스마트 모니터링 및 유연 산업용 로봇을 위한 제조 인공지능

Abstract

Smart manufacturing aims to achieve autonomous intelligence such as self-recognition, self-adoption, and self-decision making. As one of recent advancements of information and communication technologies (ICT), artificial intelligence (AI) enables to train such autonomy, resulting in smarter recognition and improved flexibility of manufacturing systems and processes. In this presentation, the research outcomes of AI for manufacturing are introduced. For smart process monitoring, Convolutional Neural Network (CNN) and Autoencoder (AE) with an internal sound sensor was utilized to estimate productivity of machines and to detect anomaly of industrial robots. Also, the method was deployed to factories across U.S.A. and Korea for transition to smart manufacturing. In addition, CNN-based robotic bin picking method in randomly cluttered space are presented as applications for flexible robotic manufacturing. The research shows how the accuracy of CNN for object localization in bin picking task was improved by self-training. Next, YOLOv5 neural network for the same purpose was trained by simple human demonstration and data augmentation, reducing data collection and annotation for training the model.