You are quite correct to highlight the differences between solid/liquid and gas- gradual displacement/volume changes rather than rapid complex physical and chemical/heat transfer changes dealt with in CFD.

In the FE viewpoint I was drawing attention to new approaches compared to historical known areas of ignorance where there was no back up data or true understanding to incorporate easily into a design.

In a strictly scientific way CFD can be adapted to cover problems of all types given that we understand as fully as possible the problem physics at all times.

How far forward are we in direct data feedback from a test bed scenario to automatic code changes for simulation online realtime.

In words can we learn directly from the test bed and produce a realtime parallel simulation - algorithmn and matching CFD Solution.

Ok, since I seem to be one of the few CFD students here, I will share my thoughts. I'd love comments on the classes you feel I'm missing or ones you think are a waste of time.

Background: I feel mechanical eng., chemical eng., aero. eng. are all good starting points (there may be more).

Necessary classes: Thermodynamics, Transport phenomenon, Conduction, Convection, Numerical analysis, Partial differential eqns, and Computer programming.

Helpful classes: Kinetics, CAD, Computer graphics, Technical communications, Aerodynamics, and Viscous flow.

These can be completed in most universities with a BS degree. However, to gain the skills to solve in depth problems, I'd rather see these people with an MS degree.

John thanks for a great question. Since we are engineers we probably need to approach your question from an engineering point of view. This entails two things the technical aspect and the commercial aspect. Some may disagree that the commercial aspect is relevant but I'd argue that it any engineering endeavour that in the long run (whatever that is)isn't commercially viable is doomed to failure. Just look on those who've tried to design and sell flying cars: technically sound idea but commercial disaster.

On the commercial side CFD must be recognised for what it is: it is a design tool nothing more. Ithink the fact that many CFDers spend a lot of time in school where the commercial aspect of engineering is ignored leads to a lack of consideration of the commercial aspects of CFD use. Firstly let me sayt that CFD is obviously a commercially useful tool. In almost any mechanical device there is some need to perform fluid analysis. This also applies to chemical processes, civil engineering project and environmental and biological sciences which are having more effect on our lives every day. So the market for CFD is there.

Having established that the market is there how do we popularize CFD or how do we get CFD to the point where it is as widely (or nearly as widely) used as say FEM for structural analysis? I think the answer to that is many fold. We need to have engineers who are cognizant of the usefulness of CFD and are skilled in its use. Fundamentally this requires us to have young engineers properly trained in plain fluid mechanics ignoring completely the C part first. Secondly we need to have commercial software that can solve the variety of problems in different industries where CFD can be used. Commercial software is necessary because if someone can't make money selling CFD codes then we won't have CFD codes to use. Wide applicabilty doesn't mean that the codes have to be huge one-code-does all types that we often see today because in many cases the codes are Jack's of trades and masters of none. Different industries require different and often very specific functionalities that would only be useless cost to another industry. Also in some cases a given industry may only need a very simple code, however these are often not available. How many commercial panel codes can you name off the top of your head? Or how many commercial codes offer simple boundary layer analysis?

On the technical side we can again use the commercial structural FEm codes as a model. I know FEM is simple compared to CFD but still we can learn from them. Using CFD is a lot more than just the solver the FEM guys realized this a long time ago. Pre and post processing are the most important steps in a CFD analysis because (1) preprocessing is where you the engineer can affect the solution and (2) postprocessing is where you determine if the answer make sense, and extract the information that is off use to you. If I have a choice of two CFD codes that give me similar results but one takes 2 months to preprocess and the other takes two days I'll take the one that takes two days every time. I've often heard people say "my code runs faster than yours and uses less memory" but in todays world where memory and computational power are relatively cheap compared to the cost of an engineers time such a statement is of minimal importance if the first code requires more engineer time than the second. In post-processing engineers need to extract the data of interest. If management gives me a target lift to drag ratio of 15 and I get 15.1 I could probably care less what the detailed flowfield looks like because I know it's probably pretty damn good. If my commercial code requires me to write whole bunch of scripts to extract the lift and drag I'd probably can it in favour of the code where it's two mouse clicks away (all things being similar). Similarly a turbine designer wants to know where on his map is his stage running right now to that he needs to extract the pressure ratio, corrected flow, stage work and efficiency in quick order - preferably automatically.

