Machine-learning based characteristic estimation method in printed circuit board production lines

Abstract

In this study, software and hardware that supported automatic optical inspection (AOI) for printed circuit board production line was proposed and demonstrated. The proposed method showed an effective solution that predicts off-line electromagnetic (EM) characteristic of manufactured components through in-line pattern integrity. A spiral antenna that represented complex patterns was used as the evaluation target with imitated production variations. Numerical evaluation on EM properties, batch fabrication, hardware setup and optimization, algorithm and graphical user interface development, machine learning and artificial intelligence modeling, and data verification and analysis were thoroughly conducted in this study. Results indicated that when the antenna showed pattern distortion, its passive capacitance, active intensity, and active frequency increased, decreased, and decreased, respectively. These results proved that the developed system and method overcame the inability of in-line EM measurement in conventional setup. The results also showed high estimation accuracy that was not yet achieved in the past. Compared to existing or similar AOI ideas, the proposed method supports analyses on complex pattern, provides solutions on target design, and efficient algorithm generation. This work also proved active and passive EM signals with evidences, and exhibited outstanding confidence levels for characteristic estimations. The proposed system and method indicated their potential in smart manufacturing.