Dear Karolis Janusevicius,

Thank you for starting this important discussion.

According to my point of view I think that the bast way to start creating a CFD code is mastering programming languages, mainly C++ and FORTRAN or even Python. There are many websites that give powerful courses in this area. Second of all, it is very important to be familiar with advanced numerical methods used in CFD, especially how to solve partial differential equations which describe most physical phenomena or how to solve an eigenvalue problem . Once the code is written, the best way to validate it, is to compare the results with exact analytical solutions if they can be found, for example, with the Poiseuille solution in a pipe flow for the laminar velocity profile.

Thank you Christoph for reply - I was thinking about definition of, correct way to start and perform CFD based work, when starting this topic. Just looking for suggestions and tips which could be usefull not only for me.

To reduce abstraction a little bit -this question comes from engineer who are not mathematican or computer engineer. Or in other words - looking for for a place to enter the CFD-forest and not get lost :)

Hi,

This is a very good question I had when I started CFD. There is so much thing to look at...

As say earlier, you have to define a clear objective like reproduce experimental data. An important thing about that, you have to be fully aware of the limits of your experimental data before making any comparison. Indeed, it will be really helpful while calibrating some submodels.

Before beginning the model, if you have enough time, you can take a quick look at a reference for CFD such as this one:

http://books.google.ca/books/about/An_Introduction_to_Computational_Fluid_D.html?id=RvBZ-UMpGzIC&redir_esc=y

It will give you the basics and some tips for the construction of your model. You'll be able to find some of the info that Christoph and Abdallah told you to get.

I think there is two most important steps that you have to be meticulous on to be sure of your results and be able to present them:

1/ Boundary conditions: according to your system, think a lot on it because even if it seems easy to put the boundary conditions, it is one of the first error in a simulation. So be careful with this.

2/ Spacial and temporal discretization: you have to study the sensitivity of your system on those two parameters. When you think your mesh and your time step are well defined, you should calculate your uncertainty using the global convergence index (GCI). You can find more information in this paper

http://fluidsengineering.asmedigitalcollection.asme.org/article.aspx?articleid=1434171

Then, you can make all the tests you want to with your model etc etc.

Hope it will help you

I agree with Jonathan

Try to follow good scientific papers!...

I recommend

http://www.cambridge.org/us/academic/subjects/computer-science/scientific-computing-scientific-software/verification-and-validation-scientific-computing

It gives a good overview on simulation verification and validation. Uncertainty analysis is indispensable as it allows you to check whether your model results align with the real word. In fact, there is no quick fix to achieve meaningful results despite of the progress CFD has done in the past decade.

HI, Mr. Janusevicius

I'm not sure about the environment in which you plan to work, but I can bring some ideas to the discussion on the industrial environment in which I have experienced in recent years (or decades). The introduction of a new technology is always accompanied by a lot of expectation which leads us to make the following basic mistakes :

1. Model a problem whose goal is not clear, for example, improve an air intake manifold : it may suffice to get an increase in static flow on a certain condition of differential pressure input against output pressure, or perhaps minimize the pressure drop across the collector for a given flow condition , but initially define what should be the output parameter before creating the model . To obtain this definition there are several options on how to attack the problem and consideration of the phenomena involved on it.

2. Often, in order to get a quicker solution to the problem, we adopt simplifications of the mathematical model masking phenomena that are important for the problem we set out to solve, for example, to consider the fluid as incompressible in areas where there is large variation in velocity.

3. Avoid, at first, look for “absolute” numerical results, for example: this concept will give us a flow of x l/s. Start setting “relative” numbers, like: the concept "A" shows an improvement in the flow of y % in relation to the concept "B" . To calculate reliable absolute values much experience is needed and mostly very accurate input data is required.

4. Start creating models that include phenomena you know. My first models had almost nothing.

go ahead!!

You might want to read this attachment.

Initially you can start with simple problems like one-dimensional steady-state heat conduction. you will get an idea about the basic rectangular grid making and use of numerical schemes like regula-falsi method.

Dear Mr. Januseviciu,

your concern is very useful to most of the people staring their CFD works. i personally would agree with most of the comments above regarding the need of knowledge of programming languages but i will try to make i easier on you in case your doing this for temporary reasons.

