(1). Right from the begining, there is always the tempatation to incorporate CFD into the design process. (2). My question is really: when would be the right time to do so? and if CFD does integrated into the design process, is it profitable to do so? (3). I think, it is a very important question to the company managers, the code developers, the cfd engineers and also CEO's. (4). Make the question a little bit simpler: Is the design integration with CFD a good idea? If yes, why? If no, why? If don't know, why?

I have not understood your question since I am not working in the industry yet. Can you please explain it.

(1). For example, when designing a car, should the engineering department include CFD in their design process? (2). For example, when designing an jet engine, should the engineering department use CFD in their design evaluation? (3). For example, when designing a water pump, should the engineering department use CFD in the product refinement? (4). There are several commercial cfd codes available on the market now, and there are also many CFD codes developed throughout the government labs and research labs in schools, how should these codes be used in relation to the product design? Is CFD still a research activity?

John,

My opinion on this issue has always been that for CFD to have the most impact on the design process, it should be used as early as possible in the process. In general, engineering analysis has the most impact in the beginning. After you've started cutting metal, depositing SiN, or whatever your manufacturing process involves, it is often too late to make fundamental changes to the design.

The FEA companies have seen this, and in the last few years they have put significant resources into developing low end tools that interface seemlessly with popular CAD packages. These tools are intended to push FEA right to the design engineer. This may not be the best idea from a technical standpoint, but it at least makes people aware that FEA can be done and should be done.

From what I've seen, CFD works best as a virtual prototyping tool. When CFD is used to identify design changes before tooling has been developed, and other parts of the system finalized, then it saves money and time.

Regards, Alton

John, which industry? Each industry will have its own requirements on the accuracy of the CFD, the time taken to generate the results, the data available to build a model etc.

It's also critical to clarify what design type questions are to be answered by conducting a CFD simulation. In design the driver is often to determine if Design A is 'better' than Design B. If the application is such that there is a real sizeable difference between A and B then the CFD employed to perform the simulation need not be that fancy.

Innaccurate results can often still be useful results!

The above ethic is that which seperates academic CFD and industrial design CFD. But there's money in the latter as we see from the success of tailored CFD codes such as FLOTHERM, CFDesign, FloWorks etc.

Whilst we debate about the ifs/shoulds of using CFD for design I think you'll find that there's been a quiet mass of people out there who've been doing it in a number of different industries for quite a while.

Fred.

(1). Well, I think, it does have the advantage to use CFD early in the product design stage. (2). Most of the time, at this stage, the simplified design codes and performance codes are used to identify the best configuration. (3). So, in a way, it is not a big issue to incorporate CFD at this stage of the game. (4). Having said that, we must be more realistic about the CFD process and the codes used. I mean one can't just grab a cfd code and ask an engineer to get the result. How do you implement the cfd in the early design process? (5). Assuming that "virtual prototyping" is the way to go, how do you make it a reality in a real company's engineering design department? I mean the step-by-step process. If the big boss says that he can't wait, it's taking too long for the 3-D case to converge, then what? Should the project management take this into account right from the begining. (there are time, cost, and accountability issues)

(1). It is really not a bad idea to push for more accurate results. (2). But, in real life, as you have said, inaccurate result can often still be useful result. I would rather use "engineering result" instead of "inaccurate result". (3). I understand what you are trying to say. So, let's say that "inaccurate results" are acceptable and useful, then how do we get cfd into the engineering design department? How can we pass this step, without telling the manager that cfd can produce a better product each time engineers are looking for a better design? (4). Can we the cfd-forum do something, so that managers will be convinced automatically and incorporate cfd in the design process? If we see only questions about using cfd all the time, then no one is going to use cfd in the engineering department. So, there is something missing, right?

Yes John, something is missing. To me that thing is a link between scientists and engineers.

For example I know a person who fully understands the physical processes, CFD equations, etc... (someone with a great science background) who asked me why his CFD solution predicts a velocity of 10^-6 m/s when he knows the true solution should be 0 m/s in that region (other regions were roughly 3 m/s for comparison of scales). An obvious lack of engineering ability there. Any engineer would say 10^-6 is 0 for all purposes.

