For using online data to diagnose the degradation status of commercial aero-engine components

an aggregated method including engine steady state modeling and unscented Kalman filter (UKF) was proposed. First

the method represents the general characteristic curves of components with nonlinear multivariate functions with respect to their working conditions

and uses machine learning methods to confirm function coefficients that fit the observed data best

subsequently acquires the steady working model of aero-engines. Then

by analysis of individual degradation properties and effects on entire engine performance of components

the key degradation factors of components were defined. Moreover

the standard UKF process was modified

by adding moving window sampling strategy to overcome low-observability problem

and replacing standard observation equation with combined model of engine steady state model and engine working conditions. At last

the acquired filter was used to process observed engine data to track the trends of component degradation factors. As indicated by the test on commercial engine dataset

the proposed method can conveniently diagnose the health status of engine components once the observed report returns to on-ground data centers

and the acquired results coincided well with the internal damage information of engines collected by visual inspection. The proposed method performed well in overcoming modeling nonlinearity

low-observability

and high data noise problems in aero-engine performance diagnosis and showed to be applicable to industrial conditions.