In addition to the mechanics of the running the code engineers also need to be able to optimise their designs, account for variation and interface their design inputs/outputs with colleagues in other disciplines. I have a buddy who was doing structural analysis on a turbine blade and it took him half a day to map tempaeratures onto his structural model! Not only that but the process was error prone due to the number of manual steps required and this is a routine analysis for a structural engineer. The FEM code vendor needs to automate this process. Similarly if i'm a wing designer I need to be able to take my aeroloads and put them on my structural model quickly and reliably or better yet solve the coupled problem.

Last but not least from the nitty-gritty solver point of view we need to both develop and disseminate new computational schemes: everything from turbulence models, to linear equation solvers to parallel algorithms, to flux techniques etc. I think the dissemination is as important as the development becasue it's all well and good for some guy in a university to say "I've vome up with the most fabulous flux scheme that captures shocks in teo grid points, has minimum artificial dispation, shows no oscillation and runs really fast" but if the commercial vendors can't put into the code that I use because the professor hasn't published any readable papers on the subject or the code vendor can't be bother to put it in his new code because he was working on a bunch of wizards or something then it won't benfit me any.

In general we need codes that have sufficient attention placed on usuablity and technical sophistication and right now in the commercial world there are few.

Well the assumption here is that it all works and therefore that the buyers will be happy with the product.

The potential risk is that you sell a big lemon and ruin your chances for the future. Then this is worse than if you had a planned, staged and simpler product which could do something WELL. Rather than a product for all, but capable for nothing (or little).

If it works, ok. Of course things are never quite so black and white in reality, but I think you get the point. The code vendors push cfd codes - and I think they have a really good product. But even just yesterday the internal support guys sent me back my question, commenting - "I'd like to know how to do that too". And its "their" code!! So you can be disappointed.

Obviously there's been lots of progress in this direction with regard the codes. But wil still need rigour in the training and science for us (engineers).

Greg

Clifford I think your points are excellent.

I think to some extent the current codes try to address all industries and provide general tools. By doing this some of the pre, post and solver technologies are then a compromise between what is required for 'your' problem and what is required for 'everyone' else's.

Its a bit like Micorsoft Word - nobody uses more than 20% of the features, but apparently, everybody uses a different 20%.

Thus in general codes you need to add your own customisation - if the vendor can provide tech support on how to do it.

I don't see this changing quickly, though a suggestion would be to somehow split up the task and allow different vendors to develop the required bits for different problems/industries. By this, we need standards, and then using these standards we could perhaps have an object framework for coupling the different parts together. I realise this is a mammoth job and not something that's going to happen quickly. But it might in time. We have already seen a standard for CFD data storage and interchange (CGNS?) which is available in some codes and perhaps more in the future. I see no reason why you can't extend this principal to include most parts of the complete package - pre, post and solver. But what I really mean by these components, are much smaller components - ie say a turbulence model, or a model for electromagnetism, or even a database of heat and mass transfer correlations. These could all be assembled somehow. Perhaps a simple metaphor is the (1) AVS program or flowsheet modelling program like (2) HYSYS where you link up visually different components to model a process (in this case a 1. post-processing process and 2. chemical plant process).

Can we do CFD by linking the required parts together?? General cfd allows us to do this a bit, by pushing buttons, but its not very flexible, really.

