挠性印制电路板焊盘表面缺陷的检测

Detection of surface defection of solder on flexible printed circuit

针对挠性印制电路板(FPC)上的焊盘表面缺陷,提出一种基于图像处理技术的智能检测方法。首先,根据缺陷的表现形式对焊盘缺陷进行归纳与分类,采用最大熵值法量化提取焊盘的颜色特征和面积特征;然后,通过评估灰度共生矩阵(GLCM)对纹理颜色变化特征与纹理结构特征量化的有效性,将其应用于焊盘纹理特征的量化与提取。实验分析显示,缺陷焊盘与非缺陷焊盘在某个或多个特征上存在着明显的差异。基于该特点,建立了BP神经网络,以焊盘的颜色、面积、纹理结构、纹理颜色变化特征值作为神经网络的输入,通过学习大量样本,获取最佳权值参数,最终实现对FPC焊盘表面缺陷的检测,检测准确率高达94.6%,50个焊盘的检测时间为300ms。实验结果表明:本文提出的检测方法不仅能够有效地对焊盘表面缺陷进行识别,而且能够满足在线检测对速度的要求。

In order to detect the surface defect on the solder of a Flexible Printed Circuit（FPC）, an inspecting technology based on image processing was presented. Firstly, all the defects on the FPC were classified into several defection sorts according to their defection characters. Then, the maximum entropy was used to locate the solder and extract the square and color characters. After estimating the effectness of the Grey Level Co-occurrence Matrix（GLCM） on the quantification for color and structure characters, it was introduced to quantify and extract colorific and structural textures for solders. An analysis on experiments indicates that the defective solder is obviously different from the non-defective solder in several kinds of quantified charaters. On the basis of the result obove,the BP neural network was established and four kinds of characters were selected as the input of neural network. After all neural network weight parameters were adjusted to the optimization through sample training, the performance of the proposed defect detection algorithm was finally evaluated in an on line testing. Test shows that 50 inspecting targets cost 300 ms, and the inspecting accuracy can reach 94.6%. The experimental result demonstrates that proposed method can detect accurately the solder defect with low false alarms, and the efficiency can satisfy the requirement of defect inspection in online and real time.