建立虛擬實境飛行動態模擬系統之現代戰機空氣動力模型

Abstract

現代的航空器系統對飛行員而言是既複雜又具挑戰的工作環境,而飛行訓練是危險性高、花費極為昂貴、人員培植不易且承擔不起重大損失的尖端科技類的訓練。因此,藉由地面的模擬飛行，達到實際飛行訓練的要求，不但節省訓練飛行員的花費亦可保障人員安全及航空器的損害。 目前虛擬實境在飛航器的模擬訓練上已經扮演了很重要的角色，而飛行模擬器性能的好壞，則取決於其是否能充分反應真實飛機的動態特性。本文研究的目的在於虛擬實境的系統上建立以國軍二代戰機為主的空氣動力數學模型，利用其風洞測試數據，進行動態模式數據處理分析並進而建立以常態多項式表示的氣動力空氣動力係數模型。藉由參數估測運算與空氣動力係數模型修正，以拉近飛行模擬器與真實飛機的動態響應間的差距，提高飛行模擬器的真實程度。 由於在一般非線性的模擬存在系統辨識中模式建立與參數估測問題，包括以一般數值解法的不易及因初始猜測值不當而可能導致之積分發散及收歛不良的問題，運用基因程式(Genetic Programming)【1】【2】的概念，一種啟發式自組織建立模型方法【3】（Self-organizing modeling）藉由用輸入變量及演化的多項式來近似地表示非線性系統的輸入輸出關係，配合適應函數（fitness function）選擇最佳結果作自我模式修正， 可快速建立空氣動力係數常態多項式數學模型。由於實際飛行資料取得不易，本文將藉由風洞測試數據來驗證本法在橫向與縱向飛行動態上其誤差在非線性空氣動力模擬應用上能接受的範圍之內，進而完成現代戰 機空氣動力模型。 本文最後將現代戰機空氣動力模型結合虛擬實境場景並利用TCP/IP 網路連線功能來與六軸運動平台控制器作連結展現戰機飛行姿態和力的表現，未來並整合力操控搖桿成為一完整之現代戰機動態模擬訓練系統。

Modern aircraft is quite complicated and poses challenge to pilots. Since flight training is highly risky and extremely expensive, flight training by simulation not only saves costs but also avoids the damage to personnel and aircraft. The aim of this thesis is to develop a visual simulation pilot training simulator with real-time dynamic model for air fighter. The criteria for judging the performance of flight simulation focuses on the fidelity of the simulation to the actual aircraft. To reduce the difference between a flight simulator and an air fighter that a technique was developed for global modeling of nonlinear aerodynamic coefficients by analyzing the subsonic wind tunnel data for the air fighter. The ordinary polynomial of the modeling functions allowed straightforward determination of an adequate model structure and the associated parameter values by cost function and parameter estimate. Normally nonlinear systems exhibit two major problems: system identification and divergence due to numerical methods and bad initial parametric guess. A selforganizing method in modeling based on genetic programming is used for wind tunnel data analysis and to construct polynomial functions. Minimum predicted squared error criterion is used to determine which polynomial functions should be retained in the model. Each retained polynomial function is decomposed into an expansion of ordinary polynomials in independent variables so that the final model can be interpreted as selectively retained terms from a multivariable power series expansion. The resultant polynomial model demonstrates good predictive capability that provides insight for the character of the nonlinear aerodynamics. In system integration that to construct the complete air fighter pilot training system, we bring the VR scene、air fighter aerodynamics model and TCP/IP as the interface between the VR scene and Stewart platform to perform the 6-DOF posture of platform by manipulating force-feedback joystick to feel fighter dynamics.