Context
The customer is a company specialized in petroleum equipment, in particular manufacture of drilling tools for wells. These tools incorporate sensors in the drilling heads to measure various parameters (temperature, accelerations, vibrations, etc.), recorded by electronic acquisition cards.
Used dozens or even hundreds of meters below the surface of the sea, these cards cannot be inspected manually on a regular basis.
However, their failure can lead to costly machine downtime. Used dozens or even hundreds of meters below the surface of the sea, these cards cannot be inspected manually on a regular basis. The customer is therefore looking for an automated, remote analysis solution capable of exploiting the data transmitted by the cards to prevent and manage such incidents.
Résultats
Needs
The customer is looking for a reliable method of detecting and, if possible, anticipating faults in the electronic cards present in the drill heads. Early detection would reduce unwanted downtime and optimize maintenance operations. The aim is therefore to develop a model capable of identifying the early signs of failure from the voltage data collected in real time by the cards.
Solution
DiagFit was used to generate an anomaly detection model from the voltage measurements collected on the post-production board test benches. This model, designed to identify weak signals indicative of anomalies, was then applied to data from cards in actual field operation, in boreholes.
Results
Among the 12 boards studied as part of this project, 5 failures were detected, 4 of which showed warning signs that could be anticipated.
If the score is below the normality threshold, the electronic board is considered healthy, whereas if it is above the threshold, an anomaly has been detected on the board.
These results show some promising predictive capability for this type of failure. However, they also underline that these warning signs are not systematic : the physical reality of the boards reveals that some failures can occur without any detectable precursors in the voltage data.
If the score is below the normality threshold, the electronic board is considered healthy, whereas if it is above the threshold, an anomaly has been detected on the board.
Want to try DiagFit ?
