Failed Component Image Acquisition Quality Optimization for Machine Vision System
With the rapid development of the global semiconductor industry, electronic products are required to have new ideas, diversified functions, and thinness and shortness. The ball grid array (BGA) packaging technique arranges solder balls in a matrix mode on the bottom of the component substrate, so as to increase the functional density. The dye stain test is extensively used for array component failure analysis; however, the operation takes time and extends as the number of solder balls increases, and the judgment result may vary with the identifier's experience or subjective consciousness. The intelligent red ink machine vision analysis system implements automatic recognition, using a camera to shoot the entire component, so as to obtain images for subsequent identification, however this will cause imaging definition loss, leading to a low recognition accuracy rate.
This study implemented the Taguchi parameter optimization design for an image acquisition system in which the image quality was upgraded, so as to guarantee the recognition correctness and stability of the image for a red ink machine vision analysis system. The judgment results of the solder ball crack degree and solder ball crack interface type were considered in the study. The control factors were parameters related to image acquisition system equipment and software setting. The noise factors included parameters of the operator or ambient environment that could influence image quality during the image acquisition. The correct ratio of the image for the machine vision system to judge failure phenomena was converted using Omega, and the result was used as the S/N. Considering the overall judgment correctness of the system, this study used the principal component analysis method for data analysis. The optimum parameter combination was as follows: a high-angle light source illumination angle, space between the light source and the component that moves with the lens, an aperture set at 8.0, and an image processing algorithm consisting of the coupled wave method with an automatic parameter tuning mechanism. The accuracy of the automatic failure recognition of the machine vision analysis system was higher than 80%..
Keywords - Failed Component, Image Acquisition Quality, Design of Experiment