Using neural networks and the rank permutation transformation to detect abnormal conditions in aircraft engines

Real world aircraft engine (gas turbine) data are contaminated with substantial noise and outliers. The rank permutation transformation (RPT), founded in some early ideas in statistics, is proposed as a way to both diminish the effect of noise and outliers, and to facilitate classification by making statistically unlikely events more pronounced. The RPT is also found to improve the performance of neural networks used for fault detection and classification considerably. Results from both real engine monitoring data for abnormal condition detection and high-fidelity simulation data for on-wing fault detection and diagnosis are presented.