John, Do you sleep at nights!!!!!

Cheers Mahesh

It is very wrong to say that commercial CFD quotes do not produce good results. But it always requires user qualified enough to handle these codes. It is this aspect which will force the demands of CFD specialist to rise. In last five years the many new applications areas are opened up for CFD with development of computing power at an affordable level. It is very wrong to say that everything is gloomy for CFD in future. I have a strong conviction it has a very bright future like FEM analysis of structures. Of course the success will be based on commercial packages.

I have heard that CFD can be useful and lots of money can be made from being an expert in this field.

(1). A real good question. (2). Why not keep track of the time when I post a message here. (3). In a war, can you sleep? You think people come here to relax to listen to the classical music? To suvrive, one must be able to feel the movement of the CFD world. (4). Recently, I sleep for three hours in the evening, then go back to sleep again at around three in the morning. My watch's two alarm will sound at six am and six fifteen (it has been this way for a long time. ) My alarm radio will give me one hour new from six am, and I will eventually get up around seven forty five for another day's work. (5). I found my brain work better at night. (6). I try not to get too involved here, so I have the full control of the situation. (6). Thank you for your message.

(1). Let's say the total cost to hire an engineer is 100,000 dollars, and the price of a CFD code license is 10,000 dollars. Then the engineer must be able to find 10 clients ( or one client with 10 licenses) in order to survive. (2). Everyday, 10 engineers in the client companies will be busy solving their problems, the he is going to be very busy supporting them. (3). Do you think that he will be able to provide first class technical support to every user? Do you think he will have time to visit the cfd-online? Do you think he will be able to learn something new in CFD? Do you think he will have the opportunity to fix a bug? Do you think he will have the time to improve the code?

John, I think you are missing a point here. Most of the CFD codes has active support from the University. Development work is concentrated in Universities (it is cheaper to do so). You can higher a Ph.D student for £15-20K in UK. For support engineer, one can always take him off the job occassionally (say 2 weeks) and train him in latest developments. Once the popularity of the CFD is growing (it is happening), it is most likely to be used for repetitive problems which will require little or no support. If we have trained people (qualified) at industries little or no support will be required.

And then what! All quiet in the Western Front! Let the sacrifice of a few CFD soldiers bring a new era in CFD development. Is is not still better!

(1) It is a very important piece of information. (2). What you are saying is, most commercial codes were born in a university by a professor. It is then raised by PhD students, and supported by part-time students? (3). Is it going to stay this way? Does a user of a commercial code understand this? (this is consistent with what I have experienced that sometimes, a support engineer or a user would say "I thought this bug has been fixed for sometimes, but it is still there." (4). I am thinking that there are probably two schools of thought that, one is the software is maintenance free, and the other is the software requires professional support. (5). I think, for a commercial CFD code to be useful, it can only collect fees based on the number of real CFD problems solved, not based on the number of licenses used, if the world is serious about the campus-born CFD codes. But then, I guess one can alway think of a commercial CFD code as a computer game, like tomb raider ?

(1). Well, I think I like your conclusion......but..... (2). But it doesn't solve the original question: Is it worth it?

