CFD Online Discussion Forums

CFD Online Discussion Forums (
-   Main CFD Forum (
-   -   Where do we go from here? CFD in 2001 (

John C. Chien December 30, 2000 03:37

Where do we go from here? CFD in 2001
(1). As the year of 2000 is coming to an end, very few people still remember the Y2k problem of the last century. The only thing you still remember is: we are still talking about CFD. Finally, we are going to move into the next century. (2). I am writing, because I am exercising my brain, and I need that whether I am working or not working. I guess, you are working because you have to support yourself or your family. (3). Before I say something about the year 2001, we need to get an overall view of the last decade, that is from 1990 through 2000. (4). First of all, after many years working with defense industries, I was finally able to put together an integrated CFD code for PC/286, including graphics on my own time, and on my own PC. It took me about 8 years to get the final idea in a working code. (from the time I bought the first Radio Shack model III sytem for about US$2500 in 1982 dollar.) It is interesting to know that it also took roughly about the same time for Microsoft to develop a good working Window 3.1 operating system. (5). So, in the last ten years, we have moved away from the super-computer systems and mini-computers into the network workstation/PC system. (6). At the same time, an interesting sytem of communication become available, that is "the Internet". (7). These two systems will start to change how we do CFD in the future. (8). In 80's, when the defense business was the main drive behind the CFD, it was necessary to work for a defense contractor to use the super-computer and to work on the CFD technology and applications. The nature of the work and the enviroment limited the access to only the best available engineers in the CFD field. So, there was no real competetion from other engineers, because they normally do not have the experience to access the mainframe or the technical problem. It provided a good shield for the CFD researchers. (9). With the support from the defense industries long disappeared, the commercial CFd codes become available for the general public, the commercial industries sector. The play field is now in the general public, and the access to the commercial codes is no longer limited to the shield CFD researchers in defense industries. This is the new play field of the post-cold-war CFD, commercial CFD codes and the commercial applications. (10). The other difference between 80's and 90's is the open-source-code vs the black-box approach. For a CFD researcher to understand the existing code, to modify the code with new ideas, and to apply it in solving the real problem, he must have the source code and the related document. This was the norm in 80's. With the commercial CFD codes in 90's, the users can no longer access the source code he is using. This coupled with the general trend of Object-Oriented Programming, and Component Objet Model for multi-tier network, the users in 90's commercial CFD codes become pure users, instead of the CFD researchers fo 80's. This is a fundamental change in the relationship between CFD and the researchers/engineers. (11). In 90's, there was also a very strong trend, which is still being continued, that is the company merger. The number of airframe companies has been reduced from 7 in US down to 3 or 2. The engine companies are also disappearing very quickly in the last 5 years. So, the CFD related departments also disappeared between each merger. As a result, the technical problem is ill-supported in the company, because the company tends to sell the codes as asset and keep the codes in transaction, without the supporting engineers naturally. (you can sell the codes, but you have to feed the engineers) There, the biggest mistake is the "separation of researchers from the codes". This is "irreversible process", and the end result is the company will have to die a second time. (12).The other big problem in commercialization of CFD codes is: it requires a lot of support engineers and sales persons, in order to keep the operation alive. As a result, the license fee is very high, relative to the 80's free source code approach. In addition to that, the research part of the code development is also limited, because of no profit no research. While the 80's approach was based on the company's contract work (or IR&D) with the government agency. There was minimum sales activities and essential no support at all. If you turn in the final source code to the sponsor, I guess, the sponsor can just read it and modify it to do whatever he likes to do. This is no so in the 90's black box approach. The user must depend on the support engineer to learn the tricks of operating the code, since the source code is not available to the user. (13). I am not about to tell you how to solve these problems, or how to approach these problems. But after reading the above message, I am sure that you have learned that the CFD of 90's is no longer the same as the CFD of 80's. It is hard to say that a CFD user of 90's commercial codes, is still part of CFD process at all. It becoming more and more like an engineering aid of 80's, although you might not agree with me in the observation. (14). And now with US$1000, you probably can get a PC with one GHZ CPU and 512Meg RAM system, which is 1000 times faster, 10,000 times more memory, two and a half times cheaper than my 1982 Radio Shack Model III system. And I suddenly realized that, I can become the CFD researcher, the developer, the boss, and the global market is just at the touch of the Internet connection. (15). The availability of the computing enviroment and the accessibility of the global market through Internet, will finally put the control of the technology development at the finger tip of an old fashion CFD researcher. (16). As you move into the next century, you will probably still worrying about the convergence issue of a commercial code. But at the same time, a brand new generation of CFD will emerge silently, across the Internet. I am not sure, whether it is still an integrated CFD code in single program, or distributed components across the Internet. I am sure that, it is not going to be the current approach. That is the rather in-efficient, backward approach of 90's. (17). By the way, you are encouraged to think about all these issues, because it is you who will be in control of the future CFD.

