Failure Root Cause Determination Through the Aircraft Fault Messages Using Tree Augmented Naive Bayes and k-Nearest Neighbors

This paper presents a method to determine the root cause of an aircraft component failure by means of the aircraft fault messages history. The k-Nearest Neighbors (k-NN) and the Tree-Augmented naive Bayes (TAN) methods were used in order to classify the failure causes as a function of the fault messages (predictors). The contribution of this work is to show how well the fault messages of aircraft systems can classify specific components failure modes. The training set contained the messages history from a fleet and the root causes of a butterfly valve reported by the maintenance stations. A cross-validation was performed in order to check the loss function value and to compare both methods performance. It is possible to see that the use of just fault messages for the valve failure classification provides results that close to 2/3 and could be used for faster troubleshooting procedures.