HPart and Condition Extraction from Aircraft Maintenance Records

Abstract

Aircraft maintenance records contain vital information about airplane parts and their conditions in free-form text that are crucial health indicators of an aircraft. Extraction of these types of information is essential to improve safety, and lower lifecycle maintenance cost, and to minimize downtime and spare parts inventory. The task, however, is challenging as it is a domain-specific knowledge discovery problem that poses unique challenges in the field of information extraction which have not been studied much. This paper discusses these unique issues and challenges and how we approach them by adapting an advanced deep learning technique that has been widely used for information extraction tasks in other domains. The proposed system has good performance on extracting part names and conditions from noisy texts and is shown to be effective in processing data sets across diverse types of aircraft systems.