I've seen other people with great engineering backgrounds try to use CFD but have no idea of the math that their programs are solving. If you don't understand the meaning of the N.S. equations how can anyone possilbly expect to use CFD well?

Then you have the managers (who know neither the necessary math, science, CFD, or engineering) making the decisions...

Until our universitites can make these connections, CFD will be mostly useless.

John,

Your first question about when to use CFD in the design process is actually a very good question. I don't think the answer to this question is well understood by most CFD professionals because they are more focused on CFD and less focused on the entire design process.

CFD can be used at the very start of the design process, but it is my opinion that it is too slow to evaluate the range of ideas and concepts that need to be evaluated when a design is started from scratch. For this stage, lower order methods will usually suffice. Remember, at this stage, you are usually trying to evaluate if a given design will get you in the right "ballpark". CFD can do this, but typically the turnaround time is just too slow. By the time you have a concept analyzed, it has already changed several times -- sometimes due to factors totally unrelated to fluid flow and heat transfer issues.

CFD really shines when it is brought into the second and later stages of a design. This is the point when you have downselected to just a few likely candidates -- all of which may be acceptable. Here, the higher order details given by a CFD code can help you discern differences in design that may not be intuitive. These fine details often are the difference between a good and mediocre product.

Your second question is on the profitability of using CFD. The easiest way to measure the profitability of CFD is by the number of testing procedures that it eliminates. Depending on the industry, a single test can cost tens of thousands to millions of dollars. If you are using CFD correctly, it can easily have orders of magnitude payback by focusing your testing efforts and eliminating tests.

As a final note, I think it is our responsibility as CFD professionals to get the word out about the profitability of CFD. If you have saved the company testing money, keep track of this and make sure the right people find out about it. CFD has a bad reputation as something that takes lots of money and only generates pretty pictures. It is our responsibility to collect the hard data that eliminates this perception.

(1)It depends on what you mean by "design process". (2)Both in designing a single component or process design, integration of CFD with process "simnulator" (say, SINDA, Aspen-Plus etc...) has been there. (3)In some cases, such as micro-TAS and some MEMS design, CFD is the only way to do design due to the scale limitation.

(1). If we take CFD as "numerical analysis and mathematical modeling in fluid mechanics" instead of running some cfd codes, then there will be a lot of thinking to do. (2). I shall call this "parametric thinking", because there will be design parameters involved in any design activities. (3). The design parameter could be geometry parameter. And in most design activities, this is one big area where the product can take different shapes. (4). At this stage, more than one thing can be done in CFD. " Is the geometry going to create difficulty in mesh generation?". "Is the mesh going to create difficulty for the solver?". (5). The integrated design process with CFD, will take into consideration of these issues, because the solution from CFD will impact on the quality of the design. Based on my experience, the large part of the product design is centered around the ability of the design to be produced (machined or ...) easily. If the design is such that it is hard to create mesh from complex geometry, then it is going to be hard to do analysis and also hard to manufacture it easily. (6). If we take CFD as " numerical analysis" to improve earning of the comapny, similar to the " stock market analysis", then it must be done early in the integration process. The key issue is: "stock market analysis" and "cfd numerical analysis" will have to be performed by the analyst. In other words, when to use the codes and how to use the codes, will be part of the job of analyst. (7). If we look at the cfd process from the testing side, the test engineer need to know whether the sub-scale model can be easily machined, whether the model can be instrumented at the location where the measurement is required, whether the model can be fitted into the existing test cell or wind tunnel, etc... If cfd process is going to replace the testing ( I don't think it is the original goal of cfd), then we still have to carry out the same step-by-step thinking. (8). In terms of the time required in the cfd code execution, I think, it is problem dependent. It also depends on what and how the code is used. Based on my experience, there is a huge room here for improvement. (it can be done) In the research world, it is all right to use general purpose codes because the research is not after the quick turn-around solution. He is mainly interested in different ways or better ways to get the solution(or better solution). (9). For this reason, general purpose cfd code is not suitable for the design integration. It is silly to carry the huge over-head load of a general purpose code to do specific design application. (geometry, mesh, algorithm, graphics,etc...all can be optimized for the application) (10). I think, in this way, there is a light at the other end of the tunnel. The key issue is still: people don't understand cfd, and don't know how to use it. If the numerical analysis is going to improve the quality of the product, then it is a good idea to start the cfd process as early as possible. (you can't ask the test engineer to run the wind tunnel test,at the last minute when everything is already designed. ) There is also the human aspect of the integration, I mean between the non-cfd engineers and the cfd engineers. (which could be a more difficult issue to handle)