Greg

In truth Greg some of the difficulty the commercial vendors face is that they are small. Sometimes we forget that. A company can have a fancy web site and advertise everywhere and give you the impression that they are a large company but they may only have 40 engineers and maybe 5 to 10 a writing code. I believe in the future we'll see some mergers and acquisitions in the CFD business that hopefully will give some of these companies the size they need.

(1). Super flexible means that it can be changed into many other forms. (2). And there is a possibility to turn that into infinite number of configurations. (3). The easy to use part simply says that the user does not need a handbook all the time. He can learn some basic operations and then he can figure out intuitively the next operation in using the program. (4). So, super flexible and easy to use is something achievable. (5). The reason why you said that "We cann't build a super flexible code to solve all the problem" is because you are thinking about "understanding the physics of turbulence". I think, if one is still trying to understand the physics, then apparently, he is not going to write down the equations and the code. So, we will have to be realistic and practical, when developing this super flexible and easy to use code. (6). Obviously, we are not trying to solve "all the problems in the whole world". With that understanding, I think, the super flexible and easy to use code is not just a dream. (7). For example, if you are ask to add two numbers together (9876543+1234567), do you have to learn the math for these two number first? No, you don't. You apply only the rule of (1+1=2). And right away, you can add any two number together. (8). Perhaps, you are thinking about solving the physics of the problem instead of solving a problem with known physics. And I think, this is always the limit of a code. A code can only repeat the same process, it can not invent the new physics. That part must be done in the research department.

(1). I think, this is being done all the time at different levels. (2). In any design, we are basically dealing with changes in configuration. That is: which configuration is better. (3). In most cases, this is done step-by-step. In this way, it is easier to control the project. You assume a configuration, predict the result, create the data base, study the implication or meaning, then make a conclusion for the next change in configuration. (4). This sequence could be programmed into a code easily. In this way, a better configuration will be derived. (automatically) (5). Sometimes, there is a problem with this approach, because the person who has the power over the project would like to know why, and also he like to give direction to the future development. So, fully automatic code is not suitable for an organization. (6). If a person can make a better design, then he sure can write down his way of thinking step-by-step into a code and make it automatic. The problem is more on the implementation side, rather than on the technical side. After all, on the other hand, if you don't know how to get a better design (I mean the logic), then it is almost impossible to write a code. (7). Some people are better thinker, so, the codes they derived will be different in flexibility and in user-friendliness.

(1). I think, CFD is " Numerical Analysis and Mathematical modelling in Fluid Dynamics". So, numerical analysis is basic requirement. (2). Mathematical modelling means that the physics however complex or complicated, must be put into a well defined mathematical model form, whether it is the geometry of the object, model of turbulence, turbulent reacting flows or multi-phase problems. (3). I think, the poor performance in CFD simply means the poor understanding of the numerical analysis, and the poor capability in the modeling of physics. (4). And if the flow is laminar, one can eliminate the turbulence modeling. In this case, it can be simplified to just numerical analysis. (5). Whether the analysis should be done only in schools or not, it really depends on the environment. In some places, the research is mainly done in universities, and little in industries. But in other places, at lot of higher level research are carried in industries. (6). So, I think, the school should at least learn, or teach numerical analysis. As for the physical modelling (or mathematical modelling), it will depends on the application itself. This is because, for some problems, the industrial experience is very important to understand the physics behind so that he can create a proper model.(in mathematical form for use in the code)