Hello guys,

i've been off for three days and this big (!!) thread pops up. Pardon me if i pirate John's numbering style but i have many points and i want them to be separated. (1) the original message asked if it is worth it going back to school to learn cfd. right know i'm doing just that. i started here doing nonlinear optimisation not cfd related at all. i took CFD courses and they were interesting. you must know fluid mechanics well (an intermediate/advanced course in fluid mechanics/aerodynamics is essential). learning CFD helps you to appreciate and think about fluid mechanics more deeply than traditional fluid mechanics courses because you ignore the silly restrictions (ideal flow, incompressibility, etc) and learn to understand physics. however i don't think that you need a MS or PhD to understand CFD (remember understanding fluid mechanics does not mean you can understand CFD, i'll get back to this later). i feel that i could have gained the undersatnding of CFD through similar courses in undergrad, but most professors are of the mindset that undergrads don't have the background, maturity, or understanding to take courses in computational methods. i disagree. i took a grad course in finite elements (structures) in my senior year and two courses in CFD since i've been in grad school. i feel i could have taken the cfd courses in undergrad with no problems. (2) i think for the professional engineer the best thing to do would be to go to your local university and take some courses part time to learn CFD. grad school (for CFD) isn't worth it unless you are passionate (as John said) or are interested in a very advanced topic like unsteady CFD in turbomachinery, aeroacoustics, two phase flow or something similar. unfortunately the people hiring CFD engineers (and FEA analysts for that matter) are stuck with the mindset that a master's degree is required to simulate flow in a pipe with a commercial CFD code. what ends up happening is that guys get MS/PhD doing advanced research into all sorts of ultra advanced topics and go out into industry doing undergrad work (sometimes). for Jason who posted this message originally i'd have to say that getting the advanced degree may not be worth it especially if you plan to stay with the same company. if you change jobs and work in a field that can utilise the knowledge you gain in grad school then it may be worth it. also you have to weigh the hidden cost of going to grad school. most schools will give you assistantship/fellowships and if you are an american you can get one of the NSF, DoD, DoE type fellowships, so you probably won't have to pay for school but at the same time you give up your 40-50K a year job for a 10 to 15K a year stipend and go back to living like an undergrad again. also while you are doing this you could be moving up in your jo etc. also there is the uncertainty when you finish of will you get a job that can utilise the skills you've learned. i have schoolmates who have MSs in aerospace going to work as analysts for insurance companies because they are pissed off with engineering. the engineering job market is good and i'm confient of finding a decent job next year when i finish but i'm pretty flexible. if you decide to do CFD you'll probably be focusing on the aeronautical industry which is different. as john said the nmber of companies is contuously dwindling and the market is by definition small (how many airliners are bought evey year-a few hundred even when the market is great). boeing and lockheed are the only two major airframe manufacturers out there. the engine business is the same with GE, Pratt and whitney and Allison being only major players in another small market. so you have to be careful about CFD. if you're an american you can go to school for free and get a job anywhere but for those two years you're in school (for a master's, about 6-7 for a MS+PhD) you'll be losing out and the job you get when you come out is not going to pay superbig dollars (ie not 70K+). (3)i think if you want to learn CFD to the point where you can understand what goes on in this forum and use commercial codes properly you can go to your nearby university and take a few grad courses in CFD/numerical methods, fluid mechanics and read a lot of texts/ papers in their library.

