Open Access

ARTICLE

Automated X-ray Defect Inspection on Occluded BGA Balls Using Hybrid Algorithm

Ki-Yeol Eom¹, Byungseok Min^2,*

1 Department of AI Bigdata, Suseong University, Daegu, 42078, Korea
2 R&D Center, XAVIS Co., Ltd., Seongnam-si, 13202, Korea

* Corresponding Author: Byungseok Min. Email:

Computers, Materials & Continua 2023, 75(3), 6337-6350. https://doi.org/10.32604/cmc.2023.035336

Received 17 August 2022; Accepted 29 January 2023; Issue published 29 April 2023

Abstract

Automated X-ray defect inspection of occluded objects has been an essential topic in semiconductors, autonomous vehicles, and artificial intelligence devices. However, there are few solutions to segment occluded objects in the X-ray inspection efficiently. In particular, in the Ball Grid Array inspection of X-ray images, it is difficult to accurately segment the regions of occluded solder balls and detect defects inside solder balls. In this paper, we present a novel automatic inspection algorithm that segments solder balls, and detects defects fast and efficiently when solder balls are occluded. The proposed algorithm consists of two stages. In the first stage, the defective candidates or defects are determined through the following four steps: (i) image preprocessing such as noise removal, contrast enhancement, binarization, connected component, and morphology, (ii) limiting the inspection area to the ball regions and determining if the ball regions are occluded, (iii) segmenting each ball region into one or more regions with similar gray values, and (iv) determining whether there are defects or defective candidates in the regions using a weighted sum of local threshold on local variance. If there are defective candidates, the determination of defects is finally made in the following stage. In the second stage, defects are detected using the automated inspection technique based on oblique computed tomography. The 3D precision inspection process is divided into four steps: (i) obtaining 360 projection images (one image per degree) rotating the object from 0 to 360 degrees, (ii) reconstructing a 3D image from the 360 projected images, (iii) finding the center slice of gravity for solder balls from the axial slice images in the z-direction, and getting the inspection intervals between the upper bound and the lower bound from the center slice, and (iv) finally determining whether there are defects in the averaged image of solder balls. The proposed hybrid algorithm is robust for segmenting the defects inside occluded solder balls, and improves the performance of solder ball segmentation and defect detection algorithm. Experimental results show an accuracy of more than 97%.

---

Keywords

---