CFD is a dynamic field which relates engineering and mathematics. But how much one should give effort on the validation?
Dear Shamudra,
To start with, your question is very much generic. It is not very clear what kind of CFD solvers you are relating to: A commercial CFD solver like Fluent, an open source library like OpenFOAM or it is about a in-house CFD solver developed by you. However, in a generic sense, Grid/Time convergence and flow solver validation is the foundation of the accuracy of the simulation results. I will suggest you to first optimise your mesh and time step through convergence studies before going for validating others results. This might save some time for you.
If you are validating your own code, it is always advisable to match with experimental results (if available) and not with another CFD results in the first priority. The other option, it if you are using any well established open source solver or commercial solver, search for the validation studies already carried out in the literature and you can cite them for your reference.
Please refer to one of my recent publication in Physics of Fluids to get an idea about how the convergence and validation studies should be performed.
1. Chandan Bose, Sunetra Sarkar (2018) : Investigating chaotic wake dynamics past a flapping airfoil and the role of vortex interactions behind the chaotic transition, Physics of Fluids, Volume 30, Issue 4, 047101. DOI: 10.1063/1.5019442.
Best regards,
Chandan
I generally use commercial software ANSYS Fluent for CFD related problems
It depends on the type of the problem you solve.
In general, for an industry level problem, mesh generation (pre-processing) is the most time consuming part. However, for a research based problem, solver convergence and validation may be time-consuming.
Debugging your code is the most time consuming part if you are developing your CFD code from scratch. For me time for debigging= 4x time writing the code...:p. Some other part also take time for example grid convergence since you need to run a lot of simulations for the same case but for different resolution in order to get a converged value. For validation, try to validate your result with some good journal papers (ex-JFM, phy of fluid, comp. phy., AIAA, phys. review...etc.)
Grid (mesh) generation takes the longest and generating a high quality grid takes time. The second is mesh sensitivity study.
Validation is very important when it comes to using a particular CFD solver of your choice (trust). Validation section has to be included in a paper if you are looking forward to publish your research especially in a Q1 journal.
First try solving a model problem, which admits exact solution. Most of the time, you will be surprised with your experience. Pre- and post-processing are mere garnishing, although one has to go through. Knowing what can constitute error for your problem, is the main issue. These few rules apply for anything you solve.
In CFD:
Grid and Mesh Generation can take up a lot of time to get quality meshes.
Sometimes, you might have to remesh the whole geometry after you run your CFD code for some time, if you have complex phenomenon like moving meshes.
Yes, verification and validation is extremely important for CFD codes, perhaps the only. If you are using a commercial code, your CFD schemes are already verified,
but you must report validation studies for publication.
For example, how does your solution vary with the grid size, and if you are using models, then how does your solution vary with different models would be of prime interest. You can look into the extensive literature of validation studies, and you will come across commonly used terms like ``grid convergence", ``grid sensitivity", ``grid-convergence-index or GCI and so on. You can compare your results with experimental data or direct numerical simulations (if you are solving for turbulent flows say).
HelloTanmoy Chatterjee !
I think your statement that : "If you are using a commercial code, your CFD schemes are already verified" is fraught with danger (perhaps ignorance about how to verify numerical schemes) for potential users and completely misleading. It can never be far from the truth. I will give you an example. Try solving the simple 2D convection equation with known exact solution by ANY commercial code. The exact solution is given by the initial condition prescribed. To make matters interesting, do not use any trivial initial condition, but, instead use a wave-pscket. I can assure you that all the commercial codes would remove the signal altogether, after you have computed for moderate times. And then talking about grid convergence, grid sensitivity etc. are realy waste of time.
By now you realize that no-one should really be bothered your turbulent flow simulation, because...
"If you cannot solve a simple deterministic linear solution, then why should anyone trust your turbulent flow simulation"?
A point to ponder!
Dear Tapan K. Sengupta , yes I absolutely agree, and verification of numerical schemes is a topic by itself. I am not a fan of commercial codes and use research codes myself, and verification of certain schemes in our code are indeed challenging. I should say that for commercial codes used by absolute beginners, the challenge of verification of numerical schemes is completely barred, since most codes are proprietary and do not allow to dig deep inside. To understand the sole concept of verification, one should have flexibility of using a code where all the terms of PDE can be coded (or available to be modified) by the user.
Dear Shamudra,
First of all, the coding of the handeled problem is time consuming along with the validation part. After that the consuming part is to explain the outcomes of the simulation in terms of flow mechanism and the physics behind the mathematical results.
- Badges
- Science topic