The field of computational fluid dynamics (CFD) is integral to engineering disciplines, particularly for designing systems that operate under complex fluid flow conditions. Accurate simulation of flow fields is essent...
详细信息
The field of computational fluid dynamics (CFD) is integral to engineering disciplines, particularly for designing systems that operate under complex fluid flow conditions. Accurate simulation of flow fields is essential for optimizing performance across a variety of applications, including aviation, automotive, marine, and renewable energy sectors. Recent advancements in deep learning, particularly graph convolution networks (GCNs), offer promising alternatives for improving simulation processes. This work introduces a novel approach to accelerating fluid simulations using GCNs for flow field initialization. To this end, two different GCN models were employed, incorporating prior knowledge of the problem like its boundary conditions, as well as residual training. Extensive experiments using over 2000 sets of simulation results of various NACA airfoil shapes and flow conditions demonstrate that GCN-based initialization significantly reduces computational resources while maintaining high accuracy, achieving a 30% – 50% reduction in simulation time compared to conventional CFD initialization method.
Large-eddy simulations (LES) are carried out to investigate the shock induced transient flow through a planar nozzle mimicking a shock tube experimental setup at shock Mach number M-s = 1.86. A fifth order Weighted Es...
详细信息
Large-eddy simulations (LES) are carried out to investigate the shock induced transient flow through a planar nozzle mimicking a shock tube experimental setup at shock Mach number M-s = 1.86. A fifth order Weighted Essentially Non-oscillatory (WENO) scheme based 3D numerical flow solver equipped with an immersed boundary method and Wall-Adapting Local Eddy-viscosity (WALE) model is used for this purpose. A comparative study is presented to show the effect of different flow initializations namely, i) without flow fluctuation, ii) with white random noise and iii) with homogeneous isotropic turbulence superimposed on the flow-field. Results are in good agreement with the experimentally measured speeds of the primary shock wave and the following secondary shock wave. It is found that an improper initialization of the flow-field may lead to erroneous predictions of the flow characteristics, particularly the location of the separation point and the unsteady shock/boundary layer interaction. Substantial improvement in the prediction of the early-stage Mach reflection, complex shock/boundary layer interaction is observed with superimposed turbulent fluctuations as an initial flow-field. This is when a homogeneous incompressible isotropic turbulence superimposed on the shocked section is assigned as initial fluctuating field. An additional test case with the latter method, having an approximately four times higher mesh resolution, is used for detailed investigation of the unsteady flow fields, turbulent statistics and boundary layer separation. Results show considerable improvements in the prediction of the secondary shock and the flow separation location compared to the previous findings which dealt with lower mesh resolution. The turbulent flow structures are depicted using the mean flow-field which is spatially averaged over the span-wise homogeneous direction. Time-averaging on the fly turns out to be inadequate, since it leads to a spatial shift of the separation bubble
暂无评论