[tchitchas]

Hello Guys, i'm starting to work on my final project and thought of making a 2D incompressible fluid flow solver using the SIMPLE algorithm with co-located arrangement and RHIE-CHOW interpolation. In case, I would develop my own code in fortran or matlab. I know there are several software that do this, but I think that would be a good way to understand the "root" of CFD and become an expert on it.

I have the theoretical knowledge behind the governing equations and the numerical methods used in their discretization, I have already worked with finite element (python) and I have knowledge in programming (Matlab, Fortran, Python).

Do you think it's an interesting job? I have a year to complete it.

Thanks for your attention.

[lcarasik]

Originally Posted by tchitchas

Hello Guys, i'm starting to work on my final project and thought of making a 2D incompressible fluid flow solver using the SIMPLE algorithm with co-located arrangement and RHIE-CHOW interpolation. In case, I would develop my own code in fortran or matlab. I know there are several software that do this, but I think that would be a good way to understand the "root" of CFD and become an expert on it.

I have the theoretical knowledge behind the governing equations and the numerical methods used in their discretization, I have already worked with finite element (python) and I have knowledge in programming (Matlab, Fortran, Python).

Do you think it's an interesting job? I have a year to complete it.

Thanks for your attention.

If you really like to understand the early nuts and bots of CFD, I highly recommend you do this.

[Ravindra Shende]

It will help you to understand how a numerical solution is obtained and the limitations of it. But this should be the only purpose of the exercise.

My technical director told me that research in CFD has reached a saturation point; there is not much left to be done.

[FMDenaro]

Originally Posted by Ravindra Shende

It will help you to understand how a numerical solution is obtained and the limitations of it. But this should be the only purpose of the exercise.

My technical director told me that research in CFD has reached a saturation point; there is not much left to be done.

I would know the name of your director to talk about ...

[Ravindra Shende]

Originally Posted by FMDenaro

I would know the name of your director to talk about ...

I do not wish to lock horns, but I do agree with my technical director.

That doesn't mean to say that I am not open to listen to counter arguments.

Having said that, I do want to listen to your views but I don't know how to ask you without causing an angry debate.

[lcarasik]

Originally Posted by Ravindra Shende

I do not wish to lock horns, but I do agree with my technical director.

That doesn't mean to say that I am not open to listen to counter arguments.

Having said that, I do want to listen to your views but I don't know how to ask you without causing an angry debate.

It is a really short sighted statement provided without much context. I'm quite surprised to hear anyone holds that opinion.

For the topic creator, the CFD field is rich in the research left to be done. If I need to point to anything, I'd point to two-phase flows and the closure relationships around them as a prominent example.

[FMDenaro]

Originally Posted by Ravindra Shende

I do not wish to lock horns, but I do agree with my technical director.

That doesn't mean to say that I am not open to listen to counter arguments.

Having said that, I do want to listen to your views but I don't know how to ask you without causing an angry debate.

There is no debate at all. At present you can use a commercial software that can give you a lot of beautiful coloured plot. There is no need to do CFD research to improve colour

[Ravindra Shende]

Originally Posted by FMDenaro

There is no debate at all. At present you can use a commercial software that can give you a lot of beautiful coloured plot. There is no need to do CFD research to improve colour

Do you mean to say that commercial software give inaccurate results so they are required to do research to improve accuracy?

[Ravindra Shende]

Originally Posted by lcarasik

two-phase flows and the closure relationships around them

Wouldn't that be an analytical work?

[FMDenaro]

Originally Posted by Ravindra Shende

Do you mean to say that commercial software give inaccurate results so they are required to do research to improve accuracy?

I have my opinion based on some benchmarks. And the improvement in the accuracy is not the only issue about.
But I suggest to check by yourself and build your own opinion.

[Ravindra Shende]

Originally Posted by FMDenaro

And the improvement in the accuracy is not the only issue about.

So I take that you agree when I said 'commercial software give inaccurate results so they are required to do research to improve accuracy'.

So what research do they have to do? Do they have to invent higher order accurate schemes? But higher order schemes are already out there, they just didn't use it for some reason.

[FMDenaro]

Originally Posted by Ravindra Shende

So I take that you agree when I said 'commercial software give inaccurate results so they are required to do research to improve accuracy'.

So what research do they have to do? Do they have to invent higher order accurate schemes? But higher order schemes are already out there, they just didn't use it for some reason.

Yes, I have seen what happens when different CFD codes are used to simulate the same problem.

Developing high order schemes working well on unstructured grids is just an example of one of the fields of research.
But often the quality of a solution for complex flow problem (turbulence, reactive flows, multiphase, shocks, etc.) does not depend only on the accuracy but also on the formulation adopted as well as the modelling of the physical phoenomena. That means we have to consider that the orginal PDE system of equations could be modified before to be discretized. Then further specific problems arise when we consider specific issues such as boundary conditions for LES/DNS, or interface conditions for embedded LES or what should be done using DES.
Then we can talk about HPC reasearch in the CFD field.

I am sure to forget many other topics...

