I have just finished my thesis and started looking for a job. Beside CFD software compagnies (grid generation and flow solver), there are not many opportunities for those who are involved in the development of new numerical methods. Thus, for those who might be more aware of the pros and cons of various numerical schemes(i.e. have a good understanding of the method they use) and some basic knowledge of the physical aspects.

I had some contacts with the industry (small contracts as a consultant), and it appeared to that they were more intersted in engineers who had very good knowledge of the physical aspects of the problem to be solved with cfd. But, not too aware of the tool they use.

From my point of view, CFD is only a tool to solve a physical problem, if you don't know how to use that tool, how can you expect very good results?

I would like some comments on the subject from those who have more experience. Thanks.

Dear Sebastien

I have to say NO. In my opinion only the (C) is the "tool" as you say. The rest, ie. Fluid Dynamics is itself physics.

You may know very well the (C) but if you don't know the (F) and the (D) you will do "bricolage".

Nevertheless, I know some people who, with some elementary concepts about the (C), can achieve outstanding results in Fluid Dynamics.

After theirs works you will see a lot of (C)ommercial FD codes with their models inside.

Good luck in your searching for a good job.

Kike

You are correct in the observation that the commercial world has little interest in people who wish to develop numerical methods. You will also find that very few people inside CFD companies develop numerical methods either. Other considerations carry much more weight in determining which commercial companies do well and which do not. If you want to develop numerical methods then join academia or a large government research instituion and get the tax-payers to fund it. The real world is not going to.

I do not quite follow the next set of points. You seem to be linking the observation that many users of CFD do not understand the models they are using (agreed) with the solution that understanding the nitty-gritty of numerical schemes is some sort of solution (disagree - not particularly important) and that understanding the physics of the problem is less important (disagree - it is almost by definition the most important because it is the "answer" what one usually performs a numerical simulation to achieve. If you can get the answer without performing the numerical simulation then you are well ahead of the game). Perhaps we are assigning different meanings to "numerical schemes".

On the plus side, as a trained intelligent user of CFD there are currently plenty of mainly entry level positions. As a CFD implementor there are also quite a few jobs around at present although one cannot expect this to last too much longer.

From my own experience it seems actually that the companies are most of the time not so interested in people who know the details of all the physics of the Fluid Dynamics (instabilities, etc...). They seems, most of the time, interested in people who have experience running industrial problems on industrial codes. There are few who are looking for a specific 'software' applications (usually for creating mesh/grid , graphics).

If they were really interested in the Physics of Fluids, they would hire physicists in that field, however they hire engineers, who are good at the practical sides, which is exactly what is needed in the Industry.

You might want to try to get something in Applied Math, or Computational Physics at some Labs in the States for example. Or also in some Universities, like for example Boulder, Colorado - they have a whole center for supercomputing and applications, etc...

Cheers, Patrick Godon

I missed my point.

Let say we have to solve an industrial flow problem. This could be done in 3 step:

1) Set up the mathematical model (it takes a very good understanding of the physical aspects (F and D)

2) Choose a numerical scheme that is best suited for the problem (finite differences, finite elements, finite volumes, implicit or explicit time stepping, do I need a very robust scheme or a more precise schemes, what are the the pros and the cons of the various discretization schemes, etc... ) (it takes very good understing of the computational apects (C)

3) Validation. Is the numerical solution in agreement with the real problem (FD). If not, is it caused by the model (FD) or the numerical scheme (C).

Thus, in order to solve the problem, it is manditory to know both the FD and the C. Sometimes, it appeared to me that the C was missing. The model was good, but the numerical method used to solve it (or the way it was used) might not have been suited for the problem.

Any comments.

In my experience the FD is lacking and the C greatly overemphasised. I have yet to come across a single instance in over 20 years of CFD where the wrong choice of numerical method FE/FD/spectral/particle,etc... was fundamental in leading to the "wrong answer". Most cock-ups are caused by running codes with the brain disengaged (whether there is no brain to engage or the interest elsewhere varies).

1) I agree.

2) Not really. There is very rarely a choice (given the short timescales) you use what is available and judge what the answers are worth. You may make recommendations but it is rare to follow them up before the next piece of "real" work arrives. The decision of what software to purchase is rarely made by the people who use it.

3) Validation does not happen much in the real world. Validation, inspecting, checking, improves reliability and sometime accuracy but is overhead (as is education). It is increasingly pushed to the limit and beyond as even a casual inspection of the latest round of major engineering cock-ups in the developed world shows.

Having said that, I am not disagreeing that nearly all predictions could have been improved by superior numerical schemes. It is simply that other factors tend to be more important.