(1). At the BS level, you really don't have the time to think indepently in terms of the problem solving and also to carry out the necessary steps to obtain the solution. (2). Let's say the you have a simple journal bearing problem to solve. This problem is fairly common in real world application. You see it everywhere, from your car to large machines. Given the training of a BS, can one write down the correct governing equations, the necessary coundary conditions for the square cavity flow problem? (3). If you pass that step, then the next one is to turn it into algebraic form so that you can solve it. This requires the knowledge of the partial differential equations and the numerical analysis. (4). If you pass this step again, you are facing two problems, one is the mesh generation and the other is the code development. Even though, the mesh generation using Cartesian mesh is easier for the cavity flow problem, it still necessary to do stretching near the wall. You could use unstructured mesh though, but the mesh generation would be more complex. (5). Assuming that you know how to write the code, and obtain the converged solution, you will be facing the problem of reading the results. This could be in digital form, line plots or contours. At this point, you are comleting the loop and connecting the results back to the problem. The problem in this case is the journal bearing problem. And I think, the basic knowledge about the operation of a journal beraing is required before one can make use of the CFD results. (6). This process is likely to take one way beyond the BS degree, I think. Someone may be thinking that a commercial code should be able to solve all of these problems, and there is no need to go through all of these troubles. Well, that could be another approach in the future. But still, the code is lifeless, the code does not provide warranty, only the user can think and make decision. The degree of involvment could reduce the time, but not the responsibility. So, training is essential.

(1). You can use a small digital calculator anywhere to perform algebraic calculations. So, you should be able to do the same with a super flexible and easy to use CFD code. (2). You can drive a car in one city, you can also use the car to travel to other city. (3). A car and a calculator perform the operations through mechanical devices and computer software. So,likewise, one can develop a super flexible and easy to use code to perform various tasks. Is a car universal in terms of its ability to travel to different parts of a country? (4). It wasn't possible a couple of hundred years ago. But it is practical today, even with a lot of computer controls in it. (5). The important thing to remember is: we are not trying to emulate the god. After all, a car is a car in the last one hundred years. The basic form does not change much. This should be applicable to our super flexible and easy to use code development. (6). After all, we are not expecting a car to be a ship or an airplane. This could be a very important guideline.

(1). Well, that's a good thinking. (2). But that is only you are sure that there are five or ten more customers out there willing to come back for that price. (3). If I put a five dollar watch in a large box and seal it, then put a picture of an expensive Swiss watch picture on the box, would you be willing to buy it for one thousand dollars? (you will be able to see the time from the outside of the box, but you can't touch the watch. not even when the battery is dead) (4). If you are looking at the larger return in the future, it is probably make sense to give away the super flexible and easy to use code at the begining, like the internet code, internet connection, e-mail, etc. I don't think "free product" is a good idea. It simply means that for now, it is being paid by someone not you. And if "free product" can promote bigger sale of other products, then it is an excellant idea. What do you think?

(1). If you are asking the cfd code to be flexible, then it is hopeless. That is because, a code is just the description of your thinking process to solve the fluid dynamics problems. (2). So, you are asking yourself whether this kind of process is possible or not. (3). The answer is very simple. If you think you can solve the problem, then it is possible to translate the process into a code. (4). The hard part is, in most cases, people do not have "experience in their data bank-in brain" to make the correct conclusion. (5). Without such experience stored in the brain, it is unkely to translate anything into a super flexible and easy to use code. So, car was invented, airplane was invented, light bulb was invented, etc... (6). At some point, something will be invented which is usually considered impossible before. The need is to think something impossible first.

John, I'm not quite sure what you're getting at.

I was thinking, (mainly academically) that perhaps one could build a cfd system in which one connected a set of well defined sub-models and sub-modules together. This is in essence what today's commerical cfd codes do, but I was thinking that it might be useful to allow the user much more control over how these cub-somponents were put together - in current codes this is decided by the vendor. And maybe a after-market for add-ins might develop.

Consider an operating system. It provides a back-bone of services etc for applications to use and share. Is it worthwhile to consider a system which provides similar utility for cfd type problems, in which case the 'applications' would be a set of connected numerical modules etc.

There exist some codes, Fastflo, in which the user can actually describe the pde's to solve and the algorithm thru a special language in which one assembles numerical operators together. I was thinking along these lines, with some additional levels of abstractions.

As clifford says - it may be difficult to achieve commerically - but at this stage I'm not too worried about that aspect.

