Reduced-order Modeling Methods For Unsteady Aerodynamics And Fluid Flows

With the need of precise design for advanced flying vehicles,developing high-fidelity tools for numerical simulation and efficient methods for multidisciplinary analysis has become a critical problem to current research community.As an important tool for aerospace engineering,computational fluid dynamics(CFD)enables accurate numerical simulation for nonlinear,unsteady and complicated flow phenomena,and is capable of coupling with other physical fields(e.g.,structure,heat,control system,etc.)to achieve multi-physics analysis.However,CFD simulation requires a large computational cost with a huge amount of resource to process flow big data,which limits the applicability of CFD techniques.To maintain a balance between accuracy and computational efficiency of CFD,since 1990 s,unsteady aerodynamic reduced-order models(ROMs)based on CFD method have been developed,which achieve a model abstraction of complex fluid dynamics and feature extraction of large and high-dimensional flow information.ROMs have similar accuracy with CFD solver,but are able to reduce one or two orders of magnitude of the computational cost compared with that of CFD simulation.Therefore,they have the potential of being a promising means of future aircraft design involving flight dynamics,aeroelasticity,uncertainty qualification,optimization and system control.This thesis focuses on the development of ROMs for unsteady aerodynamics and unsteady flows,and works on the following problems:(1)To improve the accuracy and generalization capability of nonlinear ROMs based on system identification,aerodynamic ROMs based on recursive radial basis function(RRBF)neural network are investigated.Important parameters that affect the model performance are analyzed firstly.After that,with the help of evolutionary algorithm,dimensionality reduction and multiple basis functions,several training algorithms to improve the nonlinear ROMs' performace are proposed.This allows a deeper learning from training data and leads to better prediction on complex dynamic behaviors.(2)To achieve the consideration of both linear flutter and nonlinear LCO behaviors in aeroelastic analysis,three ROM frameworks are presented by combining both linear and nonlinear system identification methods.Through incorporating both nonlinear neural networks and linear autoregressive with exogenous input(ARX)model with different frameworks and training methods,the developed ROMs are able to describe LCO responses and flutter boundary of nonlinear aeroelastic systems and eventually simulate many complex aeroelastic phenomena.(3)To analyze physical mechanisms of unsteady flows and reproduce the evolution of high-dimensional flow fields,application and improvement on dynamic mode decomposition(DMD)is studied.Through developing a flow mode test technique based on DMD,the existence of the lowest Reynolds number for vortex-induced vibrations of flow past a circular cylinder with elastic support in subcritical flow conditions,is explained.Since standard DMD has difficulties in capturing the dominant modes,an improved criterion for DMD mode selection is proposed.This makes DMD reproduce and predict the transient states of transonic buffet flow with an improved level of accuracy.This thesis provides efficient mathematical tools for modeling nonlinear unsteady aerodynamics and flows,and validates the efficacy of using these models to unsteady aerodynamic and nonlinear aeroelastic problems.