Is CFD a dead end? Concentrating on methods work within a large institution is generally a dead end. The people who rise swiftly are the people who work at rising swiftly. Methods groups are nearly always backwaters. Currently, you can use CFD to enter at ground level and then switch. If you wish to dabble in CFD long term then there really is only academia and research labs.

Often industry does not very good results. They want qualitative results. I haven't had much experience myself but I gather that running commercial codes for industry is a bit like conducting parametric studies. The only thing that seems to matter (and what I guess the industry is looking for) is the ability to do this in smart way. After all, one can not run a CFD code as many times as it takes to cover the whole parametric space (given the expense).

As far as commercial code development is concerned, most companies rely on robust methods that stood the test of time like SIMPLE (and it's many variants) and PISO or FEM based solvers. I doubt that they would risk putting a new scheme which a scheme developer like you may come up with (as great as the scheme may be). Only after a lot of people use your scheme and report its advantages over and over again, they might consider it. On the other hand, commercial code developers seem to willing to put in the most current physics models to sell their products (e.g. LES, subgrid eddy dissipation models for chemistry, RNG, PDF models both assumed and Monte-Carlo). I am not sure how much these physics models help in terms of accuracy (for problems like turbulence and turbulent combustion, sprays etc.) but I suppose not a whole lot.

If you want basic CFD scheme development (as some one already pointed out) you are looking at government organizations or academics. You can also try some small companies that do research using small business grants and consultancy. Some one in this forum can perhaps point you in the right direction if you tell us what sort of CFD work you are interested in (incompressible, transonic, hypersonic, reacting flows, spectral schemes, multigrid, internal CFD or external CFD etc.).

Good luck with your job search. Transition from graduate school to a job can be tough. It makes you take stock of your worth/abilities. Here is a suggestion that I found useful when I graduated. First job is just a job (something that puts you into the real world). It is not something that necessarily leads to a career. You can plan your career more carefully once you land "a" (not "the") job.

(1). I think most people do not understand CFD. (2). CFD is the core of aerospace industires twenty years ago. But the aerospace companies are disappearing quickly. Even Boeing is relocating the headquarter from Seattle, Washington. Not to mention the huge lay off rumoured in the news sometimes ago about the GE company. (3). So, it is the environment which is changing quickly, not the CFD itself. (4). But I guess more wars are coming, and CFD will again play an important role. (5). Sometimes ago, I had suggested that Internet is an alternative, but now, with the dot com problem and stock price falling, it is hard to make any suggestion. (6). I would say that riding bicycle in the park is a good exercise. It will keep you healthy at least. (7). So, I would say that CFD is still a very difficult technical field. Most people still don't understand it. But as the real war comes, it is something you got to have to survive. But for the time being, businessman can make more money by making products in the place with cheaper labor cost. (8). If you don't want to become someone's cheap labor, why don't you become your own boss. Two years and eight months ago, there was no cfd forum, but now it is here to stay.(I hope) So, my suggestion is: be your own boss. And write more often in cfd forum. (I am 100% sure that the whole world is watching, the east, the west and the communist.)

