Enabling Proactive Quality in Commercial Airplanes using Natural Language Processing

Abstract

Quality management systems traditionally draw insight from structured, often numerical, sources of data; unstructured, free-text representations of quality data are less frequently employed despite having high informational value, and often require additional human effort to prepare their contents for use. An ability to extract and proactively employ this information enables a richer analysis of quality performance.

The primarily free-text reports generated by Boeing Commercial Airplane's "in-service investigation" (ISI) process are taken as an example of such quality data. We investigate both an unsupervised clustering method and a supervised classification method to group these reports by the broader "quality topic" they pertain to, using semantic relationship-maintaining text "embeddings" as features. We find success in supervised classification, and describe a method to relate ISI records with quality records from other parts of the commercial airplane value stream via standardized "code" metadata.

We extend the use of similarity techniques to investigation execution and propose a "helper" tool that automates parts of the manual data collection and relationship-finding process. The benefits of using such a tool over traditional keyword searches are described through an illustrated example.