(1).. "I do a lot of CFD work and I don't think any of my solutions are grid-independent. " Well, I agree with you that it is a common practice nowadays, but, that is because someone pushed it into this situation. (2). CFD in many ways is like cancer research. Sometimes, people like to push it into application or commercialization. (even if the drug is not ready to be released to the public.) (3). I can't change how other people view the CFD from their side. So, if you think it is a tool, then it is a tool. (4). After having said that, and after having answered questions for more than two years here, all I can say is: if CFD is a tool, then it must be a very difficult tool to use. (5). From my point of view, CFD will be used in the places where it will have great impact on the safety of products, or even survival of human being. When that time comes, it would be too late to know that the mesh independent solution had a key role in the problem. (6). Perhaps, CFD is like making hamburgers. No two hamburgers are identical. Unfortunately, we know that CFD is being used in aircraft, engine, combustor, automobile, air conditioning, power plant,...etc..., where safety, efficiency, pollution etc... are very important. (7). If the solution is not mesh independent and is acceptable in industries, then we can just set the standard mesh as 10x10x10 for every cases performed throughout the industries. This would cut down the computing time, schedule delay, and finish the project ahead of time. (8). It would be nice to see that CFD can produce results with "hand-made" label on it. It is collectable and no two solutions are identical. It also comes with personal flavor on it.

John,

on this issue I think Mike and Clifford's comments are more useful and closer to reality.

This reminds me of a previous experience (long):

A few years ago as a graduate engineer in a large company I had to design a new reduction cell lining (for aluminium smelting) using an inhouse thermal code (not cfd - even though fluid flow and mass transfer are important in this application). The new reduction cell was based on a novel design concept and so very different from standard design which our code was developed for. However while we knew (some of) the weaknesses of the code, we did our best to evaluate designs - and every body from the MD, 3 GMs and just about every operator had their own pet idea of the 'perfect reduction cell'. I used the code to rank the designs and then we used some experience and basic 1d heat eqns to help us verify the design etc. Of course, I also had to explain to individuals like my GM why his design wasn't the best!

The design was built and then I and a team of people spent a few months measuring performance parameters etc. In the end it didn't meet the requirements, because we had underestimated the heat transfer co-efficient between a solidfying electrolyte and a MHD driven wave free surface (unsteady HT). So the next step we undertook was to calibrate our model to the current design to estimate/fudge factor the HT coefficient and reuse the code to design the next cell. After one iteration we solved this problem and although we had to build a cell and operate it etc, which may have cost a few $100,000 we did solve it - we also got excellent measurement data from the process.

So in this case we had a poor model but still managed to use it to help improve our designs. Of course, if we had a more fundamental model and a knowledge of the HT etc., we potentially could have achieved the same output without iterating on the design-build-evaluate loop. But for some of these processes, building good models is extremely difficult - in this case we had a multiphase system, 1000 degC, with solidification, turbulence, MHD and unsteady operation!

I still haven't seen a cfd model for the entire process - now 3yrs on! One of our senior engineers spent over a year trying to create a simple cfd model of this process and failed - not because he wasn't good enough (he had a PhD from Imperial College), but because its just bloody hard. Obviously we couldn't meet the project deadlines if we had of gone off to build a model rather than trying to build the cell.

So within the company's product timeframes, performing detailed research to support product development is not _always_ the best approach at meeting a project's short term goals.

However, one thing that this company and I'm sure others do too, is to (partially) ignore the long term development which would give you improved models over a 5-10yr time frame. After seeing the opportunity I proposed to do a PhD on this area to address this issue, but was knocked back. I left to do my PhD on something else.

So I think the challenge for management is to balance the short term needs for results, which are clearly needed to meet project deadlines etc., against the longer term ones in which sustained effort is put into improving the design system (codes, methods etc.). Its not obvious how best to do this in many situations and managers often have to balance competing interests etc. I don't think technical managers are generally as bad as John says...but I find that they can lack a bit of vision for change...

I think if you lean too far either way - like you want a perfect model but never have time to use it, or you use a crap model all the time and never improve it, you're gonna be in trouble - and out of a job!!!!

Just my thought

Greg

Actually CFD is a model - its almost never exact even with grid independence.

For even simple CFD analysis - like flow in a square duct or pipe, John do you suggest that we wait until computers can do DNS for us before using CFD??

And for complex physics, do you _know_ the heat and mass transfer relationships, correlations exactly? Boundary conditions? etc. etc. Let's not even discuss combustion and multiphase systems.

In your area of CFD - aerodynamics - cfd is very mature - in other areas like multiphase its currently very immature.