Greg

(1). You first build a house, then another house. then pretty soon, you have a row of houses. So everything looks fine. (2). Then as the population grows, you build another rows of houses , and somehow there is a street. For a row of houses, you don't see a street. But with two rows, the street is created for free. (3).When the population grows bigger, you have a town, which soon becomes a city. And you have cars everywhere. (4). If you drive a car at 15 miles per hours on the streets of this city, is it safe? I think, you will have accidents at every place. and you would say such big city with cars is impossible. (5). It is impossible. Because cars will run into each other at 15 miles per hour in a city. You probably don't agree with me at the moment. But that's fine. (6). If you look at the modern city, you see extra things, on the streets. That is the traffic control lights. It takes extra things to make it work. (7). This is one of the thing I am trying to point out here. Your idea is not working, because there is something missing in your planning at this stage of thinking. So, a very detailed plan must be carried out first before his idea can be translated into reality. It is true for making a code, career, a company or even a nation. It simply says that the failure is the consequence of inadequate planning. (8). A code came out of the vendor's machine will have negative impact on the users if it is not well planned. (9). And since a code can not think by itself, the planning has to be done by human being. This is the reason why running a code alone can not solve the problem. (10). I alway think that the best way to destroy a person, or a company is to ask the person or the company to use your code. In this way, they will have to depend on you. At the same time, you will be very busy asking the same stupid questions over and over againg. In addition to this, you will have time to do thinking, and he is not going to have time to create his new ideas or methods. Then the distance between the race will increase. (11). You see, it is a war. Somewhere along the line, you will have to stop that stupid way of doing thing and let your brain do the thinking for you. The failure usually is determined right at the start, when one decides to take the short cut (as defined by the person at that moment). (12). I am trying to expand the view somewhat, but it is essential to know that once a person accept the slave state, the failure is determined. (13). If you don't have a plan for a working code, then any level of thinking is not foing to give you the needed answer. (14). Sure, you could ask someone else to create the needed components for you, and add your logic or components to it. But, what is your goal? I think, it is the solution to the problem from the code, so that it can be used for your design. In reality, if you don't get the right answer to your problem, the responsibility is yours, not the person who provide the components. (15). The famous failure of the Mars Probe due to unit system conversion simply says that " every piece of the code or machine must be exactly to make it work" (16). And what I am trying to say is:your idea of making a program modular (like building a city) can become a reality, only when you also have a plan to design something like the traffic control system for a city. (17). If you fail to consider the traffic control system in the city planning, the city will failure, not because such city can not be built, but because there are critical parts missing. (18). If the city planner show you that the city without traffic control is working at car speed of two miles per hour, can you say that it will work when people are trying to speed up and get to work in a hurry? (19). In other words, people are taking high risk, but they are suffering and the failure will teach them how to do CFD in a better way. (20). Anything can become a reality, only when you have a complete plan to make it work. (21). In real world, thing will take its own course, even if the initial condition is not right, because the world is finite and to some degree it is self correcting. (22). But this is not the case for cfd programs and cfd solutions. Like the Mars Probe, one single factor can determine whether it will succeed or not. And to make it work, more training and experience definitely will help. (23). It is difficult, because you don't know the answer. Not because there is no answer. And many CAD and graphics oriented programs nowadays can accept plug-in modules, because it is designed to do so using object-oriented concept and programming. (24). I think, your idea is being used in computer graphic world already.

