Design and application of a real-time PCBA inspection system based on artificial intelligence technology
Abstract
To address the problems of low efficiency and missed detections caused by the reliance of traditional PCBA inspection on manual visual examination, this paper designs and implements a real-time PCBA inspection system based on artificial intelligence. The system acquires high-resolution images of the PCBA master board through a "four-region acquisition and image stitching" scheme. Taking YOLO11 as the baseline model, an attention feature fusion module is introduced to enhance the perception of tiny solder-joint defects, and knowledge distillation, pruning, and quantization are combined to achieve model lightweighting. Finally, through software-hardware coordinated driving, the system completes the full online inspection workflow of image acquisition, AI inference, and defect annotation output. Experimental results show that the system achieves a mean average precision (mAP@0.5) above 95% for common defects such as normal solder joints, bridging, cold solder joints, insufficient solder, excess solder, component offset, and tombstoning. The inspection time for a single board is about 20 s, which is approximately 80% shorter than manual visual inspection, significantly improving the efficiency and consistency of production-line inspection and providing effective support for real-time quality control on SMT production lines.