A Lightweight Hybrid Framework for Real-Time Detection of Process Related Anomalies in Industrial Time Series Data Generated by Online Industrial IoT Sensors

Abstract

Industrial Manufacturing plays an important role in the global economy, and estimates suggest that approximately 27 hours per month are lost in any major facility due to unplanned stoppages. The advent of Industrial IoT has seen manufacturing facilities deploy low-cost sensors with the hope of gaining improved visibility and thereby reducing unplanned machine stoppages and wastages. This paper introduces a lightweight, fast, easy-to-deploy framework that can be used for reliable and accurate identification of anomalies in real-time operational systems. The framework is holistic and includes data acquisition and data persistence modules to ensure that it can be deployed to a working production facility. The anomaly detection and contextualisation module is hybrid and uses statistical techniques and machine learning methods to provide fast responses while requiring minimal intervention. The framework was tested at a large metals manufacturing facility in New South Wales, Australia and the results show an accuracy of over 97% in anomaly detection.