i wanted to say more but i've gone long enough

this thread is great. a good discussion on the philosophy and economics of CFD. i'm continuing my previous message. didn't want to put everything in one message it'd be too long. (4)as for commercial codes you have to take the manufacturer's claims with a grain of salt. Dr. Sarkar made the point the the commercial CFD codes must move to the level of maturity of structural FEA codes. structural FEA is at the point where somebody who's taken an intermediate/advanced course in structural analysis/strucutural dynamics can get on and learn to use these codes with only minimal understanding of the finite element method itself. ie you only need to know the difference bewteen element types and which BC to apply and you're good to go (ie read the instructions). i took a grad course in finite elements in undergrad and from that i was able to teach myself to do analyses on ANSYS with repeatable, design quality results. the same is not true for CFD. the physics and the appropriate numerical scheme are closely linked. so taking a few intermediate/advanced course in fluid mechanics/aerodynamics cannot prepare to use even the simplest (say compressible euler) code properly. in fact the lack of true understanding of CFD is shown by some of the questions asked on this forum. these questions indicate that the users (while they understand fluid mechainics) do not have a good grasp of CFD. (5)even CFD courses in grad school may not expose you fully to the subject. for example the CFD courses i've taken have concentrated mostly on high Reynold's number compressible flow (typical in aircraft design-i'm in an aerospace department) so for the most part we concentrated on time iterative schemes. so when someone asks me about a pressure based scheme eg SIMPLE i have little understanding of that. it's just a matter of what you're doing. i'm sure if i had taken a CFD course more geared toward incompressible flow (electronic cooling, process fluids etc) i'd have similar difficulties in understanding the nuances of time iterative schemes (unless the professor was into preconditioning). (6) in the same way you have to be careful about these "general purpose" commercial codes that use one scheme to solve everything. it'll probably turn out that it'll be great for airfoils and turbomachinery but suck (or be less impressive) at incompressible flow. (7)on the other hand commercial codes often have impressive pre- and post- processing features. but if the solver isn't good it's just pretty pictures. (8)if problems are modelled carefully and the analyst understands fluid mechanics and CFD i think that most commercial CFD codes out there can give good results. i've seen impressive results with commercal codes (Fluent, CFDRC, StarCD etc) on relatively complicated problems but these were obtained by good analysts (i'd hate to see some of the trash that poor analysts come up with). it's the user not the code. (9)commercial codes are not necesarily worse than research codes (i mean the solvers) i've ssen good, bad, and indifferent results with research codes see point 8. often many research codes are old, not suitable for modern computer architectures, or just outdated. NASA is still providing old thin layer Navier Stokes codes (not saying TLNS analyses are useless but you have to be careful) and many of their codes are for vector machines when industry is using parallel machines etc. also many sponsoring agencies are not pushing algorithm/code development anymore because they feel they've spent enough money on it. i think it is up to the commercial companies to push algorithm development since they'll make money off it. but i see many commercial codes still using MacCormack schemes or pressure based schemes (from the 60's and 70's) when so many faster, more accurate and widely applicable schemes have come along since 1981 when Jameson unleashed the 4 stage runge-kutta scheme on the CFD community. (10)as an aside i see that in the future europe may take away the US's dominance in CFD code development because the europeans are still developing schemes and putting them in commercial codes while in america the codes contain old schemes and they're more interested in fancy postprocessing. america dominated teh CFD scene because of algorithm/scheme development AND IMPLEMENTATION and computational dominance. as i said above the first no longer seems to be true (especially implementation - new schemes are being developed in the US but not necessarily being widely implemented) and the computational dominance is also disappearing. now that you can string together a few hundred PCs and get the performance of an Origin or an SP2 at a fraction of the cost you no longer need a big supercomputer industry for CFD dominance. i read a webpage where some fellas at a DoE lab set up a 100 odd processor workstation cluster in a weekend. we have at least two such clusters here (penn state) and both are administered part time (and were built) by aerospace grad students (11)lastly. you have to understand that what may be a poor result in certain cases may be adequate in others. ie there is nothing hard and fast as to what constitutes a good CFD result. In structural analysis near perfection is expected butin CFD that's not the case. Case in point i took a visit to Pratt & Whitney last year. they've basically dismantled much of their structural lab because the FEA results are good enough to replace completely much of the testing they used to do. in CFD it's different. P&W has some of the most impressive CFD capabilities in american industry: they have huge computational resources through workstation clusters and have years of advanced, in house CFD code development behind them, they routinely do multistage unsteady turbine and compressor analyses. yet their unsteady CFD results aren't often right on top of the experimental results (ie their exit flow angle distributions aren't perfect or they don't get the secondary flow just right). but they like what they can do with their codes (they're still improving them of course) because otherwise they'd have to test everything out on rotating rigs or full scale engine tests like they did 20 years ago. when you're designing 100K pound thrust engines in ever shrinking design cycles this isn't possible. as their head of code development told us (a guy named Rhie i think) they have to "Just do it" because they have to be able to use these tools to produce their products in their ultra-competitve market. academics can always say "your computed pressure line doesn't lie right over your experimental points" but companies like Pratt are producing better, more efficient products at lower cost every year. sure they'll have failiures, but failiures are less common now than in the past, and nothing is perfect not even experimental results.

that's all i have to say right know. i have a great passon for CFD and all areas of computation because these are the only way we engineers are going to build better products that more people can use in the future. just remember we'll never be perfect but we should try

John, I always enjoy your non-linear thinking. I agree with everything you said. Your last statement just doesn't follow the previous statements??? Anyway, Japanese cars used to be crap. Everyone knew it. They were also cheap. Don't forget to couple price with quality in your market equation. For awhile, the Big 3 made crap cars that were cheap. They were slower than the Japanese to make reasonably priced, high-quality cars. You're absolutely right: The big three didn't improve quality and lost ground because of THE MARKET.

As for moving production to Communist (read: poor, underemployed) countries. Who cares about the motivation? We don't live in a collection of countries with different workforces. We live in a world with artificially segmented salary regions.