I think the point here is we should strive for perfection but not set it as the required benchmark. A model can be useful without being perfect - in fact by definition a model isn't a perfect represenattion of reality - and that's often its strength, and not generally/always a weakness.

Greg

Of course one cannot know all there is to know about the physics of the problem..however one thing that can be done is to determine whether the meshing is a variable in the solution..which of course it shouldn't be..so by getting a mesh converged solution you are just eliminating one unknown which is always a good thing correct?? Convergence of a discretization of a problem is something that is important in all approximate solution techniques whether it is finite elements or some other Ritz method in structural analysis or a series solution in a perturbation type problem...I don't see why the same importance shouldn't be place on the approximate cfd solution

(1). I had just proposed a degree of difficulty standard to measure the cfd problem. (2). For a problem with low degree of difficulty, one can validate the cfd solution and apply it in the routine engineering practice. This process requires the training of cfd professional. (3). In ISO900, you have to calibrate your instruments used in the design process on the routine basis, by the approved professional laboratory. (4). For high degree of difficulty cfd problem, it will remain in the research field. Trial-and-error approach might give you a new design alternative, but it can be used only in the fire-fighting situation, like NASA's Apollo-13 trip to the moon. Or the one which exploded in the mid-air when the whole world is watching. (5). So, apparently, for low degree of difficulty, or low risk cfd problems, one can push it to the extreme and use the cfd process in any way you like. But for other critical cases (sometimes it is hard to identify in the begining), it is better leaving the job to the professionals. (6). We are here to shape the CFD industries, so, if one single error in a CPU chip is not acceptable, then we should try to get the mesh independent solution in our cfd approach (at least to demonstrate that mesh issue is included in the quality consideration) (7). The other way to do is to set the mesh at 10x10x10, and calibrate the constants and the models such that the cfd solution will match the test data, thus make it useful in the routine engineering practice. In this case, there is no need to invent a new accurate model, and all one needs to do is to validate the result carefully. But for today's computer, it is easier to refine the mesh instead of refining the constants.(this was not the case back in 70's and 80's, when the memory was limited and expansive.)

(1). DNS? I am not there yet.(need to learn a lot)(2). For CFD to become an engineering "tool", it must be validated first, so that it can produce consistent results. (3). This could be a research project at a research lab ( university or government lab). This could be done in industries also. (4). Validation requires cfd professionals, and also experimental researchers. Once the particular cfd approach (algorithm and models) is validated for the particular cfd problem, it can be used in design practice on the routine basis. (5). Like ISO9000, CFD validation is a necessary step to engineering applications. We should try to shape the CFD field like the computer development (i.e. CPU, RAM, data storage, etc..), Internet development, new drug development, etc... In this way, there will be life time career for cfd students and professionals.

Mike: that was exactly my point: CFD is a means to an end. Unfortunately the approach to CFD in University is to isolate it from the engineering process. You have to be able to engineer so in that sense CFD is a tool. I get tired of people refering to CFDers as "scientists" for example in magazines or such where they say something like "NASA scientists use computational simulations of fluid flow to design aircraft engines". Scientists don't design anything. Engineers do. Continuing to approach CFD as "Science" leads to a situation where we're continuously hammering away at minutae.

You missed my point entirely John. I said there are only three people with "CFD" in their job title. We have perhaps more than a hundred engineers who use CFD and understand extremely it well. However their job title is something like "Aerodynamic engineer", "Heat transfer analyst" or "Combustion engineer". My point was we don't have a whole lot of people whose sole purpose in life is to develop CFD code (that's why we have an R&D installation). We use CFD to design. So you the prospective graduate should not look specifically for "CFD" in the job title. Engineers design.

John I know you have worked or are working for companies in the same business as I'm in and people who use CFD don't have "CFD" in their job title.