(1). "design process" is not "process design". (2). "design integration" means the use of cfd in the design of a product (a component of a whole system). (3). In the product design, normally you would use CAD. So, when cfd is integrated in the design process, you would use CAD, CFD and other methods in creating a better or new product. (4). If one use CAD to draw a pump and then run test on it, then he is not using cfd at all. (he still can look at the flow field of the pump later on when the performance is not good. But this would put cfd in non-integrated situation.)

(1). I think, you are right. (2). Basically, we have failure at the university level related to the CFD training. (3). So, in every engineering school, there should be three basic courses: (a). Design (conceptual design, engineering drawing, machine design, process design,etc. where the emphasis is on the creative use of existing knowledge and rules) (b). Solid mechanics(structure and stress analysis, strength of material, vibration), (c). CFD(numerical analysis, mathematical modeling in fluid mechanics, validation through testing). (4). To do the structure and stress analysis, one needs to use computer. To do conceptual design work, one also needs to use computer(CAD). To carry out numerical analysis in fluid mechanics, one has to use computer. So, computer and related programming will become the tool and foundation of the training. (5). Along this line, there must be a course called "engineering management" which will deal with the three aspect of the engineering activities, in an integrated fashion. (not just accounting side of management) In this way, future potential managers will have the knowledge, training and the skill to handle the engineering projects. The manager should know: the CAD capabilities, the features which can improve the efficiency of modeling; the structure and the stress analysis capabilities, the model complexity and limitations; the numerical analysis capability in fluid mechanics, the degree of difficulties involved in various phases of analysis, the modeling and validation aspect of the solution. (6). Like the failure of dot.com business, the source of failure is "inadequate training and preparation". In most cases, failure is designed into the system right from the begining, someting like: "everybody is heading in this direction, all we need to do is someting like getting a few cfd codes on the system and get the engineers trained for a couple of weeks, then we will have the capability of a modern company to create product automatically" (people like automatic mesh generation, I guess.) The failure is basically due to "no hands-on experience at personal level". (7). The old design approach through testing, can be done using trial-and-error approach. But unfortunately, the modern computer-assisted approach, using CAD, FEA, CFD to do product design, requires a different thinking. The old approach of trial-and-error is feasible, because the testing approach is always bounded by the physical boundary. On the other hand, the computer-assisted approach can give you any number including random numbers and even over-flow number. (8). If the third world countries are making progress through cheap labor force, and developed countries are facing economic problems, then it clearly shows that the management of the modern computer-assisted world must be done in a " very precise, and integrated fashion". (9).And the begining of the integration is: the hands-on experience through each components and phases. It is easier to repeat what others have said, but it is going to be a failure if one does not have the hands-on experience in this computerized world. ( it is interesting to see that the hammer and the nail can be linked together easily, you don't need much training to use it. Actually, almost no one is trained to use a hammer. But this is not the case for CAD,FEA,CFD computer-assisted approach) (10). To be a tool, the tool must produce reliable and repeatable results. To use a tool, one must have hands-on experience. And "do something like that" will definitely lead to failure in this computer-assisted world. (running a code is like key-punch girl's job in old days. her job is to repeat what has been done and produce key-punch card which holds a line of Fortran statement. The key-punch girl can't survive, because it represents only a very small step in the process. To survive, the whole system must be able to survive in the first place. And that is integration. )

John ,I agree with your basic analysis and views.They are indeed excellent points and well expressed. They are worthy of inclusion on an earlier forum question which you raised-Better Ways for doing Cfd.Welldone. I`ve been busy on my gardening so missed some good forum material.

