Cooking up some CFD
Kevin J. Farrell, Principal Engineer, Computational Simulation & Validation
Colleagues sometimes inquire about a challenging simulation they know I’m wrestling with. My short reply is generally, “It’s still cooking.”
Sometimes explaining a complex topic with a simple analogy isn’t successful, but I think using a cooking analogy to describe CFD works well.
Think of skilled chefs making one of their best dishes under the bright lights of a televised culinary “battle.” The highest quality ingredients, high-end appliances and cookware, and impeccable attention to detail coupled with years of practice all play a part in creating a successful dish on those shows.
In some ways, CFD is much like cooking. But what’s the “recipe” for getting the most out of this powerful tool?
First, consider the cookware. For CFD, that means hardware, software, and validating data. On the hardware side, the preference is for more, better, and faster, but at a minimum, you need
- 2.4 GHz CPU with at least 4 cores per CPU to allow client and server to work separately and in parallel
- 2 GB of memory per core
- 9 GB of free disk space for execution
- about 100 GB hard disk space for storing the results of your study
- dedicated graphics hardware with 3D capability, z-buffer, and translucency
HTRI uses STAR-CCM+® and FLUENT® software packages, both very capable commercial solvers and useful in different ways. Usually the cost of a code is proportional to its universality in application, so a code focused on a particular problem class is typically less expensive than a code with wide-ranging modeling capabilities. At HTRI, we simulate single- and multiphase flows in various flow regimes in our research and under contract, so we need more advanced models. Lately we’ve also used the FEA capabilities in STAR-CCM+ for computational solid mechanics (CSM).
To have confidence in the CFD simulation for a particular case, we also require validating data. If none exist for that geometry, we may find experimental data in the literature for a more canonical geometry that exhibits many of the same thermal-hydraulic features. These features may be narrow channels and clearances, adverse pressure gradients, multiple shear layers, turnarounds, or stagnation regions. Even if the geometries are substantially different, successfully predicting these phenomena even in a different (hopefully simpler) geometry gives us some assurance that our code properly simulates the heat exchanger we are studying.
Next, we need ingredients, that is, accurate geometry and process information. Our simulation reflects properly on the heat exchanger’s operation only if we specify the as-built geometry and actual operating conditions, not necessarily what a drawing or a specification sheet depict.
Finally, after we have our cookware and ingredients in place, we need an experienced chef, one who can think creatively and still follow the "right" procedures. While commercial CFD software vendors like to emphasize usability and the more forgiving learning curves offered by modern solvers and workflows, they occasionally admit that achieving and maintaining a high level of proficiency with their products requires nearly full-time use.
While all engineers should understand the capabilities and limitations of CFD, some level of expertise in using the tool and understanding the modeled physics is required for timely and accurate CFD solutions. Careful preprocessing (e.g., meshing, sub-model selection, and setting initial and boundary conditions) consumes most of a CFD analyst’s time. Along with being detail oriented, successful CFD analysts generally must manifest what I call the three Ps—patience, persistence, and perseverance.
Sometimes, though, you may have to face the reality that you cannot make that special dish exactly “right.” In that case, contact firstname.lastname@example.org.