What Actually Happens on a CFD Project
What happens behind the current when the CFD engineer goes to work? What goes into making a CFD simulation? When we understand the workflow of a CFD project, we better understand the expectations for progress of our own CFD projects. As a project manager, knowledge of the CFD workflow helps you plan project risks and understand potential cost overruns. Forewarned is forearmed when purchasing your next CFD project.
The CFD workflow divides into three major phases (Figure 1‑1):
- Model setup
- Client scope
During the geometry phase, the CFD engineer prepares the CAD geometry for the CFD solver, which requires a 3D closed solid volume for its geometry. CFD demands much higher standards in geometry quality than the typical human.
First, the engineer cleans up the CAD geometry. They redefine surfaces with simpler geometry, remove unnecessary elements such as fillets and small details. They search for gaps and holes and generally cleanup a myriad of features that cause problems for the CFD solver. The quality of the CFD mesh builds from the quality of the original CAD geometry. Limitations in bad geometry inhibit quality of downstream CFD predictions. That is why the engineer begins with the best geometry possible.
3.0 Model Setup
After importing the geometry into CFD software, the engineer constructs the CFD model. This incorporates physics into geometry.
3.1 Setup Physics
CFD simulations do not automatically load all the physics in the world; the computational load would overwhelm most computers. Instead, the engineer selects the correct physics models, inputs model parameters, and generally inputs the physics necessary for your specific CFD simulation.
3.2 Setup Mesh
The mesh divides the geometry into millions of tiny little cells; the combination of these cells with programmed physics allows the software to solve the CFD problem. But not all meshes are created equal. Simulation quality hinges on finding the correct size for these cells in the correct location. Mesh sizing is the primary control mechanism for the CFD engineer; they focus the majority of their time on this step. It requires an iterative process: try some mesh settings; test a simulation; examine the results; refine the mesh settings. Expect to spend a significant portion of the project with the engineer refining the mesh. Mesh quality equates to result quality.
3.3 Debug Simulation
Computers faithfully execute exactly what they are told, including all the mistakes made by the human CFD engineer. No simulation works perfectly the first time. The engineer checks for a host of potential problems:
- Incorrect physics settings
- Trouble locations that require additional mesh refinement
- Simulation instability
- Incorrect outputs
- Other unknown problems.
3.4 Validate Model
All simulations are wrong until proven otherwise. This is the mantra of a reliable CFD engineer. After debugging, they proceed to validate the simulation and prove its accuracy. They check the outputs for sensibility, make sure the flow patterns are reasonable. And perform a mesh independence study.
The mesh independence study systematically tests the simulation at several different mesh sizes and compares simulation outputs at each mesh size. The engineer searches for a state of mesh independence, when the outputs no longer change with mesh size. Once reached, the engineer can predict the accuracy of the simulation. (Not all simulations are equally accurate.)
4.0 Client Scope
All previous work established a reliable and accurate CFD model. This proved the model could be trusted for real world engineering. Because an unproven model had no value, and no purpose in serious engineering. Next, the CFD engineer applied that model to the Client requested scope of work.
4.1 Production Runs
The production runs are all the various conditions to simulate. Generally, each condition matches to one simulation. These can be anything, like:
- Wave height
- Ship length
When planning combinations of multiple conditions, the matrix of production runs multiplies rapidly. Imagine a test matrix of just 3 speeds and 5 wave headings. Taking all possible combinations gives 3 x 5 = 15 simulations for a simple test matrix. Imagine how quickly that expands with a larger test matrix. Data management consumes much of the engineer’s time at this stage. They simulate each condition, store the results, document any notes from the run, and troubleshoot any problems that arise. Production runs can take a lot of time, since each simulation requires 8 – 30 hours to complete on average.
4.2 Post Process
The raw output from a CFD simulation is a database of numbers, not easily interpreted by humans. In post processing, the engineer converts that database into various forms of presentation, highlighting important points. These are typically graphical visualizations. (Figure 4‑1) But they can also be tables, values at a specified location, or almost any other type of data requested.
Discuss the outputs with your engineer before you begin production runs. The engineer programs in most of these outputs into a single template file before making dozens of copies for each simulation condition. Generally, engineers are happy to add extra post processing to the template. The extra labor comes from adding in post processing after the fact, because they need to manually program the post processing into each single simulation file. Beyond the labor, that greatly increases the risk of transcription errors. You will find the entire process far easier before production runs begin.
Finally, the engineer documents everything in a report. Beyond mere results, the report should document sufficient information that a third party could reproduce the CFD simulation.
- General simulation characteristics
- Major results
- Overview of mesh
- Basic information about discretization scheme
This acts as another form of quality control. Most companies strike a balance between reporting excessive details and protecting their own expertise. In general, they will happily describe the simulation setup, but withhold details about the exact sizing of the mesh. This is because most of the simulation quality derives from mesh sizing.
Many engineering firms do not include all these extensive details in their engineering reports purely for convenience and brevity. Much of the information about a CFD simulation does not document easily in a paper report, requiring hundreds of pages. Instead, companies usually elect to provide the simulation file on request.
CFD is not magic. Nor is it an exact science. It requires an elaborate and iterative process, plus some intuition from the CFD engineer. As a project manager, understanding that process helps you plan the project and track budget expenses. Armed with your new understanding of the CFD workflow, you can anticipate the evolution of the next CFD project and anticipate all the unexpected project evolutions. You can now ask the right questions.
W. Hage, “Karmansche Wirbelstraße aus numerischer Simulation der Zylinderumströmung,” Wikimedia Commons, 27 Oct 2014. [Online]. Available: https://commons.wikimedia.org/wiki/File:Karmansche_Wirbelstra%C3%9Fe_CFD.jpg. [Accessed 31 Dec 2018].