AXI Image Quality Promotion And Automatic Identification Of Defects In BGA

The requirements of the market for electronic products tend to volume miniaturization,functional diversification. In order to meet the market demand, electronics chips are smallerin size the solder joint is more intensive, compared with the past. Therefore, at the same timeto reduce the volume of the products, welding process becomes difficult, and weldingdefects increases, while the difficulty of detection of defects is also more and more high.Because X-ray detection technology has many advantages, such as results intuitive, easy tooperate, and real-time, it has been widely used and developed. In this paper, we studied thequality promotion of image from automatic X-ray detector as well as the automaticidentification of defects.First of all, against the noise appeared in Automated X-ray detector and distortion ofimage appeared in the CCD detector, we did noise reduction and distortion correction to theimage. In the process of noise reduction, mainly used algorithms were mean filteringalgorithm based on gray-scale difference and low-frequency spatial filtering. Bothalgorithms above can effectively remove the noise in the image, and with almost no loss ofimage detail. In the process of distortion correction, the method of calibration template wasused. According to the location of feature points on template and ideal points, the distortioncoefficients can be calculated. Afterwards, transformed coordinates and reconstructed grayscale of the image. Coordinate transformation used the backward mapping method from theideal image to the distorted image. Grayscale reconstruction used the method of bilinearinterpolation.Secondly, according to the images after the process of noise reduction and distortioncorrection, we studied automatic identification of defects. First, segmentation was done tothe image to get images of single solders. Then, the characteristics of the location, size,circularity, etc. of the BGA solders were extracted. After, we fitted a grid according to thelocation information of the solder joint. The grid was corresponding to the theoreticalposition of the solder joints. Further processing was done to the feature quantities extractedbefore to construct feature vectors that can be used for the classification. Finally, a decisionprocess for automatic identification of defects was constructed by using the tree classification method.