Although CFD tools are necessary for aerodynamic design, it is still time consuming and this is one of the big problems. In order to shorten the computational time, a fast CFD code "FaSTAR" has been developed. We combined two acceleration techniques. One is a convergence acceleration technique such as the multigrid method. A coarse-grid generation method using octree data of the Cartesian grid is proposed. This method is simple and it can generate high-quality coarse grids. The other is a programing technique such as data structure improvement and performance tuning. We demonstrate the high-speed performance of the code for aerodynamic computation of a standard aircraft model, NASA-CRM. The computational time is less than one hour using 10 million cells and 100 CPU cores. It is found that the multigrid method is beneficial for large-scale problems.

To shorten the time to simulate fluid flows, we develop a fast automatic grid-generation tool, HexaGrid, that produces Cartesian grids with body-fitted layers. The objective of the present study is to apply automatic grid generation to the drag prediction of the NASA Common Research Model. First, we generate a grid following the 4th Drag Prediction Workshop gridding guidelines as far as possible and discuss the capabilities and limitations of the automatic method. Then, we validate the computational-fluid-dynamics results computed with the grid by comparison with other solvers and grid generators. The HexaGrid results agree well with the other results, with the differences in predicted drag being less than five counts except at the stall angle. Additionally, we compare separated flows at the stall angle. The separation lines and C-p distribution are found to be greatly affected by grid topology.

This paper presents an on-going study of grid generation within the context of CFD simulation for wind tunnel test. The objective is to devise a grid generation method that can support CFD simulation within the time scale of wind tunnel experiments. The challenges are multiple. From the geometry side, we need to handle complex geometry which consists of aircraft model, support system, and wind tunnel walls. From the physics side, we need to be able to handle Navier-Stokes equations and turbulence models. Finally, the time constraint is in the order of an hour. All these requirements pose a very tall order for grid generator. Our grid generation method uses a hybrid of Cartesian grid (for region far from body surface) and prismatic grid (to resolve boundary layer). The method is fully automatic and fast. Our previous study has shown that the resulting flow solution is competitive with that of grid generated manually by expert. The method has been improved to be even more resilient towards defects in surface geometry representation. This property has allowed us to generate grid without having to clean up the surface, which translates into significant time saving.

In order to develop a fast CFD code, we investigated well-known unstructured CFD codes, and we have determined the target performance and specification of CFD code. This is then followed by the development of a fast unstructured CFD code “FaSTAR”. The accuracy of drag prediction with FaSTAR is validated by the DPW4 benchmark problem. The computed drag coefficients generally agree with other results. Since we employ the Cartesian-based unstructured grid generated with HexaGrid, the predicted drag is affected by the choice of discretization method, whether it is cell-center or cell-vertex. The computational speed of present CFD code is 1.8 hour/case with 10 million grid and 100 cores for a standard civil aircraft. Although we have not achieved the requirement of 1hour/case, we believe that it is possible to achieve the target using the multigrid. The preliminary multigrid result shows four times faster convergence with global coarse grid.

In this study we proposed a robust method to automatically generate grid for Navier-Stokes flow computation, even when the input solid surface geometry contains some defects such as small gaps, overlaps and degenerate triangles. Such defects are often found in surface represented in STL (Stereo Lithography) format, which consists of triangles. This capability means that surface clean-up prior to grid generation is minimized, which is a significant advantage, as it saves a lot of time and manual labor. The grid itself is a hybrid of Cartesian grid (for region far from body surface) and prismatic grid (for boundary layer). Of particular interest is a new technique to handle concave feature of body surface. An approximation technique that minimizes reliance on the original surface geometry (which may contain defects) has been devised. Results have shown that the technique is effective. This opens up the possibility to carry out the whole process of Computational Fluid Dynamics fully automatically.

Along with the system development of “Digital/Analog-Hybrid Wind Tunnel” at JAXA, a fast CFD code is required. Based on the investigation of well-known unstructured CFD codes, we have determined the target performance and specification of CFD code. This is then followed by the development of a fast unstructured CFD code “FaSTAR”. The accuracy of drag prediction with FaSTAR is validated by the DPW4 benchmark problem. The computed drag coefficients generally agree with other results. Since we employ the Cartesian-based unstructured grid generated with HexaGrid, the predicted drag is affected by the choice of discretization method, whether it is cell-center or cell-vertex. Moreover, the reconstruction method is found to be important. This affects the drag prediction accuracy and the smoothness of surface pressure distribution. The computational speed of present CFD code is 1.8 hour/case with 10 million grid and 100 cores for a standard civil aircraft. The multigrid method with the global coarse grid shows four times faster convergence than that with the zonal coarse grid. Here, the coarse grids are generated utilizing the octree data of Cartesian grid.

Our objective in this study is to explore an automatic grid generation method for accurate Navier-Stokes flow computation, which can directly use the input of stereolithography (STL) data with defects, that is, without prior clean-up. This approach promises a significant improvement in CFD workflow, because the time and labor required for surface geometry preparation is very significant. Preliminary results suggest that the method is indeed promising. Progress is being made, especially regarding sharp feature capturing, as presented below.