(1). "If we have to solve an industrial flow problem" perhaps is the key issue. And I think, "Is CFD a dead end?" is a serious question. (2). Let me answer your question of "Is CFD a dead end?" in this way first. For example, if Boeing's business is booming, I think it will likely stay in Seattle, Washington for the next 85 years. Boeing is an aerospace giant, so we can say that CFD activities in Boeing is not helping the situation. So, what's wrong with Boeing's CFD activities? (3). Another example is the GE company. GE is also considered an aerospace giant, they produce jet engines for aircrafts. GE is in the process of merging with Honeywell company. And not long time ago, Honeywell acquired AlliedSignal company, which is considered another aerospace giant. Why is GE going through internal reviews right now, with rumours about the massive lay off? What's wrong with their CFD activities? (4). Look at Russia, is the end of cold war helps the average living standard of Russian? (5). Improvement of technology requires human participation. ( not just codes and computers) (6). Advancement of technology requires "competition", whether it is in the form of exam, job interview, product quality, the cold war, or even the real war. (7). Look at the CFD forum, it is a stage where readers can post messages, read comments and answers, and express difference in opinions. It is exactly that environment which keep the activities going. The by product is the "brain exercise" and the new idea or new understanding of the old idea. (8). This fundamental need is over-looked in Boeing's case, GE's case, and the Russia's case.(so is the US case.) (9). The space race created the communication satellites, the weather satellites, space shuttle, and the space stations. Behind that is the right stage to release the energy to create for any party involved to compete. (10). When the stage of Oscar award is no longer there, you are not going to see good movies. The same is true for CFD. When the CFD stage disappears, the company ceases to exist. (11). We know that CFD is the core technology to many important systems and products, so, the creation of a dynamic CFD stage within a company, an organization or a country is the upmost important policy to keep the company going. (12). After reading this, if you still do not understand, then you can join me and ride the five dollar Schwinn bicycle in the beautiful park. For if you don't, even the beautiful park will soon closed, for the lack of users. (13). Since most people do not understand CFD, they tend to chase after the stock market and try to push it to a higher level. But in the end, if you don not have the real technology, it can not support the market, and would naturally fall back.(like a soap bubble) (14). The CFD forum is the last front to guard the disappearing of war. (killing is not war, because after that you don't have enemy to fight against. So, killing become the end of the war. and without the war, you all become cheap labor.) (15). So, the modern war is to setup the right stage for CFD and develop the best technology. So that you can push the lower technology to a place with lower labor cost, to survive naturally. (16). The truth is that a human body is a war machine which is fighting every second in order to survive. The history of human society also reflects that nature, whether it is in the east or in the west. (17). Once you have this war stage, the need to solve the industrial flow problem will be identified. And the three stages of solving the problem will be followed. Without that, to identify the need to solve a problem, simply means that there is a need to identify the responsibility. And to show the willingness to take the responsibility, the company has to cut your job, becuase the company is operating on the "borrowed money" mostly from the stock market. (18). Most modern products and systems depend on the principle of thermo-fluid dynamics. Therefore, CFD remains the core technology of the modern society. (19). By the way, it is not my job to tell you how to implement the CFD technology, in a form of national policy? a department in a company? a code? or a forum. When it comes to your survival, the choice is yours. One thing is sure, that is, without the computer you are not going to see the weather prediction on TV.

I admitt, I loughed when I read this sentence:

"I had some contacts with the industry (small contracts as a consultant), and it appeared to that they were>more intersted in engineers who had very good knowledge of the physical aspects of the problemto be solved with cfd. But, not too aware of the tool they use."

What do you think, industry is looking for ? Industry certainly is looking for people who solve their physical problems. They don't need people, who know all about the mathematics but don't say after some days of work: ok, this configuration works with those losses and the other one with those ones. That's all industry wants !

"From my point of view, CFD is only a tool to solve a physical problem, if you don't know how to use that tool, how can you expect very good results."

If you do not know the fluid-physics, you only can demonstrate velocities and pressures and temperatures at certain points, but you cannot reduce the losses.

(1). Most people do not understand CFD. (2). What is CFD? (3). What is the product of CFD? (4). The need to use CFD come from the fact that (a). it is difficult to design aerospace systems or components by using wind tunnel (or test facility) alone. In other words, it's impossible to design and simulate the operating condition by using test facility. (b). it is not possible to simulate the weather prediction in the laboratory. (c). it is too expensive to build the testing facility. (both due to the time and cost limitation) (5). So, in these cases, you are forced to use the CFD simulation on computer. And these fields of applications are well recognized. (unfortunately, the industries in these areas are not growing, but shrinking.) (6). Outside these fields, it is not essential to use CFD at all. But for the survival of code vendors, they have to invade every other fields possible. This create some serious problems, because CFD experience in these fields is relatively short, and the products in these fields have been improved for a long time using testing approach. So, there is a mismatch here. (7). I have been saying that "CFD is not for everybody". And I am going to say "CFD is not for every field or product." This is very important, because then you can focus on the technology development and be able to see the impact and the product. (8). Use CFD to reduce the drag of a bicycle? NO. Use CFD to design a rocket? Yes. Use CFD to design a PC board? Yes. Use CFD to study the temperature distribution in a passenger car? No. Use CFD to predict the weather? Yes. Use CFD to study the blood flow in a heart? No. (where Yes: essential, No: not required. )

Currently i'm working in a group in company where we use CFD heavily and I'd say that the majority of the guys I work with are CFD users not developers. In the world of engineering design it is the FD and not the C which is most important. So most companies either buy code from vendors, acquire from NASA or universities or few of the big ones develop their own. In the case of companies who develop their own the ratio of users to developers can be a hundred to one because developers can serve many users.

