Efficient aerodynamic derivative calculation in three-dimensional transonic flow

Abstract

One key task in computational aeroelasticity is to calculate frequency response functions of aerodynamic coefficients due to structural excitation or external disturbance. Computational fluid dynamics methods are applied for this task at edge-of-envelope flow conditions. Assuming a dynamically linear response around a non-linear steady state, two computationally efficient approaches in time and frequency domain are discussed. A non-periodic, time-domain function can be used, on the one hand, to excite a broad frequency range simultaneously giving the frequency response function in a single non-linear, time-marching simulation. The frequency-domain approach, on the other hand, solves a large but sparse linear system of equations, resulting from the linearisation about the non-linear steady state for each frequency of interest successively. Results are presented for a NACA 0010 aerofoil and a generic civil aircraft configuration in very challenging transonic flow conditions with strong shock-wave/boundary-layer interaction in the pre-buffet regime. Computational cost savings of up to 1 order of magnitude are observed in the time domain for the all-frequencies-at-once approach compared with single-frequency simulations, while an additional order of magnitude is obtained for the frequency-domain method. The paper demonstrates the readiness of computational aeroelasticity tools at edge-of-envelope flow conditions.

Gust loads on aircraft

“…Thus, CFDbased ROMs generally can be obtained from either or both of the two aspects. So far, many order reduction techniques or algorithms have been developed and applied such as the convolution integral (135)(136)(137) , Auto-Regression Moving Average (ARMA) (98,(137)(138)(139)(140) , Eigenvalue Realisation Algorithm (ERA) (141) , Linearised Frequency-Domain (LFD) (142)(143)(144)(145) , Proper Orthogonal Decomposition (POD) (146)(147)(148) , Alternative Kriging Approach (149) , Sherman-Morrison-Woodbury (SMW) formula (125) , and Volterra theory (150,151) .…”

Aerodynamic Data Predictions Based on Multi-task Learning

“…Limited high-speed data are submerged in lots of low-speed data, which leads to the inability for neural networks to predict high-speed aerodynamic data. This phenomenon, more common in unsteady flow fields [2], increases the difficulty of modeling and also reduce the accuracy of models.…”