Development of Smart Factory Abnormal Detection Method Using Object Detection and AutoEncoder
Academic Society of Global Business Administration, Younsik Kno, Kwangpil Jeong, Junoh Kweon, Insu Cho, Yong Han Ju · Global Business Administration Review · 2022
Recently, artificial intelligence algorithms are being applied in various industrial fields. Under these circumstances, the manufacturing industry is also conducting research on artificial intelligence in areas such as classification, prediction, and monitoring. However, it is not easy for SMEs to invest in new areas such as artificial intelligence. Therefore, this study proposes an image-based process management system for the manufacturing system. In this paper, We utilize the object detection algorithms YOLOV5 and autoencoder in the direction of process anomaly detection. First, the object is extracted in real time from the information of the image taken at the mobile device level through the YOLOv5. Next, the extracted image is defined as an input value of the autoencoder, and a main feature is to be extracted. In particular, the performance of object detection was compared through scenario analysis and the amount of data required for learning. It is expected that the abnormality detection methodology presented in this study will contribute to the purpose of establishing a smart factory and improving productivity and yield of domestic manufactured SMEs.