Equipment was occasionally exhibiting dysfunctional behaviour in its start-up phase. Hundreds of sensors were available to monitor every phase of the equipment’s operation.
The customer wanted to:
1) Identify sensors that were highly correlated with the failure
2) Anticipate the occurrence of failures
DiagFit was used to learn the normal behaviour during the start-up phase of a healthy piece of equipment. A predictive model was then built that automatically identified the variables/sensors leading to dysfunctional behaviour.
Out of the 1,400 sensors, a dozen proved to be the most relevant and were used to build a failure prediction model. The run was performed on data anonymised both in terms of failure type, sensor significance and time stamp (blind mode). DiagFit was able to predict failures within the targeted anticipation window.