Aero-Engine Exhaust Gas Temperature Prognostic Model Based on Gated Recurrent Unit Network

Abstract

Exhaust gas temperature (EGT), a key gas path parameter, is regarded as the performance indication parameter of the aero-engine. In engineering, the EGT prognostic model provides supports for aero-engine performance assessment, maintenance plan optimization and operation schedule determination. The EGT parameters are regarded as the time series with nonlinear characteristics, which should be considered in the prognostic model. To address these issues, an EGT prognostic model based on the gated recurrent unit (GRU) network was proposed in this paper. The time series and nonlinear characteristics could be addressed by the GRU network simultaneously. For better prediction accuracy, the architecture of the GRU layer was determined by contrast experiments, in which the GRU stacked layer number, look-back timestamps and output dimension were determined. The proposed prognostic model was validated by the real-valued EGT data of a turbofan aero-engine. Five conventional machine learning prognostic models were regarded as the comparison models. The comparison experiments showed that the proposed EGT prognostic model had advantages in prediction accuracy and stability. The proposed EGT prognostic model could provide supports for aero-engine prognostic and health management in engineering.