Accurate and timely detection and identification of aircraft engine faults is critical to keeping the engine and aircraft in a healthy operating state. Early detection of faults increases the window of opportunity to schedule maintenance actions both at a convenient time and before the fault progresses and causes equipment downtime and secondary damage to the system. Typically, diagnostic models are built using parametric sensor data to infer the state of the system. However, recording and collecting this data is costly, and it is generally limited to a few snapshots over the course of a flight for commercial aircraft. Another way to recognize faults is through the use of built-in tests that produce error log messages. These tests produce data that is less information rich, but provide insight over the course of the entire flight. Each data source provides a different perspective of the state of the system. Therefore, it may be advantageous to combine information from parametric and nonparametric sources to improve fault diagnosis in terms of accuracy and timeliness of diagnosis. In this paper, we investigate integrating parametric sensor data and nonparametric information in fault diagnosis, specifically the way to parameterize nonparametric information for use in diagnostic models that accept only parametric data (e.g., most machine learning techniques). Results from high bypass commercial engines are presented.

This content is only available via PDF.
You do not currently have access to this content.