基于深度学习的断路器缺陷检测
Circuit Breaker Defect Detection Based on Deep Learning

断路器是电气自动控制系统中的重要元件之一，负责线路的接通和关断，起到保护和控制电力系统的作用。考虑到电气控制系统的复杂性，若存在制造缺陷的断路器被投入使用，不仅有出现事故的风险，也增加了检修难度。因此，断路器生产制造过程中对其装配正确性的检验工作尤为重要。传统的器件缺陷检测主要是以人工检验和物理损伤检测为主，检验过程耗时费力，占用大量人力资源。为解决上述问题，本文采用YOLOv5算法建立基于机器视觉技术的断路器装配缺陷检测模型，针对断路器装配过程缺失灭弧室和绝缘片的缺陷进行检测。经验证，该模型在检测灭弧室的mAP@0.5达到了0.97以上，而对绝缘片的检测mAP@0.5也超过了0.96。为满足实际工程需求，本文将建好的缺陷检测模型部署至NVIDIA Jetson Nano边缘AI计算设备上，实现工程应用的小型化和便利化。
Circuit breaker is one of the most important components of the electrical automatic control system, which is responsible for switching the circuit on and off. Given the complexity of the electrical control system, the electric power accidents are more likely to appear if the circuit breakers with assembly defects are directly put into use, also increasing the difficulty of overhaul. Missing arc extinguish chambers or insulators are common in circuit breaker assembly process, which greatly affects the products’ final quality. Conventional assembly defect detection is mainly based on manual inspection and physical damage detection, and the inspection process is time-consuming, labor-intensive and requires a lot of manpower. A defecting system based on the YOLOv5 algorithm is proposed to solve the above problems. The model achieves an outstanding mAP@0.5 exceeding 0.97 for arc extinguish chamber detection as well as a remarkable mAP@0.5 of almost 0.96 for detection of the existence of insulators. In order to meet the actual technical requirements, the developed fault detection model is used on the NVIDIA Jetson Nano edge AI computing device to achieve the miniaturization and convenience of technical applications.