I think, we are missing a good bridge between the pioneering theorists (Roe,van Lee, Grodonov ...) and engineers. This bridge should be such theorists who work further on the base of the poineering theorists and touch the complexities of practical problems bravely and then give engineers good and near-practice numerical methods. If every one will be a pioneer, we'll miss this good bridge for ever.

X. Ye

I agree completely. And it seems the gap between the "pioneering theorists" and the engineers keeps growing. Take turbulence modeling as an example. I think that among most of the theorists, second-moment closure has been considered almost passe for a decade now, and new research is looking at more innovative methods. But most folks in industry aren't even using Reynolds stress models because they're "too advanced" for complex engineering problems that tend to get solved with k-eps or the like. There hasn't been a "bridge" between academic model development and engineering application, but maybe some people will start to fill this niche? Maybe they're the ones who will start to turn science into art or art into science?

Keith

(1). I think the basic issue is the lack of systematic testing of the method before it is published or accepted for applications. In a way, both sides are guilty. (2). In the drug development business, it takes years to develop a new drug and it also takes many more years to do testing. Because it could be a life and death issue. (3). The same is true for the building codes. Without it, it can be a very serious problem when earthquakes hit. (4). For CFD, we are begining to see the early symptons of the problem, that is unreliable CFD results from time to time. (5). This is because the systematic testing and optimization of the methods developed,which is necessary to produce reliable CFD answers, are largely missing. (6). One can say that it is part of technology evolution, it going to take time. (7). I would say that the increasing use of CFD technology without systematic testing and optimization is going to make the CFD industries totally un-reliable. Unless one can say that each CFD problem is part of the general case, I would say that each CFD problem is by itself a special case.

(1).Strive for excellence is the major driving force for technology advancement. As a technology gets more advanced and involves in multi-discipline knowledge, people have to strive into different directions in the same subject. (2). In CFD, people working on CFD algorithm development (that was my research area 10 years ago) have different opinon with CFD application engineers (that is now I am doing). (3). CFD developers (not code developer) also have different directions to strive, some try to improve accuracy in terms of mathematic orders, some to improve the efficiency to make it run faster, and some to improve the resolution of capturing discontinuities. With those efforts, people may spend years and millions of dollers only to make the shock waves from 2 grid points to one and half, with a possible cost of 30% more CPU time. (4).CFD users (include code developers) are more interested how to solve engineering problems quickly, accuratly, and easily. (5) Some engineers also developed algorithms. One very famous professor with engineering background developed a scheme for Euler equations for a particular application, he claimed that it was the fastest solver in that area. Indeed his methods was widely used in a industry code. But CFD developers said this method does not mathematically sound and is not considered as a 'scheme'.

Come back to CFD, there are some special difficulties in CFD over CSM. The main difficulties are the advaction terms, which are the source of eddy viscosity, the numerical instability, the needs for numerical dissapation (artificial viscosity), the problem in discontinuity, and other unkown non-linear problems. Those terms are the most concerns for CFD developers. The second difficulties are the boundary conditions. The eqs. are always the same, only by changing B.Cs, we have sooooo many applications. Unlike CSM where there is a well-defined boundary, in fluids everything is so soft what there is virtually no boundary! And even though we have to define a numerical boundary we certainly don't know what exactly happens on the pre-defined boundaries. Those difficulties are the most concerns of CFD code developers. The third but not the least, the viscous effects, boundary layer, wakes, mixing, and their interactions. These factors are usually affected by grid quality and density, a partular dificulty for CFD application engineers.

Most CFD algorithms was developed between later 70's and early 80's. Some high order/high resolution methods (the Godunov-type schemes) were also developed at that time, or a little after, although the concept was dated back to later 50's. Recently, there are no major breakthrough in mathematic models in these area, except for some minor improvements (I don't know if this is the reason for me to change area from development to application). The main efforts are now shifted to 'all-purpose' code development and user-friendly interface, something more like an art. May be this is the time for new ideas which may lead to a brighter future of CFD. The hope is on the new generation who strive for excellence. There is no hope in the 'old folks' (not in terms of age) who only strive for high-salary or high-reputation.

