All About Meshing

Meshes are primarily composed of two distinct entities: the geometry (the position x, y, and z of the nodes) and also the topology (how the nodes are connected).

 

What is a Mesh?

A mesh may be a network that's formed of cells and points. It can have almost any shape in any size and is employed to unravel Partial Differential Equations. Each cell of the mesh represents a personal solution of the equation (all unknowns of equations) which, when combined for the entire network, ends up in an answer for the whole mesh.

Solving the complete region without dividing it into smaller pieces is impossible due to the complexity involved within the domain. Holes, corners and angles can make it extremely difficult for solvers to get an answer. Small cells, on the opposite hand, are comparably easy to unravel and so is the popular strategy.

 

Introduction 

The first step for numerically solving a group of partial differential equations (PDEs) is the discretization of the equations and therefore the discretization of the matter domain. The equations discretization process is expounded to methods like the Finite Differences Method, Finite Volume Method (FVM) and Finite Element Method, whose purpose is to require equations within the continuous form and generate a system of algebraic difference equations. The domain discretization process generates a group of discrete points and cells that cover the continual problem domain.

 

 

The mesh network can have many styles of geometry and topology. Often, meshes also are called grids, which is usually associated with the intrinsic organization of the mesh and/or when those meshes are associated with Finite Differences problems.

Each cell or point of the mesh will hold a neighborhood solution of the equations, looking on whether the equations were discretized over the cells or points. the selection of whether to discretize the equations over the set of cells or points could be a project decision, and plenty of aspects can affect the selection.

In general, point discretization is employed when the equations are approximated using the Finite Differences Method, where the PDEs are usually approximated by Taylor series expansions at each point’s neighbours. Sometimes the purpose discretization is used along with the Finite Volume Method, however, cells are visiting be implicitly used around points.

When the equations to be discretized are considered within the weak form, integral form, or conservative form, it's common to unravel the integrals over discrete cells. for instance, when considering transport phenomena, the Finite Volume Method is often formulated because of the discrete cells representing small volumes. Then, the fluxes are often balanced through those cells, while supposing that the answer is constant inside them.

 

Types of Mesh

It’s common to classify mesh types as being structured or unstructured. Structured meshes are historically associated with the Finite Differences Method. The Finite Volume and Finite Element methods’ nature allows for more general meshes.

 

Structured Meshes

Structured meshes, also commonly called grids, are meshes whose structure and formation leave easy identification of neighboring cells and points. This property springs from the actual fact that structured meshes are applied over-analytical coordinate systems (rectangular, elliptical, spherical, etc.), forming an everyday grid.

From a programming point of view, the cells or points that form a structured mesh are often enumerated in such the simplest way that neighboring queries will be made analytically upon the cells or points coordinates.

From a programming standpoint, there's little to no difference between a curvilinear grid and an oblong grid in terms of adjacency queries. It can easily form into an array or vector which makes computations easier.

 

 

Structured grids may also be defined in terms of boundary fitting. for instance, a cartesian grid fits the boundary of a rectangle, and a cylindrical grid fits the boundary of a cylinder. It’s possible to combine the boundary fitting and have many alternative curves or surfaces that outline the boundary. These include any reasonably parameterizable curves and surfaces, like splines and NURBS (Non-Uniform Rational Basis Spline). The meshing algorithm will decide how points are going to be distributed through those surfaces, and the way opposing surfaces are connected to every other, given what number points should be created for every curve or surface.

 

 

Unstructured Meshes

Unstructured meshes are more general and might arbitrarily approximate any geometry shape. In contrast with structured meshes, where the coordinates and connectivity map into the weather of a matrix, unstructured meshes require special data structures, like an adjacency matrix or list and also the node coordinates list. The unstructured mesh node/cell numbering will be arbitrary and sparse since it doesn't require any analytical style of adjacency query.

 

 

Complex geometries that will be impractical to come up with a structured mesh within, are often discretized using unstructured meshing techniques. The flexible nature of unstructured meshes allows various cell types to be used and coexist within the same mesh, making it possible to own a fair better geometry fitting and overall element quality.

 

 

The surface element on 2D objects is formed by a Triangular element (Tri) and Quadrilateral element (Quad). While 3D (or volume) will be represented by a Tetrahedral element (Tetra) which has four tri faces, a wedge (or Prism) element has two tri faces and three quad faces, Pyramid element has four tri faces and one quad face. A hexahedral element has six quad faces.

Mesh cells don't seem to be required to be conformal. Nonconforming meshes are those where hanging nodes occur. These nodes appear more commonly during mesh adaptation processes.

