The main objective of the study was to determine if better anomaly detection methodologies could be developed using transient engine performance data instead of steady state.
The feasibility of analyzing currently ignored transient engine flight data to evaluate engine performance and detect and/or diagnose engine failures and corresponding maintenance needs for the F108 engine was investigated by Southwest Research Institute (SwRI). The focus was gas turbine engine system-level health assessment based on performance data trending.
Why Study Transient Engine Data?
The main objective was to determine if better anomaly detection methodologies could be developed using transient engine performance data instead of steady state. If more information about the health of an engine is contained in high-stress, transient performance data, it will be possible to create methods that improve the detection, diagnostics, and prognostics of aero-propulsion gas turbine engines.
- A tool to read, format, and transfer data from the flight data recorder (FDR) files to a database was developed.
- The data was filtered to extract flight data recorded from throttle advance until the engine had stabilized.
- The transient data was taken under varying ambient conditions during multiple pilot requests via Power Lever Angle (PLA).
- The large amount of data was analyzed by an automated tool called the n-variable Statistical Process Control Tool (nSPCT™) that finds statistically anomalous observations that indicate current or impending failures.
- Maintenance records were searched to correlate performance issues with maintenance records.
High-stress transient data are more abundant than cruise data and provide more information regarding the health of an engine, but they are more difficult to preprocess for analysis. The resulting signals can be complex and require advanced algorithms for adequate fault detection. Cruise data are relatively easy to analyze with minimal preprocessing required to present the data to the detection algorithms. However, low-stress, steady-state conditions are required for cruise data acquisition. This limits the amount and usefulness of the data for analysis as these conditions occur much less frequently than transient conditions.
The engine faults that served as a test-bed for this analysis included fuel control problems, instrumentation issues, and compressor damage, all of which caused mission aborts. Despite the wide array of engine problems tested, the algorithms provided excellent results and early warning for the sample engine faults tested. For the test-bed of faults listed above, the algorithms provided a 0% false positive rate and a 0% false negative rate, resulting in 100% accuracy.
The results of this project provide a strong proof-of-concept for the analysis methodology, including the algorithms and type of data analyzed.
The algorithms developed, implemented, and applied during this project provide an accurate, automated fault detection system, especially when applied to transient data.
aircraft engine performance • transient engine data • engine data analysis • n-variable statistical process control tool • nSPCT • anomaly detection • engine flight data • flight data recorder • FDR