Combining structured and unstructured meshes: the Holy Grail for CFD engineers

        ホワイトペーパー     23.10.2020

In CFD history, structured meshes came first, and they are still in use today. Structured meshes offer several main advantages, such as precision, speed of generation, and uniform distribution of cells. These types of meshes, which AutoGrid5™ excels at producing, are perfectly suited for turbomachinery applications with any kind of blade geometries.

As the complexity of geometries began to increase (now often with more than 10k surfaces), the need for another type of mesh--one with unstructured properties--arose. The problem with certain geometries is that they lack accuracy in terms of definition (“dirty” or “unclean”), and they do not present any particular trends to which standard structured mesh topologies can be applied. CFD users have to spend a lot of time defining these new topologies and cleaning the geometry before even starting to mesh. In other words, as soon as a geometry falls outside of the classic range of structured mesh applications, the debate starts: when should we adhere to creating structured meshes versus simply switching to unstructured meshes?

This question could be answered easily if the only factors to consider were the physics to be captured and the type of accuracy needed. However, there is a third factor: flow solvers must be able to read the type of meshes behind them, and as most flow solvers only accept either structured or unstructured meshes, this is a no-win situation.

This seemingly no-win situation is where NUMECA offers an innovative solution, however. NUMECA has a reputation for providing the right technology for the corresponding application, and as such, our engineers have developed a solution within our OMNIS™ environment where users can switch between meshing approaches in a single click, enabling them to not only access within the same project--but also within the same view--all the pieces of the geometry, whatever the preferred meshing technique.

For blade-like geometries, users can apply a structured approach in OMNIS™, and for non-rotating parts, like combustion chambers, volutes, etc, users can apply an unstructured approach. As an example, a complete turbomachine is shown below (the complete simulation is detailed here). Each piece can be done separately, and the mesh generation itself in parallel. As such, 19 minutes are necessary if we can afford to run them at the same time.


Compressor mesh generation with structured and unstructured meshing combined

Hence, we are seeing more and more users applying the best methodology depending on the geometry component. We can definitely identify a few others, like an impeller and its volute, a ship, and its propeller, etc.



Pushing the idea a bit further: since NUMECA offers both meshing methodologies in OMNIS™, what about using them at different locations within a single geometry? We previously stated that blades are best created with a structured meshing tool like AutoGrid5™ (which can also do a lot of particular configurations such as cooling holes, asymmetric end walls, casing treatment, etc.). But some parts of the geometry could still be tricky for any structured mesh generator. Hence, a technique consists of removing a few blocks from a structured mesh, (away from the difficult part), and then reinserting an unstructured mesh block instead, which includes the complex tip geometry (as shown in the example below). As a result, the majority of the mesh is done with a structured grid, but the most complex part is done with an unstructured approach.


And that’s not all. NUMECA also proposes the idea of using an initial structured grid for an unstructured mesh generation. Indeed, for a volume-to-surface approach, an unstructured mesh is classically based on an initial Cartesian or cylindrical grid that is further refined. But the refinements can actually be started from any type of cell alignment: straight or curved following any kind of shape, and it can support different cell distribution. As such, IGG™ or AutoGrid™ are two very good candidates to provide such an initial grid! 

Some applications are very well suited for that concept: as proof, we used it for a hydrofoil simulation (all details here). The mesh consists of an overlapping grid over a background domain for a FINE™/Marine simulation. This overlapping grid follows the geometry curving of the foil, basing its width on the chord and its length on the span. The result is truly impressive, and out-performs any other method:

  • the global mesh quality is excellent (and so the mesh generation speed is improved)

  • The height of the viscous layers is as expected by the theory

  • the cell quality at the overset boundaries is optimal for the solver interpolation


All these concepts are very important and made possible because the data structure of the OMNIS™/Open solver is designed to understand both mesh technologies. In other words, the benefit is immediate for the user: no more conversion from structured to unstructured, the transition from meshes to the solver is direct, and the flow solver uses the best of the combination of structured and unstructured grids!