-i would start by defining the problem and try to imagine how the outcome should be in case experimental data is not available.

-i would then go to the manual and ready the sections relevant to my problem. for example, if combustion exists, i would certainly go and read all the relevant sections regarding the equations to be solved as well as the recommended models to be used.

-i would then go into solving the tutorials staring with the easiest ones if you don't have any experience with the solver and go up in complexity until you get to a case similar to yours.

-most probably, you will need to go back to the users manual to understand better the physics and the equations the solver is solving. this will always help you refining your solution.

my comments are intended for users that are after quick solutions for practical problems. if more detailed research modeling is needed, then may of the above mentioned comments are very useful. you know better what you are after so you will decide which way to go.

good luck,

hello Karolis.

Just one comment.

First: Before you start any CFD job, correct boundary conditions are a must. They predefine if your results will be realistic or not. And not always it is a simple task.

I give an example. I had an opportunity to check one CFD job from one CFD company. They solved desulfurization using rotary atomizer with several radial channels from which the lime slurry was sprayed in the absorber. And they calculated velocities of the slurry that leave the radial channels simply from volumetric flow of the slurry and the total cross section area of the channel. Big mistake. When atomizer rotates, the cross section is occupied by the slurry only in very small cross section (a kind of crescent), i.s. the slurry is pushed to one side of the channel. So the velocities of the slurry particles are much much higher and consequently, the trajectory of the particles as well and thus much more slurry can reach the wall and build up there. So in this case results were rubish.

Second: before you start, you should make a good analysis from the point of physiscs of the problem. For example, if there are heat sources (and thus natural convection may appear - depending on the "driving force") in your solution domain, you will very probably never get a steady solution and will have to switch for transient.

In one word, a good knowledge of the physics (fluid mechanics, heat & mass transfer etc.) is essential, not to just generate nice pictures.

As I said elsewhere, try simple systems first, such as my CFD Letters Editorial reprinted in my Quantum Theory of Turbulence, available from my list of papers downloadable from ResearchGate. Start with multi-specie simulations.

To start with simple simulations, you might want to read the attachment and consider multi-specie CFD simulations. The results may surprise you, a turbulence model will come out instead of massive computer effort. Pay attention to the CFD Letters reprint. The rest of the book is challenging and will require more effort. Good luck!

I think you do understand CFD when you have written one youself. Her is a link to some books, that introduce you step by step from discretization, stability, solver and so on.

http://www.eesbooks.com/booklist.html

The author offers long-distance courses for the AIAA.

When you are new in CFD it is good to make a very easy prediction which has been tested by others. On the homepage www.cfd-benchmarks.com you will find a simple 2D isothermal case which can be usefull if you are working with room air distribution. More than 50 authors has published results based on different turbulence models, different inlet conditions and so on. You can compare your results not only with Measurements but also with others CFD Work.

Foremost understand that in CFD the mesh plays a dominant role, each mesh can produce a different result. Automatic adaptive meshing is important(AAM). If AAM is not possible, try to show mesh convergence using three different meshes (coarse, medium, and fine meshes) placing the vertices where you think the gradients are important. Lastly, parallel computing is a must for unsteady computations on fine meshes.

Dear Karolis,

For brevity, I tell you the key feature that you have not mentioned in your ''Must Do'' topic. Have a profound literature study. Starting CFD analysis has its own risks. sometimes you have trouble with geometry's nature, ie, high aspect ratios in multi scale problems(namely micro scaled flows in multi domain problems), trouble with grid generation and the order and scheme that you generate elements, and finally probable obstacles in deriving equations and calculation procedure. In a sum, while you have more profound literature study, you will know most of obstacles and you only need to troubleshoot some probable ones which would be counted as your own research originality in that CFD study.

Best

Hi, we know the problems students facing in ICEM CFD. We have helped many students in this regard.

Here is our course published on udemy on ICEM CFD Hexa. Hope this will help students in learning meshing.

https://rebrand.ly/ICEMCFD

The best tips is given by :https://www.researchgate.net/profile/Christoph_Glawe