Also having been here for awhile i've come to realise that there is much more to the effectiveness and efficiency of CFD than just the algorithm because in design we make big models that have lots of complicating features so your code needs to have good features in every area: from pre processing through running the code to post processing meaningful results. the total integration of the different aspects of the process makes a more significant difference to your results than just whether your using Jameson's scheme or Roe scheme or whatever. For example the proper use of techniques like multigrid or residual averaging can (as you know) have a huge effect on your convergence and even though these concepts have been around for years they are often not well understood or used properly. When you're running big 3d jobs sometimes the little things can make huge difference to how well your programs run. recently i was in a CFD conference and a developer was talking about a researh code he'd recently acquired and how the program was difficult to restart and how poor the initialisation was.

having said that though there is a huge need for what you talk about ie developing and implementing CFD schemes. more than anything though I think one of the biggest problems in CFD right now is how many good schemes people have developed but that have not been implemented in code. The commercial vendors don't necessarily have any incentive because in most cases their customers are ignorant. It's not just interior schemes either: the most difficult aspect of hard core CFD in my opinion is boundary conditions and for the most part I think that most people have thrown up their hands for external flows it's not too bad because boundaries can be placed far away and the flow through them is often "smooth enough" but for internal flows it's a whole other kettle of fish.

back to my original point in the design environment a deep understanding of CFD is not usually necessary (just as the structural designer need not know the ins and outs of FEM) but the understanding of Fluid dynamics is most important.

(1). I must say that the questions and answers are fairly accurate. In industries or in business, people do try very hard to meet the schedule and try to push things out of the company first. (2). But, you still can return the product in seven days or thirty days, in most cases, unless it is a liquidation sale. In that case, they normally do not accept the personal check. (3). I would like to share with you the technique to handle such environment, that is, you have to steal the time to do research for the next project. (4). If it is done correctly, the company should put aside small portion of their profit in research activities. This can be evenly distributed to every engineer. And you will get, say a couple of hours a month to do research. (5). But this can also be allocated to a dedicated group of research engineers, so that they will spend say 50% of their time to do research. (6). In case there is no official rule about the research hours, then you should use my rule to steal the time and do research for the next project. This is important because it is part of your job to do research. Otherwise, everybody will stay in the stone age foever. I think, I have made the concept very clear. (7). So, you alway need to look at the current project and the future need at the same time. In other words, to be an engineer or a leader, one must have the capability to predict the future need. And the failure of a company in most cases is the failure of the company to predict the future need. (8). If you are hired to deal with only the current problem, then you know that you will be fired when the current problem is solved. So, in most cases, the current problem will not be solved in time. Who is going to solve the problem and got fired? (9). I think, the current job is always the training ground for the next job. And the current approach to solve the current problem should always include the research of the future need and the solution. This should be the implicit definition of an engineer and a leader. (10). So, how you do it, or how a company does it, is really not important, as long as it is included. (like the income tax or sales tax.) (11). The company will die without research activities. This is one of the reason why many companies changed hand in the past.

(1). Your description of the working environment is fairly accurate. (2). I predict that the company your are in will face a lot of problem in the future. (3). One thing for sure is: the current job give you the opportunity to see the difficulties and the problems in CFD applications. As the time goes on, you will discover more and more problems and difficulties. (4). So, the time is running out. (by running existing codes)

Thanks for the asnwers. But, before I go to another subject, I would like to add something.

From my point of view, there might be people on both sides of the fence that are wrong.

First the mathematicians who put to much emphasize on the equations and forget that these equations are there to model something (not just standing there waiting to be solved it the most accurate manner). The final objective is not to solve mathematical equations, but real life problems.

Second, the engineers who think that because a commercial code is big and fancy its answers are always good.

P.S. I've seed people on both sides. My bachelor degree is in mathematics, but for my master and phd I have switched to engineering (people told me that it would be better for employment.....)

(1). There is nothing wrong with math, or engineering degrees. It is a perfect combination for CFD. (2). It is just the "timing" problem. (3). And remember, supercomputers used to be designed for CFD research. (4). But nowadays, people are pretending that they can get converged solution from a network of PC. This is because the timing is not right, people are very poor right now. And I have seen many many large military bases closed in the last ten years, not to mention many large engineering companies and department stores. (5). I think it will start to change soon, but CFD will have to take a different approach.

John:

Please justify your comment regarding "people pretending to get convergerd solutions from a network of PC." Who specifically, are the "people" and what specifically are you referring to? As a user of networked PC's who takes some pains to understand the notion of "convergence" in a responsible manner, your comment comes across as a gratitious swipe. I endure and occasionally chuckle at your sweeping generalizations, but when they move past being dismissive to being pointedly insulting the effectiveness and power of your comments are lost. At least for me.

(1). I am sure that you are not working in industries. So, you are not included. (2). For those who are not in industries, CFD does not have to be justified or verified. So, they are free to do anything they like. (3). At one time, one of my PhD professor in math spent many years to develop method to calculate a constant in a solution, and the constant took several pages to print. That is just one constant. I have to go now. Bye.