(1)."We have perhaps more than a hundred engineers who use CFD and understand extremely it well. ".....HOPELESS!!!.....TOTALLY HOPELESS!!! (2)."My point was we don't have a whole lot of people whose sole purpose in life is to develop CFD code (that's why we have an R&D installation). ".....HOPELESS....TOTALLY HOPELESS!!! (3). we have engineers world wide using "CFD codes" everyday and asking a lot of questions here at cfd-forum everyday. And you are telling me that your company is such a company that more than a hundred engineers are running "CFD codes" extremely well everyday? (4). Can your company just write a short program and let the computer to run the "CFD codes"? Just write some script codes and put your "CFD codes" running on the computer day-and-night. (5). CFD is "Numerical Analysis and Mathematical modeling in Fluid Mechanics". You have probably only three qualified engineers there, and the rest of the engineers are "CFD codes" engineering aids. .....HOPELESS.....TOTALLY HOPELESS.... (6). If your engineers can't do numerical analysis, then they can't do CFD. If your engineers can't do mathematical modeling (turbulence modeling, turbulent reacting flow modeling, multi-phase modeling, etc..), then your engineers don't understand the physics of fluids. (7). You can "mass produce" the solution from cfd codes, but what about the accuracy of the solution? Your "RESEARCH CENTER" can easily write the codes such that the computer will generate the series of soluitons without human interaction at all. (8). After all, you are still thinking that "CFD" is to "get someone to run a code". Yes, you can do that. You can get someone to setup input to run different cases of cavity flows with different height and witdth. But it is much easier to write a program to change the aspect ration automatically and let the computer run it automatically. Do you see what I am trying to say? (9). If your company's definition of CFD is to train a lot of engineers to run codes developed by your "RESEARCH CENTER", then you can easily replace your one hundred engineers by simple optimization program with a little AI in it. (a parametric loop through the code is good enough.) (10). By the way, I have studied this approach back in 70's. You really don't have to change the geometry model or remodel the geometry for each case, all you need to do is to plan ahead and write a parametric model and let the program to change it automatically. Same is true for the flow conditions. This is exactly how the experimental tests are run in wind tunnels or test facilities. You simply don't run it cases by case, you create a complete map first. That map created and implemented by your "RESEARCH CENTER" will get rid of one hundred engineers in your place.

(1)."We have perhaps more than a hundred engineers who use CFD and understand extremely it well. ".....HOPELESS!!!.....TOTALLY HOPELESS!!! (2)."My point was we don't have a whole lot of people whose sole purpose in life is to develop CFD code (that's why we have an R&D installation). ".....HOPELESS....TOTALLY HOPELESS!!! (3). we have engineers world wide using "CFD codes" everyday and asking a lot of questions here at cfd-forum everyday. And you are telling me that your company is such a company that more than a hundred engineers are running "CFD codes" extremely well everyday? (4). Can your company just write a short program and let the computer to run the "CFD codes"? Just write some script codes and put your "CFD codes" running on the computer day-and-night. (5). CFD is "Numerical Analysis and Mathematical modeling in Fluid Mechanics". You have probably only three qualified engineers there, and the rest of the engineers are "CFD codes" engineering aids. .....HOPELESS.....TOTALLY HOPELESS.... (6). If your engineers can't do numerical analysis, then they can't do CFD. If your engineers can't do mathematical modeling (turbulence modeling, turbulent reacting flow modeling, multi-phase modeling, etc..), then your engineers don't understand the physics of fluids. (7). You can "mass produce" the solution from cfd codes, but what about the accuracy of the solution? Your "RESEARCH CENTER" can easily write the codes such that the computer will generate the series of soluitons without human interaction at all. (8). After all, you are still thinking that "CFD" is to "get someone to run a code". Yes, you can do that. You can get someone to setup input to run different cases of cavity flows with different height and witdth. But it is much easier to write a program to change the aspect ration automatically and let the computer run it automatically. Do you see what I am trying to say? (9). If your company's definition of CFD is to train a lot of engineers to run codes developed by your "RESEARCH CENTER", then you can easily replace your one hundred engineers by simple optimization program with a little AI in it. (a parametric loop through the code is good enough.) (10). By the way, I have studied this approach back in 70's. You really don't have to change the geometry model or remodel the geometry for each case, all you need to do is to plan ahead and write a parametric model and let the program to change it automatically. Same is true for the flow conditions. This is exactly how the experimental tests are run in wind tunnels or test facilities. You simply don't run it cases by case, you create a complete map first. That map created and implemented by your "RESEARCH CENTER" will get rid of one hundred engineers in your place.

LOOKS LIKE YOU ARE JOBLESS

(1). I have been a cfd consultant for ten years. (2). Currently, I am developing smart software for design applications. (3). Some cfd code vendors used to work for large companies, but I guess they are jobless now. They are working for themselves. (4). I just came back from one hour bicycle ride in the beautiful park. There was a rumor sometimes ago about the GE's plan to lay off over twenty thousand workers world wide. I think, it depends on the outcome of the merger with Honeywell. The decision will be in the summer, June or July. (5). Friends in Honeywell have already said the comapny's plan to reduce workers even before the merger deadline in the summer. (6). Running a cfd code will not make the engineer's job more secure, in view of the current quality of solution produced by most commercial cfd codes. (I am sure that there are exceptions, and some good results somewhere.)

(1). I have just heard that the merger between GE and Honeywell will save several billion dollars, as announced by the GE CEO. (remember that both companies have units in turbomachinery area.)

[mathew33]

Sure do you wish to do online outsourcing jobs through internet from home itself?