(1). It is a good example to look at the structure analysis side. On the computer aided activities, the structure analysis is closer to the CAD side. By linking the structure analysis directly behind the CAD, it may be possible to achieve the coupled design and analysis loop, thus shorten the design iteration cycle time. In this case, the driving force is the design and the feed back loop is the analysis. (2). And this is probably one of the major difference between the structure and CFD analysis. In CFD analysis, it is the driving force of the design. In other words, the CFD analysis must determine the optimum shape for the CAD design. This is perhaps one of the difficulties involved in implementing the CFD analysis. For example, we don't ask the CAD designer to come up with a blade shape and then run a CFD analysis.(in a way parallel to the structure analysis) The problem here is CAD designer can't come up with the shape of a blade. The blade has to be designed from the CFD side. (3). I am trying to keep the scope of discussion small here. We really have to see the whole process more clearly in order to develop more reliable CFD technologies. In the actual processes, what happens is: First, the simplified 1-D theory is used to derive the shape, Then, that shape is analyzed by more advanced theory. Then a preliminary shape is created by the CAD, and a 3-D advanced analysis is performed on the shape. (4). In this traditional approach, the analysis is finally used to check the shape of design, which is very similar to the wind tunnel testing of many concept models. By testing, it is hoping that the best can be found. It is one way street, from shape to analysis. Ideally, one would like to use the analysis to determine the shape. So, in this case, the CAD should follow the analysis. (5). From here, you can see the basic difference between the structure analysis and the CFD analysis. The current trend of commercial CFD codes, using the CAD as front end to develop the geometry and mesh is not consistent with the traditional blade design practice.(the shape comes from analysis, not from CAD design.)

Research and Development of designs associated with complex continuum systems in fluid dynamics and solid mechanics. Design/Research Objective: The design of industrial processes, systems, or components.... Research Advances to be Applied: ... through the coordinated application of new advances in physical experimentation and computer simulation at all stages of the design process.

Mechanical Engineering Department, University of Utah Sponsored by National Science Foundation

Is it not a good start to bridge the missing gap between design optimization and CFD simulation output!

i disagree with your point (5) the object of CSM and CFD is the same: produce shapes which give good performance. we can only evaluate performance after having done the analysis and the analysis should lead us to new shapes (as you said) but we do need a shape to start the analysis. in structural design we do the same thing as you said for turbine blades (point (3)) you start with a simplified structure (beams, trusses etc) and make preliminary sizing and shape which can be translated in to more 'fleshed out' structure which is analysed at the most precise level. THEN as my statement above we try another shape (ie the analysis leads to new shape). there is not much conceptual difference between the structural and aerodynamic design. the difference is practical: the structural is easier and we can often cut out some of the simplifications because of low analysis cost

on your comments about the TVD scheme (i am not a defender of TVD schemes) i can only say this: many of the concepts we use in CFD are only applicable for systems of linear equations case in point is the von neumann stability analysis that we are fed as young CFDers. so while these concepts are not perfectly applicable for the nonlinear equations we try to solve they do give use guidelines to help our analysis. indeed there are few things we know absolutely accurately about the NS equations (we can't even show they have unique solutions) but our knowledge of the realities of fluid flow (the actual fluid behaviour) and what (semi rigorous) mathematics we can apply is mostly all we have. all i'm saying is that we must be sure we use as much of it as we can else we won't be doing our best or most efficient job.

as Keith said the gap between the developers and the practisers is large. how to bridge it? in this case i think Mohammed (the users) must come to the mountain (academia). this isn't strictly true because i think there are theorists who can and will get their hands dirty. check out J. of Aircraft vol 36 no 1. there is a paper there by Jameson where he outlines the improvement of a commercial aircraft for McDonnell Douglas. the paper is very design oriented and shows how an academician can go into the industrial world and impact the design process significantly. HOWever i doubt seriously that many professors are interested in this (hence my previous statement) after all why become a professor. so it'll be up to the real engineers to go to the ivory tower and bring back the knowledge they need.

(1). I think, we are getting closer to the other core issue of design, or more specifically, the aerodynamic design. It is a much bigger issue. (2). It is not efficient to design the turbine blade or the aircraft wing from the CAD end. (3). In the CAD, the package is general, and every designer can come up with a different way to put together the final goemetry. This is not how the aerodynamic design is carried out. (4). The geometry can take different forms and formats. And the general CAD program can only handle the geometry is a general way, which does not have any aerodynamic significant at all. (5). All I can say is the geometry in turbine blade and aircraft wing takes special relationships with the aerodynamics, and it has very little to do with general CAD. So, the use of geometry in CFD for turbine blade and aircarft wing is very different from that of a general CAD. (6). That special relationship between the blade geometry (or wing geometry) and the aerodynamics is the key issue in aerodynamic design. And CFD is part of this design process. This is the reason why it is very hard to integrate the general CAD into the aerodynamic design. (it is not the right approach.)

