Defect Detection for Printed Circuit Board Assembly Using Deep Learning

In many industrial applications, the number of defect samples is often insufficient for defect detection using conventional deep learning techniques. Also, the frequent change of PCBA board types on the product line introduces new defect types and adds a layer of challenge to the detection task considered in this paper. We propose a deep learning algorithm that targets learning patterns from various defect types with unbalanced training samples in the PCBA manufacturing product lines. A novel batch sampling method is proposed for the deep learning method for PCBA defect detection. We have validated the proposed algorithm using normal and defective images. The results show that the proposed deep learning method can accurately identify defects in PBCA images and achieve an overall accuracy of 98%. This deep learning technique can also be extended to detect other surface-level defects.