Anomaly detection of technical equipment

NR conducts research on anomaly detection in time series’, including challenging scenarios where thousands of time series’ are analysed at once and in real time. The aim of anomaly detection is to forewarn unexpected behaviour in different systems, and prevent costly equipment getting damaged or harming the environment.

The image shows an example of a time series where detected anomalies are highlighted in red. The time series has a grey background and the activity is marked in black dots. — Figure caption: Processed temperature and pressure measurements of a pump over time. The red lines indicate where the algorithm has found consistent deviations over time. Figure: Martin Tveten

NR’s specialty is condition monitoring technical equipment. In these applications, sensors continuously monitor the condition of a machine, such as a water turbine or marine engine, and identify irregularities. This information is valuable as it can forestall damage, extend the lifespan of machinery, and automate and simplify maintenance work.

Unsupervised algorithms

We develop tailored solutions for various sized businesses in order to the meet the unique demands and challenges in their databases. A good example is that data or registered deviations aren’t typically available in order to “train” algorithms. In such instances, we develop and apply unsupervised algorithms. In addition, anomaly detection is sometimes critical for system security and for people’s safety. Methodologically, these scenarios demand a higher level of performance than in other cases.

A methodology for different types of numerical time series’

We have extensive experience with data from temperature, vibration and sound sensors, in addition to monitoring resource application in IT systems. On a general basis, the methodology can be applied to all types of numerical time series’.