(1). I am working in real time, and anything which is off the screen is old. (2). It takes time to find the old messages, so we will have to move on. (the old messages got pushed off the screen.) It does not matter whether it is under the better way or cfd design integration.

(1). One of the silly concept of using CFD is : to develop a code, a user-friendly code for the designer to use. (2). The idea behind this is three folds: (a). The manager can effectively control the researcher in the company who develops the code, and the designers who use the code to do the design. (b). The researcher can work on what he think is important and write the code. (he normally does not want to get his hands wet in the application area) (c). The designer think that with the code, he can design his product in his own way. So, in this way, everybody is happy. (3). The fact is, it takes time to write a code. So, sooner or later, the pressure will be on the researcher to get his code released as soon as possible. And if you have been using a code, you know that there are bugs, missing options, inadequate documentation, etc... all the time. But the researchers will be put in the position to make "promise" to the user of his code. In addition to that, for the security of his position, the researcher will also try to sell the new features of the code which might not be needed by the users. If the project lasts for three years only, and it takes one year for the researcher to write the code, he is going to figure out what to do for the next two years. The answer is normally for him to support the users. (make the code harder to use like a professional. easier-to-use means non-professional.) And this will in a way destroy the original idea of working independently in his programming field outside the applications. (4). Now for the users of the code, what the designer is going to do is to follow the instruction and try to get the answer. The users have no interest in the inner structure of the code, or the specific flags to improve or control the solution. The designer's main interest is to use the code as is. If you have been the user of a code, it should be easy to understand the pressure on you when the solution is not converging and the deadline to produce the design is yesterday. In any event, a smart designer will quitely produce the result and submit the design on time. (5). The manager using the project tool will be very happy by checking the bar chart and present the results to his client. (6). Even in the golden days of the aerospace industry, when aircraft project typically lasted for ten years, it was not working very well using this model for CFD. The reason is very simple: the design process is iterative. This is same as the iterative procedure used in the CFD code. For an iterative solution to converge, it will take a lot of interaction. By going through the researcher->code->designer->product loop a few times simply is not enough to get the design process converge to produce a high quality product. (7). So, it likely to take a dozen of iterations, or even hundreds of iterations to get a good product design. The reason why this was done in the old days in aerospace industries was because the project was typically extended, (with a lot of cost over-run) by the government. And eventually, the total project time extended to fifteen or twenty years from the start of concept. That's long enough to keep everybody happy, because they all have work to do. (8). When moving through the fast-pased commercial product, this model is not going to work. As a result, the division will try to get rid of the research group so that the unpopular research work can be done somewhere in the company. Or, the division will try to get the code from outside the company, free from the university or government laboratories. The new idea is: give me the code, I will pay for it now. I don't want the schedule delay because of the code schedule slip. This is very funny, because from CFD side, it is like "give me the best initial guess, so that all I need to do is to run the code for one iteration". The idea of using a CFD code, without interacting with the original code writer a few dozen times, is simply not a practical idea. The design has to be changed because of the nature of the design, and the cfd solution has to be changed, along with the iteration control, the mesh generation, not to mention the geometry itself. (9). To write a code which will be used in an ever changing design environment, and be able to produce good design is nearly impossible. It is like training a car to take you to work everyday, through a computer program. (10). If the CFD solution must be done iteratively within a code, then the design application using the code also must be done within the same group. This means that researcher and the designer must work on the same design in the same group. So, that a few dozen interactions can be performed to get the design finished, between the researcher and the designer, if you insist on using such organization. (researcher and designer) (11). From the company's point of view, it is the good product which is important to the market, not the code which is developed, not the efficiency of a designer to use this user-friendly code. This leads to the notion that only the integrated approach can produce the high quality product. The old concept of dividing the process into many sub-processes and try to improve efficiency is not going to work in the computer-assisted world. Because, in the computer-assisted world, the information must be exchanged many dozen times to reach a converged state. (unless one can write a code which will pass a lot of information in parallel between the research and the designer through the code. And the only one which can achieve this is through the interaction of human being, instead of through a code.) (12). Another level of integration is to combine the researcher and the designer together, to produce a superman. A good example is Bill Gates. This is why you have company merger. To get people integrated until effective communication can be established to produce good design through iterative process.(you can copy someone's product to eliminate the design iteration loop, and this is frequently done by many people and companies) (13). The distributed computing is not going to work, without efficient network system. The same is true for the CFD integration. The researcher->cfd code->designer+cfd code->product loop is not going to work because it is hard to develop an effective loop for such organization. There, the end result is to compress the linear organization (merger, down sizing) to reduce the communication time or to increase the workload of each phase(hire more engineering aids to run more cases) But this approach is hopeless. The lack of communication will not produce a converged solution, or a good product, either inside a code or a company. (14). Only a highly integrated company can survive in the computer-assisted world. (The linear process of going from CAD->geometry->mesh code->mesh generation->solver code iteration->graphics->modified geometry->new mesh->....will lead to company re-organization, merger and down sizing, by the outside pressure to improve internal communication.)

