Residual life prediction from statistical features and a GARCH modeling approach for aircraft generators

Abstract

Condition-based maintenance is currently widely used in the aviation industry with diagnoses obtained from the performance data of the aircraft. Online assessments of the real-time condition and predicted residual life have been of great importance for both mechanics and pilots, especially during flight for the latter. Statistical distribution and feature parameters are believed to be crucial criteria of performance degradation, which facilitate making practical component replacement decisions. Furthermore in terms of observations featuring performance degradation, time-series analysis provides feasible forecasts of residual life from the available working time of aero parameters. The recorded data from constant speed generator drives of aircraft generally demonstrate these characteristics, are non-stationary and have time-varying variance in time-series analysis. The generalized autoregressive conditional heteroskedasticity approach is appropriate to the situation to obtain prediction results. The suitability of the proposed method has been examined through calculating prediction errors with data from an actual life experiment of aviation generator.