Jim Park December 30, 2000 12:30

Re: Where do we go from here? CFD in 2001
Well John you sure brought a lot of the past back.

I remember, as you do, the hassle in getting 10 minutes on the main frame overnight, the messenger coming down the hall with my card deck and green-bar output. I remember doing hex arithmetic to track down the constant error codes, then running to the key punch to replace the bad statements with what might be a correction. Then submit the thing to run 10 minutes the next night. Gone but not mourned!

Recently, I was able to repeat a benchmark calculation on a 2-d code from the 70's. The sample problem (a thermal driven cavity) took about 260 seconds using the CDC 7600, the mainframe of choice at Los Alamos in those days. When I ran the same code, same input, on my 450 MHz Pentium II system, it took 26 seconds.

I agree that use of the commercial codes requires some caution. In my experience (with 3 commercial codes at various times), the front line of technical support is often a green PhD with a dissertation using CFD but damn little experience in fluid mechanics. These folks are surely very smart, but the user must do his own benchmarking and form his own conclusions about the results he gets. I think your (quite valid) concern is that the typical CFD customer doesn't have the budget (or background?) to do that.

I hope that you're a bit pessimistic.

In any case, with the power we have on our desks, we should do a lot of cross-checking of the results from the commercial codes. Perhaps there's a cottage industry in doing this?

Happy New Year!

John C. Chien December 30, 2000 13:28

Re: Where do we go from here? CFD in 2001
(1). Pessimistic? No. (2). Unless you are in the ".com" business, or stock market in a foreign country, such as Taiwan, where the stock index droped 40% in seven months this year. (3). The current computer and commercial codes still can not solve the fluid mechanics problem by itself. So, it all depends on the training of the engineer who is using it to solve the problem. (4). The basic problem is: you had a chance to see the source code and use a key punch machine, while the 90's generation has no idea of what is inside a code. (even as an engineer working for a software vendor, he probably does not have a good picture of the whole software package. He probably has access to only a small portion of the code.) (5). Anyway, I will get into the software side, when I have the time. I still have to move the two small trees from the front yard to the backyard today.

Des Aubery December 31, 2000 02:50

Re: Where do we go from here? CFD in 2001
Hi John,

Interesting how history comes to repeat itself...

A good example of the "internet approach" would be the FeatFlow software from Dortmund University.

This software is freely available over the internet - I am currently working with it for research purposes.

Best regards,

Des Aubery... (East London, E.Cape, South Africa)

ken elms December 31, 2000 10:22

Re: Where do we go from here? CFD in 2001
Hopefully,CFD can maintain its rate of progress inkeeping with the rapid change in home computer power.The successes of its uses too must be more forthcoming in product areas-I`m still struggling to get more technical references into my rotodynamic pumping site.

Advancements being made in meshing and codes too would be welcomed.

CFD must be allowed to flourish ,be seen to flourish in as many forms healthy to better understanding-limitations included -tradeoffs.

This site must be congratulated in its open-ness of debate too.

Ghanshyam Singh December 31, 2000 10:35

Re: Where do we go from here? CFD in 2001
If we look at the history, 90s approach was responsible to convert "Costly Fluid Dynamics" of 80s to a "Commercial Fluid Dynamics" the so called black-box approach with the aid of beautiful graphics. Some people call it "Colorful Fluid Dynamics" but I call it "Click....." or rather "Clerical Fluid Dynamics"!!! But it will not take us any where.

Being a great admirer of "Linux", I believe in the open source or as you call it "Internet approach". In this regard I have deep sense of respect and appreciation towards all those who have made their contributions. I wish that we must proceed in this direction, if we want to achieve something substantial. It is good that CFD has come-out from defense industry and it is being used as an "optimization tool" in other places as well. I do agree with John that CFD has produced more colorful pictures than useful design, but I have a hope. Let me take this opportunity to wish you all the best and happy CFDeing!!!!! in the new Millanium.......


