基于深度学习的航天板级组件缺陷检测

Method of Defect Detection of Aerospace Circuit-board Based on Deep Learning

面向航天柔性电路装配产线的板级组件缺陷智能诊测问题,采用基于YOLOv3的深度网络方法,解决传统AOI检测精度不高及配置效率低下等问题。通过对应电路板的*.PcbDoc文件解析板上元器件位置和尺寸,并据此将待测图片缩小为单网格的元器件待检区域,从而提升缺陷检测算法的通用性。通过研究YOLOv3的航天产品元器件网格区域缺陷检测原理,完成YOLOv3深度网络算法建模,搭建该样机在某型号电路板产线的原理演示验证系统,提高缺陷检测速度和精确性,实现航天系统内电路板故障/缺陷检测的高精度、高效率识别,为将来航天智能制造转型之路夯实基础。

The board level component defect detection of the intelligent production line of the aerospace flexible circuit assembly is a serious problem in manufacturing.This paper provides a new method which is YOLOv3 to use it to improve the poor detection accuracy and low configuration efficiency of traditional AOI.The image which needs to be tested can be narrowed down to a small gird component through analyzing the location and size of the PcbDoc file of the circuit board.Thus,the universality of the algorithm can be improved.Besides,deep detection of network algorithm modeling of YOLOv3 can be completed by learning the principle of deep detection of YOLOv3 space product.The aim of setting the verification system of circuit board production line is to improve the speed and accuracy of defect detection test.It is also an efficient way to provide a high precision and high efficiency identification for circuit board and this method will be the foundation for the digital transformation in the space industry in the future.