CFD Project Accuracy vs. Effort

This is a thought experiment about the relation between CFD accuracy level and the corresponding effort needed for achieving it. My working name for this article was CFD at Confession, but in the end, I decided to keep it more formal. I’d like to shed a light onto CFD results and show that CFD is not a calculator, but it is rather its users and their effort who produce the results.

In the last decade, we’ve encountered a large rise of new CFD codes and new CFD business models. It is partly caused by the rise and availability of open-source codes [1],[2]. And partly it is caused by the trends going on in the engineering simulation marketplace, like for example democratization, software as a service, simulations in a web browser, and cloud computing [3]. 

CFD market is greatly overpromised

The whole CAE market [4], where CFD belongs, keeps growing well. CFD is not just a science anymore. CFD is a business. And where there is business, there is a competition. Where there is a competition, the marketers keep promising always a little bit more than their product can deliver. If this is happening over a long period of time, the overall image of the market is significantly distorted. I am pretty sure that this is going on in the CFD market at the moment. The CFD market is greatly overpromised. Both in terms of capability to cover complex physics and also in user-friendliness and accuracy. It became normal to present the best as a standard. When browsing the internet, it makes an impression that CFD is just a calculator (an expensive one) where (any)one just imposes the model data and CFD delivers the results. In fact, it is correct, but what is the results’ accuracy? Or at least, what is the level of accuracy?

Over the last ten years, I have been personally involved in more than 50 commercial consultancy projects, at all levels and roles, from the engineer to the project manager. Some of the projects were very successful, some projects were less successful. Nonetheless, all the projects had one thing in common. The results had always depended on the amount of effort that has been spent.

What is meant by accuracy?

The CFD results have always been compared to physical measurements (experiment) [5]. Maybe you know this old engineering joke: Nobody trusts cfd simulation, except the one who made it. Everybody trusts the measurement, except the one who made it. CFD is not perfect, neither is the measurement. But it is the CFD, who has to chase the measurement results and never the other way around. This is the game we play and these are the rules, no matter how much we like it or not at all. So at the end of the day, the CFD results’ accuracy is always judged with respect to the experimental measurement.

What is meant by effort?

Because every project is specific (the devil is hidden in the detail), for every project, a CFD user has to make several specific decisions and resolve them to a particular quality. Nowadays, most of the CFD tools are robust and automated. Users can decide how deep to dive into the CFD or not at all. That is a large benefit and a double-edged sword at the same time. I will give you a couple of examples of what decisions are to be made and what is the impact on the results’ accuracy level.

Example 1: Let’s consider a simple simulation of external aerodynamics of an object exposed in the fluid flow, static mesh, single phase, incompressible flow, no radiation, no heat transfer, virtual tunnel. And the result (the simulation goal) is the total force acting on the object due to the fluid flow. In this first example, the user picks some highly automated CFD tool, the typical decisions to take are: simulation domain, mesh size, time management, mean-flow boundary condition, fluid properties, and the number of simulation cores. Those are just six issues to be resolved. Such a simple simulation is easy to set up, but the resulting accuracy level will be around 20%. Single one simulation run will be enough.

Example 2: The same case as example 1., but this time the aim is the accuracy level (bandwidth) 10%. This time the user has to resolve, for example, the domain size, model roughness, mesh size, mesh refinement, inflation layers, time management, basic boundary conditions, fluid properties, turbulence approach, convergence criteria, number of cores, averaging method, evaluation formulas. Those are 10+ issues to be resolved. And that likely meets the 10% accuracy level. A few simulation test runs will be needed. 

Example 3: The same case as example 1. and 2., but this time the aim is the accuracy bandwidth 2%. This time the user has to resolve the domain size, model components, model roughness, mesh size, mesh refinement, inflation layers, y+, mesh dependence, time management, fully resolved boundary correct condition for every quantity at every boundary patch, initial conditions for all quantities, fluid properties, Newtonian fluid, cavitation, turbulence approach, turbulence model, turbulence model coefficients, wall treatment, walls roughness, relaxation factors, time step, write control, averaging window, spatial numerical scheme, temporal numerical scheme, limiters, linear solver settings (tolerances), convergence criteria, number of cores, averaging method, evaluation quantities, evaluation formulas, report units, probe locations, efficiency interfaces. Those are 30+ issues to be resolved. That should be enough to meet the 2% accuracy level. Several simulation runs will be needed.

CFD is not a calculator

We could continue with many other much more complex examples (you know, how passionate we are, for example, about rotating objects) but I am sure the point I just made is obvious. CFD is not a calculator. Or, at least, it is not a classical calculator. With a classical calculator, you put in the exact data and get back the exact data. With CFD, we never impose the exact, correct data, because they are not known exactly. Moreover, all the CFD methods are just a certain approximation of natural laws, which complexity is almost infinite. That means that in CFD, there is always a lot of uncertainty involved and always will be. That is bad news. 

Successful CFD project is a result of focus, skills, experience, patience, and dedication

The good news is that we still can acquire very accurate results and methods (either by ourselves or get it outsourced [6]) if we understand, that the results are created by people, not by the codes. CFD codes are merely tools. That is the reason why every successful CFD project is a result of focus, skills, experience, patience, and dedication. 

I would like to wish all of us a lot of successful projects and finish with a link to my favorite YouTube video of drawing Spiderman in a limited time (effort). Maybe it will remind you of some projects you have seen in the past.

https://youtu.be/x9wn633vl_c