Unsupervised context-sensitive anomaly detection on streaming data relying on multi-view profiling

With the rise of Industry 4.0, many industrial assets are increasingly being monitored, generating vast amounts of data. At the same time, progress in machine learning and AI facilitates the exploitation of the gathered data for anomaly detection. Both are essential for improving condition monitoring of industrial assets. Current approaches typically consider a single asset at a time, while in reality, assets often operate next to one another. Thus, it may occur that the behaviour of one influences the behaviour of another or more assets. Moreover, current approaches are rarely addressing the detection and mitigation of dynamic changes induced by changing operating contexts. In this paper, we propose an unsupervised multi-view methodology for anomaly detection in streaming data, which embeds context-awareness to operating conditions. Results show that our approach is outperforming conventional non-contextual approaches. In addition, we demonstrate that our approach is able to maintain a high degree of interpretability which is enabled by splitting the data into two separate views and by keeping track of changes in the underlying operating behaviour. The real-world applicability of the devised anomaly detection workflow is demonstrated and validated on a real-world industrial use case.