CASE STUDIES

They raised concerns over a lack of ability to predict failure within their systems. Fixing problems in retrospect was impacting productivity, increasing recovery costs and causing negative environmental influences. The aim of the project was to predict when maintenance was required, eliminate false alarms and help to identify and reduce issues with faulty sensors.

Using historical telemetry data and incident reports, our goal was to investigate the possibility of predicting the triggering of alarms and anticipating the necessary maintenance ahead of time. We were also provided network data from the previous 4 years, including sensor measurements and alarm logs, to investigate reconstructing missing or incorrect measurements from faulty sensors.

Our data model was able to show that it was possible to anticipate alarms and allow preventative maintenance. It also proved it was possible to reconstruct and correct measurements from faulty sensors using nearby/related sensors. We were able to build universal models of network behaviour that were then used to predict sensor data and network behaviour. We were able learn rules for alarm triggering within the systems and use these models to anticipate alarms indicating approaching equipment failure.

We successfully predicted future sensor measurements in order to decide when an alarm would be triggered. We also showed that it is possible to predict future measurements for sensors using AI & Data Science models.

Using propagation algorithms, we were able to divide the sensors into groups with similar behaviours and use sensors within each group to infer the measurements from one of them (a “faulty” sensor) using the other ones.