Best Practices for Meshing AEC Applications in Simulation CFD

AEC applications can be challenging to mesh due to their sheer size. By following fundamental best practices for mesh convergence and employing a strategic mesh refinement scheme, even the most complex situations will be mesh efficiently and accurately.

Before proceeding, it is recommended to review the online help material for meshing AEC applications.

The primary objective for every simulation is to obtain quality performance insight as quickly and efficiently as possible.  Since meshing directly impacts both accuracy and solution speed, it is a very important aspect of the simulation process.

Typical AEC Characteristics

AEC application domains can range from the internal flow of small data centers or lab spaces (e.g. 30’L x 20’W x 10’H) to the external flow around an entire hospital site (e.g. 1000’L x 1000’W x 1000’H).

The common characteristics of most AEC applications include:

  • Relatively large fluid volumes
  • Smaller localized areas with finer details

Despite potential differences in dimensional scale, the same fundamental meshing principles will always apply: using enough mesh and focusing finer mesh in regions that need it the most.  AEC applications with larger domains with more local detail will ultimately require more elements compared to those with smaller domains or less geometric detail. 

Sufficient Element Count

By strictly following the principles of the mesh convergence process, a quality mesh containing a sufficient number of elements for a mesh independent solution, will always be obtained. Although tempting because the solver runs are much faster, care must be taken not to base final design decisions on a mesh of insufficient quality.  Final decisions should only be taken from a mesh that has been validated by the mesh convergence process which requires iterative mesh refinements.

1. Default autosize mesh of data center contains 120,000 elements.  Solution will run faster but results will be misleading because of poor overall mesh quality.
2. With only 6 elements from floor to ceiling, it will be challenging for the solution to resolve recirculation or vertical thermal stratification.  
3. Only 2 elements span across the air conditioner inlet which would result in a poor flow profile in this area.
4. Refined mesh with all volumes adjusted to 0.5.  This mesh now contains 1.4 million elements (more than a 10x increase from the default mesh).  Solution time will be longer than the default mesh, but much more accurate.
5. Refined mesh now has 12 elements spanning from floor to ceiling.  This increase in mesh resolution is needed to capture subtle flow details such as small areas of recirculation.
6. Refined mesh now has 6-8 elements spanning across the conditioner inlet an)d can more accurately represent the flow profile.

Strategic Placement of Elements

For AEC applications, a very important consideration is the mesh refinement strategy.  Ideally, elements need to be focused only where they are needed the most.  Due to the sheer scale of many AEC simulations, over-refining the mesh (i.e. using too many elements) has the implication of generating millions of elements that are not needed.  Although accuracy is good, the solver run will take longer to complete, which does not align with the primary objective of obtaining insight as quickly as possible.

As experienced is gained with completing simulations and understanding what a quality mesh looks like, the strategy for mesh refinement will become more clear;  expert CFD consultants will automatically start thinking about mesh refinement while still preparing the CAD model.

When reviewing the model geometry and visualizing results, the following areas should be thoroughly evaluated for local mesh refinement:

Fastest flow velocities

  • For a typical room, the fastest flow would be located at the inlets and returns (e.g. place your hand over an inlet while it is blowing air and then move your hand away from it).
  • Internal momentum sources such as computer fans, fume hoods or ceiling fans would all have higher local flow velocities than the surrounding air.

Narrow flow channels

  • These are areas where the flow has to go between a relatively thin gap.  For example, when modeling chemistry labs, a small gap under the door may be modeled to make sure air is always being pulled in from the hallway to eliminate odors or contaminants from escaping the room.
  • Depending on surrounding conditions, narrow flow gaps may also correspond to areas of highest flow velocity.

Highest component temperatures

  • Objects that are hotter than the surrounding fluid may create a local natural convection engine (i.e., as hot air rises off the component, it entrains colder air from the immediate surroundings).  An example would be a large motor or engine on a factory floor.

Complex geometry

  • The autosize feature will automatically focus more elements in areas that have small edges or lots of curved surfaces, such as those found in architectural features.  However, it still may need some additional help from local refinement.

To demonstrate the process of strategic refinement, consider the following AEC application of an air curtain, which helps to keep ambient air out of a conditioned space when the door is open.  Assume that the ambient wind speed is 10 mph and the air curtain inlet velocity is 3000 fpm.

1. Ambient air with prevailing wind
2. Doorway with air curtain.
3. Conditioned retail space

An initial default mesh is generated, reviewed and then refined by adjusting the mesh sizes on the volumes.

Default autosize mesh (left) is the starting point.  Adjusting the volume size to 0.25 increases the element count (right) for better results fidelity.

The refined mesh appears to be a good start, so a simulation run is initiated and the results are reviewed.  As depicted below, the results appear questionable in the vicinity of the air curtain

1 Air curtain inlet only has 1 element  across, which is not enough to accurately capture the flow profile.
2 The velocity of the air drops off quite suddenly once it leaves the nozzle.

Even if the expected results from prior physical testing or hand calculations are not available, the air curtain region should still be considered for refinement for two reasons: 1) It is the highest velocity fluid in the model (the air curtain is 3000 fpm, 10 mph converts to only 880 fpm); and 2) The air in the curtain inlet has to go through the narrowest gap in the model.

A number of suitable mesh refinement strategies could be used in this situation.  In this particular case, a total of 3 mesh refinement regions were added as depicted below.  The regions are cascaded to precisely control the transition from small elements to large elements.  This method (sometimes referred to as the Russian doll approach) helps to conserve elements by avoiding using the same small mesh size used for the inlet in areas which are some distance from the inlet.

Refinement regions can be overlayed on top of each other (left) to control element transition from smaller to larger.  Note the finer mesh in the region of the air curtain and that there are now 6 to 8 elements spanning across the air curtain inlet (right).

A comparison of the results below reveals the impact of proper mesh refinement on the results.

1 Default mesh  Air curtain velocity fades rapidly at the inlet and the plume does not extend to the floor.
2 Strategically refined mesh.  The air curtain plume extends nearly all the way to the floor.

The video available here further illustrates the refinement process on the air curtain.

An extreme example of what NOT to do is to take the small mesh size required around the inlet and apply it to the entire domain.  The resulting mesh, shown below, is so dense that it appears all black when zoomed out.  While this mesh is certainly accurate, it would be considered as over-meshed and a waste of elements since the resulting run times would be much longer than needed.  

While accurate, the mesh depicted here would be considered as over-meshed and impractical since similar results could be obtained with much fewer elements.