Cadence has sponsored this post.

Prepping this model for CFD mesh generation was likely the most time-intensive portion of the simulation process. (Image courtesy of Cadence.)

Many CFD analysts dread seeing model preparations on a to-do list. Often what should be an exciting chance to explore a new design space is instead delayed by a scavenger hunt that could last hours—or potentially days.

Rather than using engineering expertise to develop, optimize and assess interesting designs, the analyst scours 3D solids for slivers, cusps, gaps and anything else that could break a mesh.

“Mesh generation for CFD is often cited as the most time-intensive portion of the CFD simulation process.,” said Nick Wyman, a manager on the Product Development team at Cadence (formerly Pointwise).

As simulation-led design and optimization strategies encourage more and more iterations of a product’s design, a larger portion of the development cycle will be devoted to this game of Where’s Waldo—or Where’s Wally, for those across the pond.

So, what is the problem here? Wyman noted that it can’t be placed on any one individual or segment of the process. In fact, to avoid laying blame, he has come into the habit of using the term “model unsuitability.” After all, he explained, the challenges of translating CAD models into CFD meshes are ubiquitous, so the issues must be systemic.

Model unsuitability is often produced by three issues that stem from conflicting model intent. These are:

Model interoperability and translations.
Surface intersections, trimmings and tolerances.
Geometric detail mismatches.

We discuss these challenges below, or download Cadence’s white paper “Preparation of Geometry Models for Mesh Generation and CFD.”

How Model Intent Creates CFD Geometry Interoperability and Translation Challenges

The intent of CAD software is to create geometry. This might seem like an obvious statement—even a frivolous one—but it has downhill implications for the software that uses that CAD geometry.

These complex models will have some ambiguity, which makes file translations complicated. (Image courtesy of Cadence.)

“In nearly all cases, regardless of the geometry modeler used, information must be translated from the creation software to the meshing software,” Wyman elaborated. “This is because most meshing packages use a geometry kernel optimized for geometry query operations—and during meshing, many millions of queries are performed—versus geometry creation operations.”

Transferring a model from a software optimized for geometry creation to one optimized for queries is not as easy as one may think. Regardless of the engineer that created the original model, it will have ambiguities that affect the translation. It’s like translating a book from UK English to American English; it’s easy to change all the “colours” to “colors,” but phrases like ‘Bob’s your uncle’ get ‘a tad wonky.’

“While there are standards and specifications for file exchange, variations in standards interpretation, common practices—and plain old bugs—result in model inconsistency if care isn’t taken,” Wyman explained. “It is not uncommon for a geometry model to be passed to, and from, more than one software package during its lifetime with each translation incurring a little more error along the way.”

“Proprietary formats can be worse as there is no published standard; instead, interoperability tools are used to reverse engineer the data format,” he adds. In these scenarios, third-party tools and interoperability libraries do their best to translate and approximate the proprietary file into a standard file. Not only is that problematic in and of itself, but it also adds additional translations to the lifecycle as the standard file will inevitably be converted into something else.

“It is at this point that CAD-embedded CFD has an advantage, because meshing uses the originating geometry kernel, thus avoiding translation errors,” said Wyman.

How CAD Model Intersections, Trimming and Tolerances Complicate CFD Meshing

It turns out that most CAD models are a lot more complex than they seem on the user interface (UI).

Wyman explained that these models are kind of like a quilt; a bunch of mathematical shapes and surfaces are knitted together by Boolean operations and rules that define a boundary and an overall shared surface. Once all the excess surfaces are cut away, you end up with a part that is well defined for analysis and manufacturing.

Comparison between the CAD model in the graphical interface, and the actual surfaces that define its shape. (Image courtesy of Benjamin Urick, nVariate, Inc. and Benjamin Marussig, TU Graz)

“The most common form of geometric modeling uses boundary representation topology, in which a collection of trimmed surfaces are logically joined at shared edges to bound a volume, or solid part,” Wyman explained. “When you show all the mathematics behind it, you see that complex part.”

The challenge is that CAD models were never intended to define these intersections analytically. If they did, the computational resources needed to fully define the intersections would not be practical.

“Rarely are intersection curves between surfaces defined analytically—particularly for the organic shapes commonly seen in modern designs,” Wyman said. “Instead, surface intersection curves are determined by approximately defining the junction of two surfaces within a prescribed tolerance. This means the intersection curve does not exactly lie on either of the two surfaces.”

Since the intersection curves used to trim the boundary are approximations, they pose a problem for engineers wishing to simulate watertight solids. Specifically, the curves can return inconsistent answers, within a tolerance, which could invalidate the mesh.

“Geometric inconsistency between trimmed curves and surfaces results in the loss of robustness during the meshing process—particularly when the element size is on the same, or lower, scale as the intersection curve’s tolerance,” Wyman explained. “When this occurs, the analyst must manually intervene to control mesh behavior and maintain mesh validity.”

The gap created by the approximated intersection curve of two surfaces. (Image courtesy of Benjamin Urick, nVariate, Inc. and Benjamin Marussig, TU Graz)

How Model Details Complicate CFD Meshing

The amount of detail within a model is also dependent on the intended use of the geometry. “Geometry models have many end users,” Wyman elaborated. “So, it is tempting to include as much detail as possible out of ‘completeness.’”

He added, “Some geometric details needed for the manufacturing of a part are unnecessary when performing CFD analysis. For example, geometry in excess of the wetted surfaces must be omitted during CFD meshing—a process which can be arduous and time-consuming for detailed thin parts.”

The wetted surfaces, like the hood of a car, might be smooth on the outside. But under that surface, the model contains struts and other structural geometry. These shapes would not interact with the air, but they would produce a complex mesh. “To get rid of these structures,” said Wyman, “You sometimes resort to clicking on individual parts of the geometry thousands of times.”

In addition to the surfaces that won’t interact with the fluid, there will also be—intended and unintended—geometries that have insignificant effects on the fluid flow; this could be a nut, a bolt, or slivers, cusps and knife-edges. Though these geometries will have little effect on the results, they also complicate the mesh and must be eliminated.

Excessive information is not the only ‘level of detail’ issue that can pop up. Wyman noted that, “The converse is true as well. Omitting necessary-for-simulation features, such as farfield boundary faces, or common faces between parts in an assembly, causes an interruption in the CFD meshing workflow as these features require manual definition by the analyst.”

In the case of missing geometry, we aren’t talking about wheels, windshields or mirrors. The CFD geometry that is missing is in addition to the design itself. As an example, engineers will need to define features that can affect the fluid flow around the design—such as the ground. Since these parts are missing, and require a manual touch, the process of adding them is hard to automate.

How to Simplify Geometry for CFD Meshing

Engineers have a few tricks up their sleeve to simplify geometry for CFD meshing:

Topology abstractions treat the mesh as a single component, ignoring boundaries.
Repairs or healing create approximations of local surfaces to improve model validity.
Shrink-wrapping reduces the size of a watertight mesh, until it lays on top of the model.

Even machine learning (ML) tools can be used to automate much of the geometry fixing. However, this method limits the control engineers have on the model.

Wyman noted that Cadence offers many of the tools that engineers need to get the job done, “We offer sophisticated tools to automatically resolve geometric validity issues due to model tolerances. Our productivity tools are designed for working with excessively detailed models, and our fault tolerant meshing tools allow mesh element sizes to be smaller than model tolerances.” He also noted that Cadence offers interoperability tools between CAD and analysis packages.

For a more in-depth discussion of how to prepare geometry models for CFD mesh generation, download the Cadence white paper “Preparation of Geometry Models for Mesh Generation and CFD.”