Defect Inspection of Voltage Control IC in Electric Vehicle Chargers Using Surface-Mount Technology

Ensuring the reliability of solder joints is essential for stable operation in electric vehicle chargers, particularly for components assembled using surface-mount technology. Therefore, we developed a defect inspection system for welding joint defects using a Faster Region-based Convolutional Neural Network model to classify results as insufficient defect, shifting defect, and normal (pin-qualified) on voltage control IC pins. The model was trained on 72,000 pin samples and achieved a training accuracy of 99.93%. Evaluation of 65,700 pin samples resulted in an accuracy of 98.89%. The experimental results demonstrate that the system provides stable recognition of reflective solder joints, reliably identifies critical pin-level defects, and is suitable for deployment in practical inspection environments.