Reduced-Order Modeling of a High-Fidelity Propulsion System Simulation via Probabilistic Principal Component Analysis and Neural Networks

Abstract

Ever stringent aircraft design requirements on simultaneous reduction in fuel consumption, emissions, and noise necessitate innovative, integrated airframe designs that require concurrent engine designs. In order to ful ll these design challenges, aerospace engineers have relied on a physics-based engine modeling environment such as the numerical propulsion system simulation (NPSS). To expedite the use of NPSS in aircraft design, this research proposes a methodology for the reduced-order modeling (ROM) of NPSS by incorporating the following two techniques: probabilistic principal component analysis (PPCA) for basis extraction and neural networks for weighting coe cient prediction. To e ciently achieve an empirical orthogonal basis, this research capitalizes on an EM algorithm for PPCA (EM-PCA) to handle NPSS engine decks that typically lack some data due to failed o -design performance analyses. In addition, to e ectively explore a weighting coe cient space, this research utilizes neural networks to deal with six NPSS engine modeling parameters. As a proof of concept, the proposed NPSS ROM method is applied to an NPSS turbofan engine model usually employed for conventional civil transport aircraft. Comprehensive prediction quality investigations reveal that engine performance metrics estimated by the reduced-order NPSS model show considerably good agreement with those directly obtained by NPSS. Furthermore, the reduced NPSS engine model is integrated with the ight optimization system (FLOPS) in lieu of directly using NPSS as an illustration of the utility of NPSS ROM for aircraft design research.