This technology presents an AI-based defect inspection platform with root-cause analysis and few-shot learning capabilities. By integrating zero-shot learning and LLM-based reasoning, the system can perform real-time defect recognition and anomaly analysis even with limited defect samples. It effectively detects tiny irregular cracks and complex surface defects while maintaining highly reliable inspection performance under low-light and high-interference environments.

This technology pioneers the integration of LLMs into industrial AOI systems, enabling not only defect detection but also root-cause analysis and explainable AI reasoning. By combining Few-shot and Zero-shot Learning, the system adapts to new products with minimal labeled data, reducing retraining cost and deployment time. The proposed PRB-FPN, multi-scale fusion, and Siamese network architecture can detect ultra-small 10×10 pixel cracks with high accuracy and low latency on Edge AI devices.

This technology addresses industrial inspection challenges in semiconductors, PCB, automotive electronics, and petrochemical industries, where tiny defects can cause product failure and yield loss. By integrating LLM reasoning, Zero-shot Learning, the system enables real-time defect detection, anomaly analysis, and root cause inference. It can detect ultra-small cracks and irregular defects while supporting Edge AI deployment, real-time inference, and smart manufacturing applications.