更加快速准确并稳定地实现电机的非侵入式故障诊断。
In order to overcome the shortcomings of traditional motor fault detection methods,namely low accuracy,intrusive measuring process and excessive reliance on prior knowledge,a non-intrusive motor fault detection method based on convolutional neural network was proposed by the author in this paper.The total power signal of the motor and other equipment when they jointly operate was used as the detection sample in non-intrusive detection.And neural network training was performed based on the convolutional network structure featuring residual minimization.Finally,the non-intrusive detection and classification of motor overload,single-phase short circuit and inter-phase short circuit faults were realized.The results show that the fault recognition accuracy rate is raised to 96.79%by using the method proposed,thus enabling quicker and more stable non-intrusive motor troubleshooting.
作者
刘琦昊
许盛之
俞梅
赵二刚
杨松泽
张建军
LIU Qi-hao;XU Sheng-zhi;YU Mei;ZHAO Er-gang;YANG Song-ze;ZHANG Jian-jun(Department of Electronic Science and Technology, College of Electronic Information and Optical Engineering, Nankai University, Tianjin 300350, China;Key Laboratory of Photoelectronic Thin Film Devices and Technology of Tianjin, Tianjin 300071, China;Engineering Research Center of Thin Film Optoelectronics Technology, Ministry of Education, Tianjin 300350, China)
出处
《科学技术与工程》
北大核心
2022年第6期2326-2333,共8页
Science Technology and Engineering
基金
国家自然科学基金(61974074)。
关键词
卷积神经网络
电机故障诊断
残差网络
非侵入式
convolutional neural network
motor fault diagnosis
residual network
non-intrusive
