UNet3+深度神经网络在FPC开孔检查中的应用

Application of UNet3+Deep Neural Network in FPC Hole Inspection

开孔工艺是柔性电路板(FPC)制造过程中的一项重要工艺流程;当前工艺流程的加工过程中,会形成毛刺以及偏位、破孔、堵孔及变形等缺陷。针对当前FPC生产过程中的开孔质量问题,提出了一套检测方法及实践,通过研究开孔毛边、毛刺、孔破、堵孔和孔变形等缺陷的形成机理和成像特征,基于UNet3+深度神经网络实现了通孔区域和毛边与毛刺区域的逐像素精确分割。提取出了毛边宽度、毛刺长度、孔圆度、孔偏和孔破等参数。给出了具体的实现方法和过程,整个FPC检查流程包含训练过程及检查过程,其中训练过程包含样本标注、学习训练、模型转换;检查过程包含开孔区域分割、缺陷区域判定、缺陷参数提取以及缺陷判别。实际生产数据验证结果表明:所提方法孔缺陷检查漏检率小于0.01%,误检率低于5%,满足FPC的自动化工业生产中质量检查需要。

The punching process is an important process in the manufacturing of flexible printed circuit(FPC).In the process of the current workflow,burrs and defects such as deflection,broken hole,blocked hole and deformation will be formed.Aiming at the current FPC hole quality inspection problem,the formation mechanism and imaging characteristics of defects such as hole burrs,burrs,hole breaks,plugged holes,and hole deformation,are studied,based on UNET3+depth neural network,pixel by pixel precise segmentation of through-hole area,burr and burr area by is realized.Parameters such as burr width,burr length,hole roundness,hole deviation and hole breakage are extracted.Specific implementation method and process are given.The whole FPC inspection process includes training process and inspection process,in which the training process includes sample labeling,learning training and model transformation.The inspection process includes the segmentation of the hole area,the determination of the defect area,the extraction of the defect parameters and the identification of the defect.The actual production data verification shows that the proposed method has a hole defect inspection with a missed alarm rate of less than 0.01%and a false alarm rate of less than 5%,which meets the quality inspection needs of FPC automated industrial production.