Analyzing Aircraft Maintenance Findings with Natural Language Processing

Abstract

Aircraft maintenance at the technical base includes both scheduled and unscheduled repairs. When performing repairs, there are detailed inspections of aircraft, which are ordered in advance by the customer. During these inspections, it is possible to detect other unsatisfactory condition of the aircraft in terms of continuing airworthiness, which needs to be corrected. These unplanned failures are findings and are entered into the maintenance organization's system, which is required to keep records of activities performed on the aircraft. Finding descriptions are created by technicians and are usually stored as unstructured text that does not allow subsequent analysis with a database of such records. The presented work performs automatic text analysis of findings using a dedicated software tool. For the purpose of the analysis, a lexicon was created, with components and failures appearing in maintenance records. The lexicon was used in automatic text analysis, where the output was a statistic of the most frequent components and failures in the stored maintenance data. This statistic can be used for maintenance planning optimization and better prediction of unscheduled repairs.