Shulou(Shulou.com)11/24 Report--

CTOnews.com, November 7, according to the Rong Media Center of the propaganda Department of the Yanta District CPC Committee, the 2023 China (Xi'an) artificial Intelligence Summit Forum was held on November 4 at the High-tech International Conference Center of Yanta District, Xi'an. At the meeting, Qinling soaring, the first large model of hydrodynamics for aircraft, jointly developed by Northwestern University of Technology and Huawei, was officially released.

Professor Zhang Weiwei, School of Aeronautics, Northwestern Polytechnic University, Ding Cheng, General Manager of MindSpore Business and Director of distributed parallel Computing Laboratory of Huawei Central Software Institute, and CEO Qi Haiping of Xi'an Pengteng Intelligent Technology Co., Ltd.

The large model of "Qinling Flying" is an intelligent model for aircraft fluid simulation, which is jointly developed by the International Institute of fluid Mechanics Intelligence of Western University of Technology and Huawei AI4Sci Lab on the basis of the domestic open source fluid computing software Storm, relying on the surging computing power of Teng AI and the framework of MindSpore AI. The official introduction to CTOnews.com is as follows:

The large model aims to realize the accurate prediction of the whole scene flow field by creating an intelligent and general fluid mechanics software platform and the full scene application base of the fluid industry. At the same time, combined with the industry-leading data assimilation, AI turbulence model, rapid flow field prediction and other technologies to support the infrastructure of the large model of hydrodynamics.

Specifically, the large model adopts self-developed multi-level distributed parallel adaptive framework, multi-level integration of the classical theory of fluid mechanics and artificial intelligence methods, the construction of mathematical and physical correlation features, the development of multi-paradigm integrated modeling, the construction of invariant realizable multi-modal unified framework. At the same time, innovation and verification are carried out in the design of model algorithm, acceleration of mixing accuracy, and coupled parallel optimization of numerical solution, and the reconstruction of flow field with high confidence, the solution of turbulent flow field in full velocity domain and the near real-time prediction of complex flow field are realized.

Large hydrodynamic model module 1 of "soaring in Qinling Mountains": AI turbulence model

AI turbulence model is an artificial intelligence turbulence model for engineering with high Reynolds number. This module can completely replace the traditional partial differential model in the form of data-driven and physical embedding. Based on the leading modeling framework and deep understanding of fluid characteristics, the AI turbulence model can solve the multi-scale complex turbulent flow field with higher accuracy. In the case of two-dimensional S809 airfoil separated flow at high angle of attack, compared with the experimental data, the average relative error of the AI turbulence model is less than 5%, and the simulation accuracy is more than 3 times higher than that of the traditional SA turbulence model. In the case of transonic flow around a typical three-dimensional wing-body combination with tens of millions of grids, the average aerodynamic error is less than 1%, and the solving efficiency of the turbulence model is improved by more than 4 times.

"Qinling soar" hydrodynamics large model module 2: flow field prediction large model

The large model of flow field prediction is a fast prediction model of end-to-end complex flow field. Based on the deep integration of deep learning method and hydrodynamics theory, the module carries out flow field modeling from the perspective of transform domain, and realizes the efficient and high-precision prediction of the complex flow field of two-dimensional airfoils and millions of grid wings. The large model of flow field prediction includes end-to-end modeling, near real-time reasoning, wall turbulence characterization and full flow field prediction. In a typical case, the large model of flow field prediction shortens the time of a single wing sample from hours to seconds, achieves five orders of magnitude acceleration, and the relative error between distributed load and concentrated force is less than 1%.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.