Context
From time to time, certain anomalies occur in autopilot systems, causing discomfort for pilots during aircraft flights. As the autopilot system is monitored by several hundred sensors (400+), it is sometimes difficult and time-consuming for maintenance to make diagnostics to identify which sensor(s) are responsible for these inconveniences.
Need
Firstly, the customer wanted to know which sensors were most correlated with anomalies, so as to be able to diagnose his system. Secondly, for each anomaly, they need to know whether the beginnings of the problem can be identified by analyzing the previous flight(s), in order to avoid costly breakdowns and guarantee the safety of their pilots.
Solution
DiagFit was first used to find correlations between the 400+ sensors and the anomalies observed. All sensors were then ranked in order of correlation importance. Finally, 20 sensors with a correlation of over 80% were selected. These sensors were then used to build a model to identify weak signals of anomalies in the analysis of previous flights.
Results
The classified sensors enabled the customer to diagnose the root causes of the anomalies observed. The Research and Development team used the results to identify potential changes to the autopilot system to avoid such anomalies in future generations of aircraft.
As for the model, it enabled the maintenance teams to analyze every flight made to determine the risks for following flights. Not only did this reduce the number of costly breakdowns, it also ensured the safety of pilots by, for example, cancelling certain missions.
