基于深度学习的线束端子外观缺陷检测技术研究

Research on Detection Technology of Wire Harness TerminalAppearance Defects Based on Deep Learning

为解决接触件端子与线束压接过程中会出现未成功压接、压接不到位、芯线外漏等质量问题,提出一种基于传统机器视觉和深度学习相结合的方法实现线束端子外观缺陷智能检测。首先搭建视觉检测系统获取高质量的线束端子外观图像,接着应用传统机器视觉中的图像预处理、图像滤波、最小二乘法实现剥线长度的在线检测,然后人工标注线束端子外观缺陷并构建样本数据集,最后利用深度学习算法实现线束端子外观缺陷的智能检测。试验结果表明,该视觉检测系统与人工检测对比误差小于0.01 mm,模型缺陷识别准确率为99.33%,漏检率为零,单张图像推理耗时5.6 ms。该系统运行稳定可靠,满足实际生产需求。

In order to solve the quality problems such as unsuccessful crimping,incomplete crimping and core wire leakage in the crimping process between the contact terminal and the harness,an intelligent inspection method based on the combination of traditional machine vision and deep learning was proposed to realize the appearance defects of harness terminals.First,built a visual inspection system to obtain high-quality appearance images of harness terminals.Then,applied image preprocessing,image filtering,and least squares methods in traditional machine vision to realize online detection of stripping length.Then manually labelled the appearance defects of harness terminals and built sample data sets.Finally,used depth learning algorithm to achieve intelligent detection of appearance defects of harness terminals.The test results showed that the contrast error between the visual inspection system and manual inspection was less than 0.01 mm,the model defect recognition accuracy was 99.33%,the missed detection rate was zero,and the reasoning time for a single image was 5.6 ms.The system ran stably and reliably to meet the actual production demand.