Fred Uckfield December 31, 2000 12:25

Re: Where do we go from here? CFD in 2001
I don't see anything too wrong with a black box approach (rather the desire to achieve one). Why would you expect a secretary who wishes to use MS Word to write a letter regarding an invoice to know anything about the inner workings of what she is using. All (s)he cares about is probably what font to use, the general style, how to print and where to save to.

When applied to CFD you would like an engineer who has a working knowledge of fluid dynamics (just as a secretary has knoweldge grammer, document styles etc.) to be able to perform computational ananlyses of fluid systems.

CFD vendors would also like to achieve this as there are many more engineers than CFD researchers, the former having far more disposable wealth than the latter (and all companies wanting to make more and more money, what with money making the world go round).

I know there are some obvious holes in the above analogy but the vision is true. There are many CFD researchers high in their towers who are guilty of working for the sake of remaining aloft. The new breed of CFD in the 90s, now and beyond requires (academic) CFD research to be refocused so that CFD itself may be applied in anger by the masses.

Happy new year and BRING IT ON!!


John C. Chien December 31, 2000 13:38

Re: Where do we go from here? CFD in 2001
(1). I think, the need to explore the inner structure of a code, comes from the curiosity of the failure of the code,in the form of solution convergence,accuracy, etc... (2). Luckly, a secretary using MS Word does not have to worry about the mesh generation, solution convergence, etc... (3). And I think, even a registered PE (professional engineer) will have difficulties in these areas when using a commercial code. (4). Just read the commercial code specific forum here, then you will be curious why there are so many questions related to the operation of commercial codes. At the same time, no one seems to have trouble with Internet or e-mail. (5). The reason why a secretary can use a postscript font in her document by selecting it directly from the MS Word, is simply because: there was a company invented the postscript, and each character was modelled using parametric curves and saved, and the postscript language was made available to every printer. These activities represent the "Research , Development, and validation". (6). So, for CFD to reach that stage, there is a long way to go. It is like asking the secretary to use a CAD program to draw the characters in her document. I think, the commercial code should invest more in research and development, if survival in the next century is one of the goal.

George Bergantz December 31, 2000 16:19

Re: Where do we go from here? CFD in 2001
I completely agree with John's remarks. This whole business has a somewhat different expression in applied and theoretical earth sciences than it does in engineeering practice. There are (at least) 2 cultures with very different needs, and expectations. One is the hazardous waste and groundwater community, where Darcy flow with advection/reaction is normally used. The issues in 2001 wrt CFD here are not related to issues of code usually as the equation are of such low order that they do not cause as many problems. Rather it is the interface between models and non-technical decision-makers that drives discussion of role, utility and expectations of CFD. In this group you have people running codes who have NO training in fluid dynamics, or maybe even calculus! Rather scary...

The second group is a different intellectual culture, based on engineering training, and is involved with multiphase flow and reaction at widely varying Reynolds number- volcanic eruptions, sediment transport, etc. These do not scale well for laboratory experiments so some other means, CFD, is needed to simply get the first-order kinematics and fluxes of the flow. But this requires better resolution of phase-volume-fraction than is standard practice. It also requires some kind of parallel implmentation so that the inherent scales of the *natural* system can be incorporated. Wish-list for 2001:

1) Better LES models for use in multiphase flow

2) Better physical models for inter-phase coupling, especialy in turbulent flow

3) Seamless parallel implementation

4) Better Lagrangian tracking features

5) Higher order forms are defaults for all variables

6) MOST IMPORTANT: Much better documentation about what commercial codes are doing and what things like 'residual', 'convergence', and so on mean and how that changes with choices in grid, time step, etc. I have not seen a commercial product yet (sorry) that really provided the end-user with a concise, error-free, clearly-stated, presentation of the limitations and operating concerns.

This forum has been very useful for me in 2000. Thank you all.

Fred Uckfield January 1, 2001 06:03

Re: Where do we go from here? CFD in 2001
A long way to go, granted, but a goal that must never be lost!