(1). I think, at this point, there are serval issues raised. (2). One is the code should not be free, and it should be priced at $1000. This is one better solution to the current CFD industries. So, we can say that the current commercial CFD codes are over-priced. Because of this, the CFD suffers. A CD costs less than one dollar, so charging anything over $1000. is excessive. I think, it is a good idea to lower the commercial CFD codes. Although the number $1000. is kind of arbitrary. (3). a BS degree is the minimum requirement to get into the CFD field. a MS degree has a potential to get a good solution. But I think, a PhD in CFD would be ideal. I am thinking about the extra time provided by the PhD program to do hands-on exercises. Without that extra time, one will be using the employer's time later on to do the same thing. (4). The development of a super flexible and easy to use cfd code seems to attract some negative comments. I think this is because I have not define the exact specification yet for this code. If we let the common sense to guide us, then apparently we are not talking about creating a car which can also fly like an airplane and sail like a ship. (5). Even in a school, there are different departments. And in the same department, there are usually several different fields. We are not talking about creating a cfd code to solve both the pump problem and the combustor problem at the same time. And this seems to be the weakness of the general commercial code vendors, because you almost have to hire code programmer and developer from every field of a department, and every department of a school. (6). And this seems to be the case with some vendors. It seems to me that as a first step to do better cfd is to define the field first. It is nonsense to talk about cfd for weather prediction and the pump design at the same time, even though it is fun to listen to the voices from different field of applications. This kind of focus is important. And you can see why by looking at the desk top computer and the laptop computer. And I don't think it is a good idea to develop an assembly line to produce desk top computer and the laptop computer at the same time, or even different time. (7). We will be better off in the future to include the field of application in the specification of this super flexible and easy to use code. I think, within this context, the super flexible and easy to use code concept can still apply. Or you probably have different opinions and ideas to say, I hope.

(1). There is really another bigger issue in CFD, that is the code-user interaction. (2). In the case of a digital calculator, the user is not interested in the hardware and software underneath the cover. He is mainly interested in the input and the output. (3). This is also the case for a car, except for those who are involved in racing. No one really is interested in improving the gas milage of the car his is driving, and trying to look underneath the hood. (4). For CFD, this is different. Here, one is mainly interested in the better way of getting more accurate results. Therefore, every bit of the detail is critical to that goal. (5). If you have been reading the forum, you should already have such feeling. For example, no one is really discussing about the issue of changing a few degree of a pump inlet angle and getting x% increase in efficiency. On the other hand, most people are asking questions like, how to get converged solution, which model is better for his problem,... and so forth. (6). So, the statistical data shows that users are interested in every bit of the code, because he is having trouble with it. I think, this issue must be resolved first, otherwise, you will end up like a cfd code slave asking endless questions. I think, it is miserable for a CFD engineer to be traped in that position. (7). So, this idea of a super flexible and easy to use cfd code, must solve this code-user interaction problem first. (8). Unless, of course, you love this type of work. I am looking only from my point of view. So, I could be biased.

John your analysis in this fascinating question indicates that given the right impetus of ease of availabilty at a sensible cost to the user CFD can provide the solutions to a whole range of scenarios.

You have concluded that a PH D backround will prove more helpful than just a B Sc for the more demanding areas of physics flow related problems.

You have also ascertained that a super code could very well emerge provided that the full data of the background work of the associated problem area has been correctly performed and moreover well understood.

What is very difficult to assess is the in house code maker versus a black box containing numerous side packages that is user friendly but likely to cause headaches because it is too general and open to misuses.

I sense that you are leaning toward a CFD much like it is now but with better[significant?] refinements toward better accuracy and reliabilty.

What also is worth reviewing is: 1. what is the biggest advantage of CFD. 2. which type of problems lend themselves to CFD. 3. given the resources what would cfd users like to happen to stimulate its uses. 4. who currently leads and drives CFD -research or industry. 5. what is the biggest problem currently facing CFD. 6. would numerically computed fluid dynamics[NCFD] be more helpful and better name than just computer fluid dynamics[CFD].

(1). I realized that CFD is at its cross-road now. (2). And your comments and questions do make a lot of sense. (3). Since I am running out of my free Internet time, I will have to answer your questions off-line first and copy it back here. (4). I don't have a fixed plan or picture right now. It is like that there will be several possible models for CFD, including its new name.

Look at the market around us, they use the same strategy. Once again, this target needs a good wells to carry it out if we still have some.