Defect prediction of electronic equipment
From time to time, equipment shows a disfunctional behaviour in its start-up phase. Hundreds of sensors are available to monitor each operating phase of the equipment.
Customer wanted to: 1) identify sensors strongly correlated with the failure 2) anticipate occurrence of failures
Diagfit was used to learn the normal behaviour during the startup phase of a healthy equipment. A predictive model was then built that automatically identified variables/sensors leading to a disfunctionnal behaviour.
Among 1400 sensors, a dozen has been found to be the most relevant and used to build a failure prediction model. Operation was run on anonymised data both in terms of failure type, sensors meaning and time stamps (blind mode). DiagFit has been able to predict failures within the targeted anticipation window.