Analogies are good, especially the CAD one. Your insights are good John matched only by your inability to use line breaks.


sam January 1, 2001 12:09

Re: Where do we go from here? CFD in 2001
I can only offer my 2 cents from an auto industry perspective:

Considering the upcoming stringent government and EPA regulations (in US and ROW), the needs are:

1- Faster transient algorithms: CFD algorithms for transient solutions take too long and therefore not production feasible. Need to come up with solutions in hours, not days.. 2- Better sub-models: especially for chemical reactions, combustion and turbulence. Oh by the way, they need to be fast and accurate too (see item 1). 3- CAD import: already mentioned.

Up until these issues are resolved, we will stick with 1D gas dynamics.


John C. Chien January 1, 2001 13:51

Re: Where do we go from here? CFD in 2001
(1). Your comments are really very important. (2). In 80's, super-computers were used throughout the industries to handle the speed issue. The hardware cost was very high, and the cost to run was also very high, about US$1.00/second. (3). In 90's, the general commercial CFD codes , in general, need one megaRAM/1000 cells. It takes a lot of memory to run a commercial code, and in general requires at least overnight to get a converged solution. In many cases, it took several days to get a good solution, based on my 3-D experience. (4). I think, in routine design and trade study, the tool is still 1-D analysis, not just in auto industries. I have been thinking about this issue for a long time. And I think, it requires a different approach. (5). By the way, even performance codes are taking hours to run. For CFD to have direct impact on the day-to-day design activities, the approach needs to be changed to improve the speed.

John C. Chien January 1, 2001 14:32

Re: Where do we go from here? CFD in 2001
(1). You are right. I think, it comes from the habit I developed in the last couple of years when helping a very very old company (over one hundred years old). (2). I had to use a very slow Sun workstation as a terminal to run the 3-D CFD code on other machines. Most of the time, I was worrying about the hard disk memory and I had to delete or compress the files every week, otherwise the machine would overflow and stop. (3). Now you see the real picture of industries, they are the opposite extreme of defense industries of 80's. They are rather poorly equipped to do CFD.(for a very long time, the company could not even afford UPS to keep the computer running. So, it wiped out long CFD jobs when the power was out.) (4). Since this forum is free (sort of), I think, it is important for me to conserve the resources. It shouldn't be difficult to write a program to look for (xx) and rearrange the text in a way you like. If you have worked in my enviroment and also in the computer graphics field, then I think, you will appreciate the need to use "compressed file". (5). By the way, my comment is really not on the "compressed file", but on the user's enviroment in industries. In general, the industry is not equipped with machines to run commercial CFD codes which generally require a huge amount of RAM to run and hard disk space to save the output files. (I normally print the colorful results and delete the picture files and output files. A 3-D output file can easily become a couple of hundred megabytes, and a 3-D picture with some particle path tracing can easily exceed 10 megabytes. And a job with two dozen pictures will occupy a couple of hundred megabytes. And if you run a series of cases, you can easily overflow the machine. Now, if you are from the vendor's side, the situation is probably different. ) (6). The message is: the users side is trying to save money (by using CFD), they are not trying to spend more money to support CFD.

Greg Perkins January 1, 2001 23:14

Re: Where do we go from here? CFD in 2001
The whole purpose of CFD in engineering is as an effective design tool. So John's comments on accuracy, transparency etc. are very important. If its wrong - its not going to be very effective.

Moving along from that, I think that there are several areas which when addressed can help improve it as a tool and therefore drive growth in its use and application. These are:

1. multiphase modelling - as John pointed out early CFD was mainly about flows over wings and other aerospace components - simple geometry, simple physics (ok simpler relatively speaking!). Now use in chemical industries, biomedical, environment, and energy require much better multiphase modelling capabilities - this requires fundamental research and then application in codes

2. systems integration - we often talk about CFD as a stand-alone tool - its not - I think its just one tool - with strengths, limitations (mentioned above). For large scale design of any system - say a car, plane, aluminium smelter, power station, oil field etc., CFD currently is used to model details at a component level - but we need to consider how this design feeds back into the system design - which is usually the important one - we want to optimise the system - not necessarily one component. So we need to have better integration of models - forming a heirarchy of models in fact - since you can't develop a single CFD model for the whole system and solve it anytime soon. Linking all this up with material databases, cost estimates etc. seems to me to be one area for major improvement. How much time is spent - stuffing around with file conversions, manual inputs etc. etc. - all potentially leading to errors! Here commerical CFD codes seem to me to be fairly primitive - although I guess companies often develop their own systems and integrate the codes into them.