For the sake of argument, let's look at it from your isolationist viewpoint. Let's say a guy in Detroit "works" 40 hours per week. He averages an additional 20 hours overtime. Tightening the same 3 nuts over and over again on a Big 3 assembly line, he's likely to bring home $70k+ per year. GM, for instance, has a couple of choices. 1) Set up a manufacturing plant in, say, Tennessee where labor costs are more reasonable. 2) Set up a manufacturing plant in, say, Juarez Mexico where labor costs are CONSIDERABLY less expensive. 3) Close up shop and stop feeding anyone.

You seem to be saying that choice 3 is unacceptable. I agree. I'm guessing you don't like greedy motivations behind choice 3 You seem to think choice 1 isn't so bad because it's keeping jobs in the US.

Get real: there's is no ethical difference between choice 1 and 2. Sure, it's going to have a immediately painful effect on some. Over time, though, it evens out the "disparity of wealth". How's that for a free market making a nod towards one of the supposed goals of Socialism?

Well, we've obviously gone way off topic. Feel free to reply, but I'm done with this thread. Time to get back to what we all love: CFD!

Jeff W

(1). I have just realized that there is no fun in working on CFD. (2). By talking to a human being on Internet, I could see things more clearly. (3). And this is the first time that I check into the web search for the words " non-linear thinking". I didn't know that thinking can be non-linear at all. I had the impression that thinking is always parallel. So, I was wrong. (4). Yes, you are right about what you have said. So, it is hard to change someone's mind in the first place. I agree with you that it is time to move on to other subjects. Who cares about whether it is worth it or not. He probably has already made up his mind long before asking the question. Good night! (or good morning!)

(1). A real good first-hand report indeed.

(1). I would strongly recommand that this article be studied in great detail. (2). I do have some comments on it. But in view of the situation, I will just mention a couple of problem areas. (3). Currently, the thermal stress problem in turbomachinery hot sections, combustor, turbines is still a real problem. Ability to run a commercial structure code does not mean that these problems can be or has been easily solved. It is not uncommon for the hardware to crack and melt. (4). The computer hardware has been a driving force behind all the modern computer assisted analysis and design. Its speed will reduce the design cycle time. But it can also produce garbage at much faster speed. Garbage-in-and-garbage-out will only overload the system. So, the speed along will not solve the problem. (5). I have been getting relatively reasonable flow separation and secondary flow results in turbomachinery flows. I don't think you need to use super-computer to get good results. But it is essential that one does understand the physics of the flow, the turbulence models, and the mesh. I don't think running more cases, or using parallel computers will solve the problem. (6). I would like to repeat what I said before, " if you know how to solve a problem, then the problem is nearly solved, so there is no real need to actually solve it". ( do you have to check out the calculator for all the mathematical operation? No! because you know it can be done.) (7). P&W is merging its two divisions. I think, in CFD areas, it is important to seek good engineers from outside the company. Like many large organizations, the key problem is always with human beings. This is especially true for CFD. CFD is neither a code nor a computer. Without experienced, well trained CFD engineers, nothing will be done, and nothing will be solved.

