Hits:

Affiliation of Author(s):能源与动力学院

Journal:Aerospace Science and Technology, 2017, 64, 223-236.

Key Words:Low-bypass turbofan engine Performance degradation LPV modeling Kalman filter Kernel extreme learning machine

Abstract:A key to achieve reliable model-based engine control, diagnostics and prognostics resides in in-flight engine model with high confidence level. Presented here is a new lifecycle real-time model to describe turbofan engine dynamic behavior called ALPVM (Adaptive Linear Parameter Varying Model), and the issues of engine/model mismatch compensation and performance degradation adaption are focused on. This methodology is different from the widely used STORM (Self Tuning On-board Real-time Model) presented by Pratt & Whitney, and the ALPVM is proposed on the linear parameter varying framework. The system matrices of ALPVM are computed using simultaneous step response data, and the polynomial LPV model is designed by the sets of polynomial fitting curves with scheduling parameters of engine operation in continuous forms. The IR-KELM (independent reduction kernel extreme learning machine) is developed to improve computational efforts without prediction accuracy reduction, and it serves an empirical model for polynomial LPV model mismatch compensation. The mechanisms of predict error control and linear dependency are considered in the IR-KELM, and it leads to decrease the hidden node number and simplify the IR-KELM topology. Kalman filter is employed to tune the health parameters of LPV model over its course of lifetime. Finally, the IR-KELM performance is confirmed by the benchmark data, and the simulation results from the ALPVM application to track a low-bypass turbofan engine dynamic behavior in the flight envelope indicate the effectiveness and usefulness of the proposed approach. (C) 2017 Elsevier Masson SAS. All rights reserved.

ISSN No.:1270-9638

Translation or Not:no

Date of Publication:2017-05-01

Co-author:Qian, Junning,Huang Jinquan,Qiu, Xiaojie

Correspondence Author:Feng Lu