3. optimisation - once we have the tools for 2, we can then look at optimisation etc. There's been work on this over the past decade - but again both integration and speed appear to be limiting (assuming your model is accurate).

So by the 2010 this is what I want:

a) commerically available CFD code with validated multiphase models for all types of chemical reaction, fluid flow etc. etc.

b) seemless integration with other codes - eg solid mechanics etc. etc.

c) seemless integration with some-sort of automated feedback mechanism within a heirarchy of models - eg. networked transient plant model (for say an entire oil refinery) with CFD models for major components (reactors, heat exchangers etc) and lumped models or correlations for off-the-shelf componets. I imagine the automated feedback requires substantial data reduction/expansion at the CFD/simple model conceptual interface - how to handle??

d) a much bigger computer to run it on !! - or perhaps a code that runs over the internet and has is a free-ware screensaver frontend - so that it computes MY problem on your computer whenever you're not there - and maybe charges the use back to my credit card - assuming I've got any credit!

Regards Greg Perkins

Lars Ola Liavåg January 2, 2001 05:49

Re: Where do we go from here? CFD in 2001
Although I've seen my share of FORTRAN lines, I have spent the main part of my career trying to put (commercial) CFD codes into use for the benefit of my employee's technology development. As such, I suppose I'm one of the engineers mentioned briefly in a few of the previous postings. Other people may fear for the quality (or whatever it is they are afraid of) of CFD in the future. My opinion is anyway that CFD will not develop along with other technologies unless it is taken into widespread use, and it won't be taken into use on a broad front if a PhD is required to produce results.

For sure, like all of us, I've produced some bogus results in my days (and more are to come) due to lack of time, resources, experience, and knowledge. However, I know for sure that this work has also been of use to the company (it shows no sign of dying yet). One reason is that the results shed light on areas that were previously entirely in the dark. If spread-sheet calculations or 1D modeling cannot do the job, and if the experimental resources aren't available, affordable, or developed, ANY CFD results might make a difference for the better. In fact, for most applications I've tested, the emphasis has been on qualitative result evaluation, trend-seeking and acquiring basic knowledge about the processes. For such purposes, even a solution that isn't grid indipendent can be of much use. In my experience, you can ALWAYS learn something from your simulation results.

Another reason why poor results have sometimes turned out to bee good results, is that they have elliminated end- and fruitless discussions and enabled decisions to be made and the work to carry on. Such fights are all too often driven by professional opinions formed on the basis of more or less qualified guesswork, belief, gut feeling, in short, what we in deep admiration frequently call experience. Different people have different experiences, and unless these experiences can cooperate around some common perception of the reality, they tend to pull in different directions sooner or later. THIS can for sure destroy a company.

I feel a strong urge to remind a few people that the choice is not always between poor results and good results. It may also be between no results at all (complete dark) and some results (dawn, where we can see enough to get an idea about where to go from here). My whish for the continued development of CFD is for it to become widely used, making it an indispensible tool and a competition edge for technology companies of all kinds and sizes. This, I believe, would eventually boost the development of CFD methods and softwares (commercial as well as others) more than any other single factor.

A happy new year to you all,

Lars Ola

Fred Uckfield January 2, 2001 07:21

Re: Where do we go from here? CFD in 2001
"All models are wrong, some are useful"


A very good and appropriate quote!

The interesting question is how wrong/innacurate can the model be and still be useful. Now THAT is something that should be researched in the future.