[JBeilke]

There are very different tasks you can do in the world of cfd:

• writing a code
• properly using a code to produce "correct" results
• solving engineering problems by using cfd codes

[JBeilke]

Without having a plan, what problem you want to solve with your work it is just an programming exercise or the 500th reinvention of the wheel. You might be able to download such a code.

I would suggest you to search for an unsolved problem and try your best to work on it.

[Ravindra Shende]

Originally Posted by FMDenaro

Developing high order schemes working well on unstructured grids is just an example of one of the fields of research.

The schemes do not work well because of the dispersive nature of the discretization error. To control this error one needs to come up with good limiters and this does require research. The outcome of this research will be method that will lead to more expensive simulations. The cost that went in that research is also going to be large.

Originally Posted by FMDenaro

But often the quality of a solution for complex flow problem (turbulence, reactive flows, multiphase, shocks, etc.) does not depend only on the accuracy but also on the formulation adopted as well as the modelling of the physical phoenomena. That means we have to consider that the orginal PDE system of equations could be modified before to be discretized.

This is not a research in CFD, this is research in FD and mathematical modelling of fluid flows. Mr. Carasik was talking about the same thing in his last reply. Learning the roots of CFD is not going to enable one to do research in this field.

Further, any outcome of that research is going to be an equation with convective, diffusive or source terms and CFD already have well developed methods to numerically solve these equations. And that is what CFD is about - Numerically solving the equations from the mathematical models. Research to come up with methods to numerically solve the equations is research in CFD.

Originally Posted by FMDenaro

Then further specific problems arise when we consider specific issues such as boundary conditions for LES/DNS, or interface conditions for embedded LES or what should be done using DES.

These techniques are again more expensive and the research required in them is also expensive. Any outcome of the research is again going to lead to more expensive simulations.

Originally Posted by FMDenaro

Then we can talk about HPC reasearch in the CFD field.

Here again you are talking about research in computer science and not research in CFD.


Research in any field is driven by the cost required to do the research over the savings in cost the outcome of the research is going to produce. That is why research in batteries, super-capacitors, electrification of vehicles is in full swing.

The costs involved in doing research in CFD (Developing high order schemes that work well on unstructured grids, LES/DNS/DES, etc) are too high and the outcome are going to result in more expensive simulations. That is why research in CFD has slowed down. And that is what I meant when I said that research in CFD has reached to a saturation point.

[FMDenaro]

Originally Posted by Ravindra Shende

The schemes do not work well because of the dispersive nature of the discretization error. To control this error one needs to come up with good limiters and this does require research. The outcome of this research will be method that will lead to more expensive simulations. The cost that went in that research is also going to be large.



This is not a research in CFD, this is research in FD and mathematical modelling of fluid flows. Mr. Carasik was talking about the same thing in his last reply. Learning the roots of CFD is not going to enable one to do research in this field.

Further, any outcome of that research is going to be an equation with convective, diffusive or source terms and CFD already have well developed methods to numerically solve these equations. And that is what CFD is about - Numerically solving the equations from the mathematical models. Research to come up with methods to numerically solve the equations is research in CFD.



These techniques are again more expensive and the research required in them is also expensive. Any outcome of the research is again going to lead to more expensive simulations.



Here again you are talking about research in computer science and not research in CFD.


Research in any field is driven by the cost required to do the research over the savings in cost the outcome of the research is going to produce. That is why research in batteries, super-capacitors, electrification of vehicles is in full swing.

The costs involved in doing research in CFD (Developing high order schemes that work well on unstructured grids, LES/DNS/DES, etc) are too high and the outcome are going to result in more expensive simulations. That is why research in CFD has slowed down. And that is what I meant when I said that research in CFD has reached to a saturation point.

Now I see that you have an idea of the research in the CFD field that is totally wrong. Just use the index of the book of Peric & Ferziger to see that the tasks in CFD cover all the field you do not consider to be CFD.

A good researcher in CFD has to work without frontiers between mathematics, physics, numerical analysis and information technologies.

You talk about "The cost that went in that research is also going to be large." That means almost nothing in the research field where you cannot say what you will discover and when.

[arjun]

Originally Posted by Ravindra Shende

It will help you to understand how a numerical solution is obtained and the limitations of it. But this should be the only purpose of the exercise.

My technical director told me that research in CFD has reached a saturation point; there is not much left to be done.

The fundamental CFD problem if it could be tied to solving of Navier stokes then it is not solved AT ALL,

The pure navier stokes when written in matrix form are saddle point problem and there is no good and efficient way of solving them, And indeed it is a big area of research,

What commercial packages solve is modified form of navier stokes where they add terms to remove that 0 block of saddle point. There by changing the definition of problem (means that original problem is not solved at all, we have not even started :-) )

Second even in second modified form where that extra term is added (sometimes called Rhie and Chow dissipation) there is no closure to how much this term has to be.

This specially manifests in break down of cfd solvers at very small time step sizes (you would think it shall become stable and accurate as time step reduces). This problem is unsolved.

If he knows a solution then I (and prof Milovan Peric) would love to hear it.