Anomaly Detection for Injection Molding Using Probabilistic Deep Learning
Vili Ketonen, Jan Olaf Blech · 2021
Data collection from industrial devices is becoming more popular with the advent of technologies and trends such as the Industrial Internet of Things (IIoT) and Industry 4.0. We propose a deep learning-based approach to detect anomalies in real-time from multivariate time series data and interpret the detected anomalies' root causes. We apply the method to real-world industrial data collected from two injection molding machines. Additionally, we evaluate the method using artificially generated multivariate time series data. We compare the performance of the method to five baseline algorithms from the literature. Our results indicate that the method can detect anomalies in the injection molding machine data and interpret the root causes of the detected anomalies with high performance. Similarly, the method works well on the artificially generated multivariate time series data, demonstrating that the method is also applicable to other multivariate time series data problems.