CFD is Non-Linear and More Difficult to Tame than Structural FEA
Unlike static structural analysis and other physical modelling regimes, the Navier-Stokes formulae are partial differential equations that for most interesting geometries and realistic flow conditions, do not have an analytical solution. For realistic problems, you can’t just plug algebraic terms into Matlab. Numerical methods for solving these equations have been under development for more than a half-century. The fundamental idea for the most popular methods is to discretize the flow domain around or within the object under study: the physical space is divided cells as small as millimeters on a 1000 meter long supertanker. Time is also broken down into very small timesteps, sometimes on the order of microseconds or even nanoseconds.
Solution methods with names like Finite Difference, Finite Volume, Finite Element and Direct Numerical Simulation are then applied to the millions or billions of cells, timestep by timestep. (Techniques such as Lattice-Boltzmann, Particle-in-Cell and Smooth Particle Hydrodynamics are also used, but tend to have more specialized applications). Each of these methods has advantages and disadvantages in terms of memory needs, computing power requirements, stability (meaning: do we get an answer or a program crash) and most importantly, accuracy.
Solving differential equations has another important requirement: boundary conditions. For example, what is the speed of the airplane or the temperature and pressure at the inlet of a jet engine combustor? These conditions are natural to us in the real world, but expressing them accurately as inputs to the solution program (known as the solver code) or even measuring them accurately, can be very challenging. Even if we could manage boundary conditions, discretization and solution method, there is a trade-off between what can be directly solved and what needs to be modeled. Turbulence, the tendency for many flows to exhibit an almost chaotic behavior, exists at many scales and directly impacts lift and drag, supersonic combustion and other phenomena. The quest to understand and model turbulence has been an ongoing pursuit for more than a century.
These facts explain why compute requirements for CFD are more demanding than those for structural or thermal analysis. In addition, the size of files that need to be post-processed for a single, unsteady analysis can be terabytes in size.
Even if you are successfully using CFD, ask yourself these questions:
- Am I waiting a long time to even read my CFD results?
- I’m using the post-processor that comes with my solver code, but what am I missing?
- I spend all day copying CFD results from my server, is there a better way?
- My competitors are showing their CFD results with great looking images and animations, what do they know that I don’t?
- I have no time to read and examine all the solutions that I have, how can I automate that?
- My solver runs in parallel. Can I post-process in parallel too?
- I have a great CFD product such as ANSYS Fluent, SIEMENS STAR-CCM+, Metacomp CFD++, etc., but am I getting to all the value in those results with the built-in post-processor?
- How can I get great post-processing for my open-source or government CFD code, like OpenFOAM or OverFlow?
- What are the error bars on my quantities of interest? Can I really trust the results to be predictive?
Products and services from Intelligent Light have been specifically developed to help CFD users get more reliable results in less time from their CFD investments. Take a few minutes of your valuable time to check our solution information for the answers to your questions and consider a pilot project with Intelligent Light – we work along side your team to bring the best of our solutions to you so that you can get the most from your CFD.