Daniel Joseph Cleary

A novel approach to aircraft engine anomaly detection and diagnostics

Accurate and timely failure detection and diagnosis is critical to reliable and affordable aircraft engine operation. This work describes a statistical and fuzzy logic based approach that analyzes multiple engine performance parameters for trend recognition, shift evaluation and failure classification. It integrates the statistical data analysis and fuzzy logic reasoning processes and provides powerful data fusion capability. The system captures and diagnoses failures as soon as the engine performance-shifting trend is recognizable, based on customizable probability. This approach improves upon current diagnostic processes in a number of ways. First, the dimensionality is increased so that multiple relevant parameters are integrated into the diagnosis. This helps reduce single dimension false alarms. Second, this approach effectively handles the noise in engine performance data. Many diagnoses depend on detecting changes in the data that fall within three standard deviations of the pre-event data, historically leading to false alerts and diagnoses. Finally, this approach seamlessly integrates the noise in the data with the uncertainty in the diagnostic models, rolling it up into a single score for each potential diagnosis. This increases consistency, and removes a substantial amount of subjective judgment from the diagnostic process. This approach has been successfully applied to a series of General Electric commercial airline engines, demonstrating high accuracy and consistency. The methodology is expected to be generally applicable to a wide variety of engine models and failure modes.