data mining; machine learning; aircraft health monitoring; component failure prediction; apprentissage automatique; surveillance de l'état de l'aéronef; prédiction des pannes
The operation and maintenance of modern sensor-equipped systems such as passenger aircraft generate vast amounts of numerical and symbolic data. Learning models from this data to predict problems with component may lead to considerable saving, reducing the number of delays, and increasing the overall level of safety. Several data mining techniques exist to learn models from vast amount of data. However, the use of these techniques to infer the desired models from the data obtained during the operation and maintenance of aircraft is extremely challenging. Difficulties that need to be addressed include: data gathering, data labeling, data and model integration, and model evaluation. This paper presents an approach that addresses these issues. We also report results from the application of this approach to build models that predict problems for a variety of aircraft components.