thanks for your comments john. this is what i wanted sometimes we don't get good discussion on these topics. in response to your point (3) i would say that these structural problem areas (turbomachine hot sections - combustor, HP turbine) are areas where heat transfer analysis is critical. indeed if you think about it the structural analysis is only as good as the fluid dynamics calculations which produce the thermal boundary conditions. in my experience the ability of cfd codes to produce this thermal information is (woefully) inadequate. one of the reasons (personal opinion) is the difficulty in obatining experimental comparisons to validate cfd heat trans results. i've read many papers where cfd code developers validate there codes using surface pressure distributions alone. but in my opinion this only really shows that the inviscid part of your code is working well (ie you have a good euler code) it doesn't tell you much about whether you are predicting viscous features ( which affect heat trans) well. i agree much work is to be done here but i think it is on the cfd side not the fea side. moreover (as i said before) commercial fea code developers are producing good products whereas the same isn't true (in general)for commercial cfd code developers. so in short i agree with you that many new (and even older) engines have hot section structural problems but i think it is more a deficiency on the part of the fluid analysis than on the structural analysis. as to your points (4) and (7) i think you're reiterating what i said and i agree that the results can only be as good as the analyst. aside from that i chose the example of P&W because i'd been there and so i have been told directly by the people there what their issues are. i agree that a large parallel (or distributed) computer isn't needed to get good results but my understanding was that in their case to obtain results fast enough to fit into their compressed design cycle they needed this kind of performance. also i wanted to stress that unlike a monolithic super computer ( say a big sgi or SP2) this computational power comes at low cost since they're just linking the (sun in their case) workstations that everybody already has on their desks. also i used them as an example because they were a case of a company that is in market where cfd is essential to faster design cycles and better products, and that P&W (along with the other big engine companies) have been using these tools to produce better engineered products (faster). i've also read papers written by people in the airframe market who are not as high on cfd because for them the windtunnel is still more cost effective and reliable than cfd so cfd is not as dominant a tool in their case. it's a matter of perspective and the nature of your business the sircraft guys can get their results quickly and acurately via experiment and so they still rely on it (nothing's wrong with that) but for turbomachinery guys experiments are expensive and time consuming so they'll tolerate the shortcomings of CFD because they have to.

(1). That was exactly the point I was trying to make, the efficiency, the reliability, the life of an engine depends only on one thing, that is CFD. (2). So, an engine company without a strong group of CFD researchers and programs will fail sooner or later. ( I can only guess that when there is a failure in structure, the structure group gets more support to do more work. And no one in aero/thermal would say it's their problem.) (3). About the cost, I can tell you that the current implementation of CFD and the design process in most companies are highly in-efficient. It has become a huge dianosaur. That is a much bigger issue which is going to take a lot of discussions separately. The current form of CFD and CFD codes are highly in-efficient.

Getting a bit off-topic here, but anyway...

Predicting hot-section life in turbomachinery only using CFD/FEA is far off what we can do today. I've spent 4 years doing research on turbine blade-heat transfer simulations and the current state of the art still has problems to predict heat-transfer in a single linear cascade with homogenous turbulence coming in ... trying to predict heat-transfer in a 3D multi-stage environment is at least two or three classes more difficult, and if you also add combustion things become virtually impossible. The FEA side of life prediction is also often very sensitive to small differences in surface temperatures and a small errors in the CFD prediction can thus lead to large differences in life prediction.

The problem is not that there is no experimental data, the problem is that this is just a d*mn difficult physical phenomena which is very tricky to simulate. Current turbulence models and CPU resources are not able to handle it. To design these sections you have to rely a lot on experience, experiments and ad-hoc models (transition models, turbulence models, correlactoins etc) which were tuned for a similar case. As soon as you want to do something which is different from the typical cases you have to rely on experimental testing to validate your design. One of the most common misstakes I see other CFD engineers do is that they trust their heat-transfer results almost as much as they trust their pressure distributions etc.

You seem to blame these inabilities of CFD on the commercial codes. I can't see the logic - it is not their fault that this is a difficult problem. State-of-the art research codes also have difficulties here. Actually, I've run a few of the test-cases that I worked with for my PhD in one of the larger commercial CFD codes and the results are quite acceptable. Of course you don't get the small fixes and tunes that you have in a research code but results seemed consistant and not that much worse than what I got with my own code.

(1). I think, readers are very lucky because we have a webmaster who really understand the degree of difficulty in CFD prediction of skin friction and heat transfer. (2). I remember a few years back, I had a discussion with a support engineer about the inconsistent skin friction prediction using their code, he simply said the pressure field is accurate and many users like to use their code. These codes are being used in the real world environment (both in-house and commercial codes) to provide design information for real products. And in some cases, plastic models are used to obtain the needed heat transfer or aerodynamics information. (3). The point I was trying to make is engineers would use any method available to him especially the problem is very difficult. Then, they will run into G_I_G_O problem. (if you have convinced the company to use your code, then it is difficult for you to say that you can't handle the problem. And even if the company accept you statement, there is no other alternative because the code was suppose to replace the testing. Going back to more testing? )

No.

Get out of engineering. Start your own company that allows direct interaction with the public, e.g. plumbing, TV, AC or PC repair.