While Industrial machines do publicly display abnormal behaviour from time to time, sometimes it can’t be detected even by a human expert.

This undetectable behaviour, potentially caused by problems such as structural defects, electronics malfunction e.t.c, can go unnoticed for long periods of time. All the while spoiling production output or causing more damage that could lead to the shutting down of the entire production line.

If there’s one thing that I believe machine learning can readily solve in industrial facilities, it is its use in identifying such anomalous behaviour.

Because as it turns out, Anomaly Detection, unlike Predictive Maintenance, doesn’t require huge amounts of multivariate data over extended periods of time to effectively predict.

All you need to do is to pass a stream of what you consider normal behaviour data through an ML algorithm, long enough to form a model that detects events which do not conform to an expected pattern.

In the video below I demonstrate how to use Azure Stream Analytics to detect an anomaly in temperature sensor data and send the results to a PowerBI dashboard to allow an expert to approve the deviation as an anomaly.