**Introduction **** **

This user guide outlines how to use Ennova to generate high quality, hybrid, and structured/unstructured meshes for CFD. Traditionally, producing high quality meshes for CFD has been difficult and complex and often is a major bottleneck in the CFD process. Ennova allows a mesh to be generated very quickly.

The basic meshing strategy of Ennova can be thought of as five steps, however these steps are closely coupled:

Figure 1. Ennova Basic Meshing Flow Chart

** **** **

** **** **

**Basic Ennova Meshing Process**** **

The imported CAD data is assembled into a virtual distributed assembly. This allows a path to full parallelization without the bottleneck of single processor geometry operations. Using many tests, the geometry is interrogated and reduced to N closed volumes with (usually) the largest volume used for the aerodynamic analysis. Once the geometry has been automatically cleaned and made watertight, the topology is again tested and divided into isotropic and anisotropic regions and surface meshing rules are applied. Mesh sizing is applied simultaneously based on these tests. For example, for prescribed y+ and N layers, the surface size of the element can be deduced based on local curvature, growth rates, and solver dependent aspect ratio constraints. Once a surface mesh is obtained, viscous boundary layer prisms are grown in an advancing frontal manner until they become isotropic. From here, an advancing front mesh method can be used to fill in the isotropic background mesh.

CFD meshing offers many challenges. For example: large data sets, features of many orders of magnitude difference, and very fine spacing required for viscous sublayer effects. Clearly unstructured methods can deal with the complexity of geometry, but often both isotropic and anisotropic elements are needed for an accurate mesh. It is very desirable to split the domain topologically into unstructured and structured regions. For example, it is advantageous to use a structured topology for wing leading and trailing edges where three distinct length scales exist: span-wise, chord-wise and normal to the surface. The three linear dimensions of a hexahedral element here will be orders of magnitude different.

In order to obtain a useful mesh, understanding the topology of the geometry is crucial. This can be done manually on small models, but needs to be done automatically for complex models or parametric studies. Automatic block structured topology generation is still very immature. We present an alternative. Here we make the observation that in the vast majority of cases the underlying CAD surfaces are parameterized in the directions of principle curvature. For example, a rectangular leading edge will have control points on its underlying NURBS surface aligned to the span-wise and normal to flow directions. It can be meshed with a structured mesh very efficiently in its parameter space. Similarly, a wing may start out as a sequence of 2D airfoil curves lofted in the span-wise direction, allowing the structured mesh to simply be aligned to the CAD model parameterization. However the above comments apply when the geometry can be sufficiently cleaned up. In the case where this is not possible, and the CAD parameterization cannot be used directly, an alternative method is needed.

In this case, the geometry that cannot be cleaned is shrink-wrapped. Here, using an Octree approach to meshing the geometry creates the virtual assembly. The final Octree volume mesh is discarded and only the closed surface volumes are kept. The advantage of this is this surface can then have its topology extracted geometrically (rather than underlying CAD parameterization) and the resulting surface can then be topology meshed as normal.

The complete process is shown in detail in Fig 2 below.

Finally, this meshing process is designed to be flexible and iterative. First, an automatic mesh of lower quality can be obtained very quickly allowing an initial CFD solution. Once this solution is available, any part of the automatic process can be manually over-ridden to improve the mesh. For example, unstructured regions can be forced to be structured and anisotropic. Extra refinement can be added or removed as necessary. Often in meshing complex geometries, meshing can take weeks if not months before the first mesh is available for solving. In our approach, an initial mesh is available almost immediately and can be incrementally improved using the solver results as a guide.

Figure 2. Ennova Detailed Meshing Flow Chart