To shorten the time to simulate flow, we develop an automatic and fast grid generator, HexaGrid, that produces Cartesian/prism hybrid grids. The objective of this study is to apply it for drag prediction on NASA Common Research Model (CRM). First, we compare the generated grid with the DPW gridding guideline and discuss about the capability and limitation of HexaGrid. Then, we validate computational results with HexaGrid, comparing with other solvers and grid generators. HexaGrid results agree well with the other results, where the difference of predicted drag is less than 5 counts except for the stall angle. Additionally, we compared the separated flows at attack angle of 4°. The separation lines and Cp distribution are largely affected by grid topologies.

To improve efficiency of CFD analysis, we develop an automatic and fast grid generator, HexaGrid, that produces Cartesian/prism hybrid grids. We applied HexaGrid for the NASA common research model employed for the fourth drag prediction workshop (DPW4). We investigated the grid dependency of the separated flow at attack angle of 4°. In particular, we focus on the effect of wake resolution behind the main wing. The grid in the wake region was refined using the refinement box of HexaGrid. The wake is smeared for the original grid, whereas it is clearly captured over the span for the grid with refinement box. However, the side-of-body separation bubble and its wake are not changed significantly. Therefore, the force and moment are almost same. Finally, although the refinement box does not change the result for this problem, it is useful for the problem that includes the flow interaction between main and tail wings.

Digital Wind Tunnel (DWT) is being developed in a project entitled Digital/Analog Hybrid Wind Tunnel in JAXA. The main objective of this project is to sophisticate the transonic wind tunnel in JAXA by applying IT technologies. We have several hurdles to establish DWT such as the developments of an automatic grid generation tool and a high performance CFD solver. In this paper, the prediction accuracy of aircraft loads by using the automatic grid generation tool and the modeling of porous wall for the CFD calculations of whole wind tunnel including an aircraft model, support instruments, and walls are discussed.

To shorten the time to simulate flow, we develop an automatic and fast grid generator, HexaGrid, that produces Cartesian/prism hybrid grids. Since the prism cells are also hexahedra, the generated grid is hexahedron-dominant. The objective of this study is to extend the method so that it can produce grids with sufficient quality for Navier-Stokes simulation. The transonic flows around ONERA-M6 wing and DLR-F6 wing-body configuration are computed using HexaGrid and an unstructured-grid flow solver, JTAS. The lift and drag computed with the grid are compared with others’ results. The agreement is reasonably good, which shows that HexaGrid becomes a powerful tool to accelerate a CFD process.

The method outlined here is capable of automatically generating hexahedra grid around solid surface described as overlapping components. This offers a significant benefit in design environment, because the geometry of individual component and their relative positions can be readily modified. The method is based on Cartesian grid method, which is well known for its efficiency and speed in filling a computational domain. The cells near solid surface are hexahedral cells of general shape, which includes degenerate forms such as pyramid. Because the cells are generated from the interior of computational domain toward the solid surface, the method tolerates surface defects such as small gap and overlap, which are sometimes found in surface grid obtained from CAD. In this study this property is further extended to direct treatment of a configuration consisting of intersecting surface components, that is, without trimming and regenerating the whole surface. The geometry of DLR F6 with engine and pylon has been used to demonstrate the capability of the method.

A hybrid gri d generation method based on prismatic and Cartesian grid is being developed. The main objective is to generate grid around c omplicated geometry for compressible and viscou s flow simulation , especially that with high Reynolds number . The method is based on unstructured grid, which treats cells as arbitrary, unstructured polyhedra. By invoking a combination of Cartesian, prismatic, and cut cell generation modules, user can produce grid for specific simulation needs. At present, development is focused on pris matic grid generation. An advancing layer strategy that incorporates a number of advancing steps to make a prismatic layer is discussed, along with a smoothing technique. Results seem to be favorable, although computation can be time consuming. 1. Introduct ion As CFD plays an increasingly greater role in flow analysis and design, it is demanded to handle even more realistic cases. One of the cases that still pose a computational challenge is simulation of viscous flow at high Reynolds number. At this condit ion isotropic grids normally employed in inviscid and low - Reynolds -number flow simulation will perform very poorly, especially in region close to body surface. When the grid is non -body -fitted, as in the case of Cartesian grid, the situation becomes worse , due to irregularity in cell size at body surface. This is true even when anisotropic grid is employed. Thus a hybrid grid method to carry out such simulation is being developed in National Aerospace Laboratory (NAL), Japan. This study is the continuatio n of the previous in anisotropic Cartesian grid for inviscid flows. 1

Simulation of the flow around a moving body is a challenging problem, especially in 3D. The type of grid employed can have a significant effect on computational efficiency. Body-fitted type of grids such as structured and tetrahedralbased unstructured grids move with the body, so that global grid modification is necessary during the movement. On the other hand, non-body-fitted type of grids such as Cartesian stay stationary while the body moves across it. As a result, it requires only local modification in the vicinity of body surface, which saves significant computational effort. This paper discusses the method proposed in the presently on-going research to treat moving body problem. The current approach seems to offer a significant improvement over an existing method.