ScholarWorks@성균관대학교

초록

This paper presents a novel deep learning framework for anomaly detection and reliability assessment in Overhead Hoist Transport (OHT) systems used in semiconductor manufacturing. To overcome the challenge of limited fault data, we utilize generative AI techniques to augment virtual fault instances that resemble real anomalies. These synthetic data are used to train a semantic segmentation model for identifying abnormal regions in OHT wheel images. Our approach aims to replace traditional visual inspections and scheduled maintenance with an automated, real-time fault detection system. Experimental results demonstrate the model's effectiveness in detecting current and potential detects, offering significant improvements in operational efficiency and downtime reduction in manufacturing processes. This methodology not only provides a way to detect industrial anomalies in extreme data situations, but also provides a scalable solution that can improve overall manufacturing reliability and productivity.