AI soldering inspection for X-D0240-326 in Antenna Potting line

Abstract

The project began with the research, selection, and integration of hardware components, including an industrial camera, suitable lenses, custom camera jigs, and a central processing unit, to create a complete inspection station compatible with the existing production line. We utilized the NVIDIA Jetson as the image processing server, connected to the camera to capture data, analyze it using an AI model, and transmit classification results via an internal network, ensuring fast processing, low latency, and high stability in the production environment. Real-world production images were collected, labeled, and used to train a deep learning model, fine-tuned to detect weld imperfections with high accuracy and reliability under factory conditions. The system was synchronized with PLCs and actuators, enabling real-time inspection and classification of products as pass or fail, automatically sorting them into separate trays before proceeding to the next stage. Despite time and resource constraints, the project achieved positive outcomes, including reducing one manual inspection worker, shortening inspection time, and improving defect detection consistency. The project demonstrates the feasibility of integrating AI into traditional manufacturing, laying the groundwork for future intelligent quality control systems.