John C. Chien January 2, 2001 12:56

Re: Where do we go from here? CFD in 2001
(1). There is probably a missing part to the short phrase. (2). For CFD to be widely used, it must be included in the product design loop. (3). With that in mind, the statement becomes " All models are wrong, some are useful, but none is acceptable." (4). In today's enviroment, there exist many commercial CFD codes which provide many models. And no one can say that they are 100% correct.(bug free) (5). It is common for a user company to have a couple of commercial CFD codes on their computer systems. If one has trouble in getting converged solution with one code, it is likely that he is going to try the other program. If he is lucky, he will get a solution. So, we can say that some codes are useful in getting converged solution for someone's problem. (6). But, in the design review, there will be test data presented by engineer-A, 1-D solution presented by engineer-B, 3-D CFD solution presented by engineer-C. It becomes an issue of "which is acceptable?", if they don't match. (7). Is a couple of hundred tire failure out of a million tires used acceptable? The answer is apparently no. But if it is a secret project, it might be acceptable. (especially, during the war.) (8). So, whether the CFD will be widely used or not,I think, depends on the user's community or society. If the user's society is such that it encourages new technology through trial-and-error and accept failre, then CFD will be accepted by today's standard. The driving force can be a company policy, or a law. Someone has to make a decision there. (9). Whether CFD will be widely used in defense contractors or not, depends on the decision from the "president". And whether CFD will be widely used in a company or not, depends on the decision from the company "CEO". And whether students in schools should receive CFD training or not, depends on the job market, which in turn depends on the "president" and "company CEO". (10). In a war, neraly everything will be acceptable. When a war ends, the problem begins. (a war is not defined in the conventional sense,as a fight between two persons. It is between a person and his goal.) (11). In other words, CFD results can only be improved in a war enviroment. So, in the next round of design review or battle, a better CFD result can be presented. A limited war for CFD result to be perfected first?

ken elms January 2, 2001 13:36

Re: Where do we go from here? CFD in 2001
This is a stimulating debate which ought to reach out to those companies and even research workers to reassure them that CFD will continue to have a big impact on product design and development.

The overall impression from reading through the question/reply thread is that CFD has more than helped out to a better understanding of flow model situations and behaviour.

However it needs harnessing and care because it still is a sophisticated technology. It still gives best at time solutions that prove effective with positive benefits to the overall design process scheme.

Maybe it needs like many other factors entering a real gain from evaluation-more dollars and confidence from investors. CFD has come a long way and will continue to have its rightful place as a technology in its own right.Those who thought it would totally replace actual testing of models etc have themselves to blame for expecting too often the impossible.

What else is there to replace it?

John C. Chien January 3, 2001 14:31

Re: Where do we go from here? CFD in 2001
(1). The wind tunnel testing has been around for a long time. Can we use the wind tunnel test data to design an aircraft and jet engine without the final flight test and validation? The answer is NO. Because the very dangerous flight test is the job of very brave flight test pilots. And this has to be done on every new aircraft. (2). Coming back to CFD, it becomes obvious that to have impact on the product design by CFD, CFD validation must be carried out for each calculation. (3). Would you like to be the test pilot of a product designed solely on CFD results? That's challenge. If one is talking about the CFD research and paper writing, then it is a different story. (4). So, one of the most important aspect of CFD activities in the future is: every CFD calculation has to be validated. So far, running a CFD code is much easier than taking a driver's license test...very, very funny indeed.

Greg Perkins January 3, 2001 21:57

Re: Where do we go from here? CFD in 2001
I'm not convinced that you NEED or even would WANT to validate EACH CFD "calculation". Anyhow, how do you validate your results for a potential design you may never build????

As previously mentioned, qualitative results are often very useful - of course we are striving for quantitative results.

I think that a reasonable process methodology would go like this:

1) Develop model - and validate it over range of conditions, geometries etc. of interest for the design. The more fundamental the model the better since this will give you a better confidence envelope for extrapolation of the model to different conditions.

2) Use model for design - by exploring designs similar to those for which the model was originally validated. Use of the 'model' in this situation, leads to some uncertainty in the results - since you can never be sure that the model is capable for this design - until you build a prototype, which is the next stage . . .

3) Built prototype and confirm or otherwise your design predictions - no model will ever be perfect - but we strive for adequacy for the task. If need be repeat process.

The skills then in this process, come from applying engineering judgement to a) include dominate physical mechanisms in model b) design appropriate validation cases to test model phsyics and represent likely design conditions c) 'judging'/'determining' the envelope over which you can safely extrapolate the model to explore different design conditions - eg. if your model asumes laminar heat transfer co-efficient and you run a turbulent design option - the results will be outside the design envelope for which the model is capable. And you made a very bad engineering judgement.

In many cases c) might be most difficult - and that's where we need to think about additional methodologies to determine acceptable risks etc. Here results of sensitivity analysis, the type of design safety required etc. could be incorporated.

Greg Perkins

All times are GMT -4. The time now is 14:46.