Knowledge Discovery for Avionics Maintenance Support

In avionics maintenance complex and time-consuming actions have to be taken to return faulty equipment to a fully functional state. The objective of this work is to support the maintenance activity by providing technicians with cues of actions to take, in order to repair a faulty component. We use an ontology to model avionics maintenance, and discover new concepts in the ontology characterizing the equipment failures. In a further step we associate these new concepts to a set of corrective actions, and we use them as suggestions to support the technicians in the diagnosis process. The method intends to explore only those concepts expressions that are relevant i.e. that are related to some sample. We provide our own algorithm for concept learning that allow us to explore a (potentially reduced) space of concept expressions, and to trace the reason (the properties in the samples) that leads us to select each expression. A prototype for avionics maintenance diagnosis support has been implemented, where given an equipment test as an input, the suggested corrective actions are returned as output. The prototype uses information from a Thales Avionics (France) repair shop, with whom we have developed the model and selected the data. The final implementation is hosted in Thales Research & Technology (France) using a BigData platform, allowing massive processing and remote access. In this paper, we introduce the use-case and the data and then position the solution, the most relevant notions and the algorithms used. At the end, we present the implementation of the prototype for our use case before concluding and the plan for the upcoming work.