![]() ![]() optimization under subsonic, transonic, and hypersonic flow using GP. Also, GP is widely used in optimization, such as Jouhaud et al. constructed multiple response surfaces for airfoil design with a multiple-output Gaussian-process-regression model to predict lift, drag, and pitching-moment coefficients. utilized it to predict prediction in fast fluid structure interaction simulations. The Gaussian process (GP) regression, also known as Kriging in geostatistics, and the surrogate model in the fluid mechanics and computer experiments field, is a non-parametric statistical model which has been extensively used in various cases. In the era of big data, data-driven methods tell many successful stories while leaving many challenges. Researchers have performed machine learning techniques to improve the accuracy of Reynolds-averaged Navier–Stokes (RANS) models. So far, there have been many applications of machine learning in the field of fluid mechanics. Once the model learns the aerodynamic features, it holds the fast prediction ability, which satisfies the unseen data from a validation dataset. Unlike CFD methods who strictly follow the physical laws, ML can blindly do pure input-output (I/O) mapping, without incorporating any priori knowledge. The techniques of machine learning and the increase of numerical simulation data amount naturally lead to the applications of data-driven modeling in physical systems, including the aerodynamic design. With the rise of machine learning (ML) theory and computer science, there has been a rapid development in the field of data processing. Therefore, there is a demand for a fast and accurate pressure predicting method. For example, an inverse problem requiring possibly thousands or more simulations, the overall computation time would be orders of magnitude higher than real-time requirements. ![]() However, CFD simulation is relatively time-consuming. Generally, computational fluid dynamics (CFD) solver handles the computation of aerodynamic forces by solving Navier-Stokes (NS) equations. A successful target pressure distribution renders favorable reasonable geometry and aerodynamic characteristics. In the aerodynamic design optimization, specifying airfoils with well-performance pressure, distributions at the cruise point is an important aspect. Given the unseen airfoils from the validation dataset to the trained model, our model achieves predicting the pressure coefficient in seconds, with a less than 2% mean square error. Furthermore, we utilize a universal and flexible parametrization method called Signed Distance Function to improve the performances of CNN. Given the airfoil geometry, a supervised learning problem is presented for predicting aerodynamic performance. Therefore, this paper presents a data-driven approach for predicting the pressure distribution over airfoils based on Convolutional Neural Network (CNN). As an alternative, deep learning approximates inputs-outputs mapping without solving the efficiency-expensive physical equations and avoids the limitations of particular parameterization methods. Surrogate modeling can leverage such expense to some extent, but it needs careful shape parameterization schemes for airfoils. Conventionally, the pressure distribution is solved by computational fluid dynamics, which is a time-consuming task. In the aerodynamic design, optimization of the pressure distribution of airfoils is crucial for the aerodynamic components. ![]()
0 Comments
Leave a Reply. |