(1)Process design includes design process (2) CFD has been integrated into designing new products (3) CAD (or E-CAD) can be used to draw conceptual design, CFD then be used to reveal more detailed flow fields (or heat and mass transport), then CAD or E-CAD would be used to modified the original design. They can be used iteratively or can be integrated.

(1). CAD/CAM has been integrated. So, CAD is the definition of the final product. (2). In the design integration of CFD, CFD geometry model must be developed so that it can be used in the numerical analysis. (3). Depending upon the level of complexity of CFD analysis, the CFD geometry must also change to reflect the need. So, there will be several levels of geometry models for several levels of CFD analysis. (4). The use of the conventional CAD thus should be reserved for the final product definition. It is just too slow to use the conventional CAD in the design integration loop with CFD. (5)."CFD has been integrated into designing new products". I think, I kind of agree with you. But, the design integration of CFD I am talking about is the "routine design integration" of product design. (6). In other words, you can go to an engineering company and find people using CAD on the routine basis, but you are unlikely to find people using CFD in the same product design loop. (I know that some companies are trying, but most are not doing.)

John, your lenghty insight may well be applicable to your own industry, rather your own application of CFD, for the required accuracy and solution time. All you are saying though is that the technology is as yet not mature enough to be integrated within your design process.

I'm surprised you assume that a researcher is the one to produce the code and that it is the researcher who must iterate with the user within the loop. If an entire organisation was focussed on providing the appropriate technology and support then many of your arguments would become moot. As to whether there's enough money in proving application specific design focussed CFD technology...... if the appropriate CFD technology is mature and application validated then there's LOTS and LOTS of money to be made, I'm sure!!

I don't think the design requirements of the automotive or areospace industries can be satisfied by current CFD technology, research and qualification of future designs yes; product based design, no.

John, your job is safe (for now), don't worry

Fred.

(1). If we are still talking about the convergence issue of a CFD run, then there is no way that CFD can be used in the product design loop. (2). And even if we make it a project of CFD analysis, there is no guarantee that the solution will converge within so many number of hours. (3). So, before CFD can be used in the product design loop, the issue of solution convergence must be solved first. The issue is rather complicated, because it is also a strong function the the CFD geometry, CFD mesh, and CFD turbulence model, not to mention other physical modeling required. (4). And if we are using the test data to validate the CFD solution, then we might as well forget about the CFD solution and run only testing. (5). CFD is still a research field of "Numerical Analysis and Mathematical Modeling in Fluid Mechanics". If it is carefully validated, then it can provide global and local flow field information which is difficult to obtain using testing approach. Unfortunately, even in this case, most of us are interested in the global performance only.