 

 

Mesh Adaption

Since on unstructured meshes the points and adjacencies don't follow any quite global structure, it's also possible to feature or remove mesh cells and points. the method of dynamically adding, removing or displacing mesh cells and points is termed mesh adaptation.

Depending on the problem's nature, mesh adaptation techniques are required to get accurate solution fields while maintaining the computational cost of the matter low, by keeping the general number of mesh cells and nodes in check.

In general, the degree of refinement required is related to the error estimated supported the equations to be solved. Therefore, regions with higher error will find themselves accumulating more mesh cells.

Mesh refinement is split into several general types. most typically, these types are remarked as h-type, r-type, p-type, derefinement, and combinations of these.

 

Mesh Refinement

Mesh refinement, which is additionally called h-type refinement, is predicated on the addition of cells or points, reducing the local characteristic length of edges. this system can cheaply increase local mesh resolution, but also will increase the number of simultaneous differential equations to be solved since it increases the number of degrees of freedom of the system.

On unstructured meshes, the addition of cells or points is easy, because it involves the reconnection of the modified cells. Refinement on structured grids isn’t straightforward since the addition of cells could break the mesh's regularity. It’s common to permit nonconforming meshes when using adaptive structured grids.

 

 

Similarly, mesh refinement techniques, in general, leave mesh derefinement, which might be wont to decrease the number of cells in regions where the estimated error is extremely low. this enables computational power to be more efficiently used, reducing costs and simulation time.

 

Mesh Movement

Mesh movement, or r-type refinement, is completed through the movement or displacement of mesh cells and points. during this case, the quantity of cells and points remains constant, while also sometimes keeping the connectivity identical.

Below is depicted an example of an r-type mesh refinement process that might be associated with an undulation propagation problem, for example, where the mesh resolution is kept higher in regions of high variation on the answer.

 

 

Other Mesh Adaption Techniques

Other common mesh adaptation techniques might include p-type refinement and adaptive remeshing. On p-type refinement, which is related to the Finite Element Method, the complexity of the form function is increased, while keeping the identical mesh. The adaptive remeshing technique is employed to get a brand new mesh, supporting the estimated error. this permits for the most effective overall mesh quality, and fewer points to be used. On the opposite hand, the overhead of mesh creation may be large. Adaptive remeshing is often used locally, generating new meshes only on regions where the estimated error is just too high or low.

Combinations of the aforementioned techniques are often used. as an example, the mix of r-type and h-type refinement is often cited as the rh-type refinement, where nodes can both be displaced or created on the mesh.

Frequently-Asked Questions in Meshing

 

1) How does the FVM solver work on a mesh?

A general FV-based code doesn't make difference between structured and unstructured grids. It works by doing a flux reconstruction on each face of the finite volume, regardless of the shape.

The solver forms control volumes around the nodes, which implies for tetra meshes the control volumes are polygonal. If we use a structured mesh the control volumes also are hexes, which makes the control volumes have fewer faces (so fewer fluxes to stay track of, less memory) and also the control volume faces can potentially be closer to perpendicular which helps reduce numerical diffusion.

2) How meshes are ordered?

Structured grids are ordered-type by indices such (i, j, k) so any value for these integers corresponds to a spatial location and its topology is well related to the following indices. Unstructured grids don't have any such ordering, therefore a node is related to one integer index and therefore the topology is thought of by associating the connections of the node with the neighboring nodes.

3) Can a tet mesh be structured?

A common misperception is that a hex mesh is structured and a tet mesh is unstructured. it's accurate to mention that a Tet mesh is unstructured, however, a hex mesh may also be unstructured. The difference between a structured hex mesh and an unstructured hex mesh is just in how the information is stored.

A structured mesh may be composed of triangular elements. Mesh is claimed to be structured if one can find a primitive motif that's regular in space. this suggests that as an example node connectivity incorporates a fixed pattern. But the geometry maybe not be constant, for example, your mesh may be deformed from place to position (for instance to match a particular boundary). the sole we all know needless to say when the mesh is regular is that the node (or elements) connectivity is usually identical. 

4) What are the important differences between structured and unstructured?

A mesh is claimed to be structured if its topology is regular, i.e. have a well-known pattern (such as cubes in a sugar box). On the opposite hand, unstructured meshes have an irregular topology, meaning that it has to store the topology for every single node (or edges, facets, etc.). This specific feature makes unstructured mesh more flexible and more memory-demanding addition.

It is hard to mention normally that structured or unstructured mesh is simpler to form. Structured meshes are difficult for complex objects, and generating unstructured meshes could be a very cumbersome task. the sole important thing is that, on complex objects, it's generally true that unstructured meshes create fewer degenerate elements.