Study On The Nonlinear Aerodynamic Online Modeling Method Based On Wind Tunnel Flight Test

In recent years,with the emergence of various new concept aircraft,the traditional "serial iteration" development process has become inadequate to meet their development and iteration needs.The "parallel autonomous" development process can improve the safety,reliability,and design efficiency of aircraft,and non-linear aerodynamic online modeling is one of the key technologies.This technology can not only avoid inherent errors in wind tunnel testing and CFD calculations,improve the fidelity of aerodynamic models,but also accelerate the model iteration speed and shorten the aircraft development cycle.Therefore,studying aerodynamic online modeling technology is of great significance for the future development of aircraft.This thesis takes the International Common Research Model(CRM)as the research object,and the specific research content is as follows:(1)Establishes the CRM mathematical model based on three commonly used coordinate systems and transformation matrices.By analyzing the waveforms and spectra of various excitation signals,the characteristics and applicable scenarios of each signal were studied,and orthogonal multi-sine excitation designs for the three sets of control surfaces of CRM were carried out using Particle Swarm Optimization(PSO),resulting in excitation signals with relatively small relative peak factors.(2)For the quasi-online aerodynamic modeling method,offline model structure identification was first performed using Principal Component Analysis(PCA)and Stepwise Regression(SR).Then,online parameter identification was performed based on Recursive Least Squares(RLS)and Augmented State Extended Kalman Filter(ASEKF),respectively.The results show that both parameter identification algorithms can quickly converge and obtain accurate identification results under noise-free conditions,but ASEKF algorithm performs better under noisy conditions.In addition,a solution is proposed for ASEKF by substituting the estimation of and with their ratio /,which can also achieve good identification results.(3)For the fully online aerodynamic modeling method,two algorithms-"Recursive Orthogonal Functions plus Recursive Least Squares"(“ROF+RLS”)and Online Sequential Extreme Learning Machine(OS-ELM)were proposed,and their characteristics were analyzed through simulation experiments.The results show that both algorithms can perform real-time operations with extremely low latency and establish aerodynamic models with good fitting and prediction effects.Furthermore,the impact of hyperparameters on modeling performance was also investigated.(4)Two fully online aerodynamic modeling algorithms are verified by wind tunnel virtual flight test.The results show that both algorithms can quickly complete aerodynamic online modeling under test hardware conditions,and the obtained models can also predict aerodynamic data generated by other excitation forms,demonstrating their engineering feasibility.