Prediction of Transonic and Subsonic Wind Tunnel Aerodynamic Data by Neural Networks

應用倒傳遞類神經網路預測氣動力試驗數據研究

Jun-Kai Ouyang(歐陽駿塏) ； Yi-Ting Liao(廖翊廷) ； Ying Li(李穎) ； Wei-Hsiang Wang(王威翔)

《中國機械工程學刊》 43卷6期 (2022/12) Pp. 509-520

摘要

This study aims to build the backpropagation neural network model to predict wind tunnel aerodynamic data. The experiments were performed to obtain the pressure fluctuation on compressible cavity flows in transonic wind tunnel and used different sweep angle delta wing model to obtain the aerodynamic data in subsonic wind tunnel, respectively. The experiments data is used as the training parameter for neural network to decide the neural network structure, tuning the adjustable hidden layers and neuron number parameters of the neural network. The Levenberg-Marquardt (LM) technique is adopted as the weighting training algorithm to minimum the cost function. This article have established the neural network model to provide good agreement with experiments result. By using neural network technique, the wind tunnel test efficiency and aerodynamic data analysis can be significantly improved.

關鍵字

neural networks ； wind tunnel ； cavity flow ； delta wing

並列摘要

本研究目的在建立倒傳遞類神經網路模型來分別預測穿音速風洞與次音速風洞試驗的氣動力數據，穿音速風洞以戰機內置彈艙外型簡化為簡單凹槽幾何之概念進行可壓縮流凹槽壓力量測試驗，並獲得動態壓力變化試驗數據，次音速風洞以不同後掠角角度的三角翼進行試驗，利用內置式力平衡儀量測模型的升力係數與阻力係數。分別將穿音速風洞與次音速風洞試驗結果作為類神經網路模型的訓練參數，因類神經網路訓練結果會因為不同的隱藏層／神經元數量而影響學習速率與正確率，本研究利用試誤法進行參數調整並決定所需神經網路架構，並以Levenberg-Marquardt演算法作為權重更新的方法以最小化誤差函數，本研究已完成倒傳遞類神經網路模型建立，經驗證後可應用於預測穿音速風洞凹槽模型特定位置於不同馬赫數條件之壓力變化，及藉由改變三角翼外型幾何參數預測次音速風洞不同後掠角角度的三角翼升力係數與阻力係數，所獲得結果比較試驗結果均具有一致性，利用類神經網路方法可減少風洞試驗前準備的前置作業時間，預測結果可作為氣動力數據分析的參考資訊。