Two Phases Anomaly Detection Based on Clustering and Visualization for Plastic Injection Molding Data

The importance of smart factory is increasing. Among them, the importance of sensor data-based anomaly detection research is increasing. Accordingly, this paper presents a framework by using sensor data of plastic injection machine and deep learning models to perform anomaly detection and cluster secured outliers. This study aimed to be a practical solution and proposal for factories that want to introduce smartfactory system. According to this goal, three main contributions are assumed. The three main contributions are as follows. First, assuming a situation suitable for the manufacturing site, and appropriate approach was used. Second, it is possible to secure outliers by applying Auto Label using with pseudo labeling technique. Third, the decision maker can identify the potential cause of the defect. These three advantages can be found in the Two Phases Anomaly Detection system architecture. The artificial intelligence model used as a classifier in the pseudo labeling technique was LSTM and Accuracy showed high reliability of more than 90%. And through the SOM algorithm, clustering visualization of defective data is shown.