(1). There are at least two issues here. One is the integration and the other is the location. (2).In some countries, the research is carried out mainly in universities (they have the money, the PhD level professors, and the facilities). In other countries, the real technology is in the hands of private industries. And in some other countries, the industries work with the universities in the research subjects. So, there are many ways to bridge the gap. (3). In the first case, the CFD technology developed will be more tuned toward student training, or the paper publishing. So, the output of the CFD code( or technology) is linked to the general features of the problem, for training or publication purposes. (4). In the second case, the output of the CFD code will be linked to the efficiency or performance of the product (or design), to impress the potential buyers (private, government, or foreign customers). (5). In the third case, the CFD output will be expressed in terms of the mutual relationship, hopefully a long and good one. (6). The fact is many countries have been working on this issue with different approaches. (7). My personal feeling is: It is the engineer's total responsibility to solve his problem. One can not say that it is someone's geometry problem, mesh generation problem, turbulence modeling problem, algorithm problem, or even the plotting problem. In CFD, one really have to become a superman (or superwoman) to be able to solve the problem. (it can be a group of engineers or a department in a company scale).

(1).The turbulence modelling is the key to the future of CFD. So, it is a very critical issue. (2).If one is using a commercial CFD code which has four or five turbulence models in it, he will be dead if none of the turbulence models listed can provide a good solution. (3). So, it is very important that a CFD engineer must have the skill to do the turbulence modelling for the types of the problems he is interested in, whether using algebraic models or two-equation models. (4). I don't know exactly why the stress models are not being widely used. It could be because most problems do not require high level model to achieve the accuracy. Or it could be because it is hard to understand like a black box. (5). So far the platforms provided between the algebraic Baldwin-Lomax model and the two-equation k-epsilon low Reynolds model is wide enough to cover a wide range of practical problems and at the same time allows the flexibility for modifications. (6). Instead of waiting for someone to invent a new set of coefficients or models, I think, we should train the student to modify the existing models for particular applications. This is the future of CFD. It is not a small market of future CFD. (7). Even though there are other important issues in CFD, such as mesh, resolution, and algorithm, but I think, these issues are mainly related to the hardware limitations,which are disappearing very quickly every year. So, the main issue will be back to the "turbulence modelling". That is a systematic modelling process for a particular problem at hand, not just a selection of existing models or coefficients. We want to solve the problem and the turbulence model must be further "modelled" to achieve the goal. The ability to implement the model implies that the engineer must have the minimum ability to write the code (or modify the existing code). The ability to model the turbulence and solve the problem will be experience related. That is related to specific applications. (8). The 3-D mesh generation will become more and more automatic. It will be of benefit to the automatic mesh code developers, but not to the advantage of the users' experience. (9). In addition to the turbulence modelling issue, the fundamental issue of aerodynamic design (or thermo/fluid design) will require more thinking or smarter approach.

ok i understand you know.

(1).I think, we are heading into the dark age again. It is not because of the Y2K problem which can cause the power disruption, but because we are not focusing on the technical problems. At least from CFD point of view, the same old Baldwin-Lomax model ,and the two-equation models are still being used. They are at lest a quarter of a century old. (2). Running more CFD calculations or computer runs will not improve the quality of the solutions or products. It can only waste more resources. (3). As the computers become more popular and affordable, I wonder why people are not taking advantage of this opportunity to bring new ideas to the CFD solutions.

(1). This is true, if the world is steady-state. (2). In reality, the world is not a steady-state. (3). This can happen in stokes market, in Asian economy, American defense spending, new policy of the newly merged large company, reorganization, cut-back to improve the financial performance, ...etc... (4). There were successful examples in other country, where a school was created by the company interested in training the needed engineers. In that case, the company remain stable and the relationship between the school and the company has been good. (5). This is not the case in America, where the change in the market can easily affect the survival of the comapny, thus affect the relationship between the two parties. (6). So, I would tend to believe the superman (or superwoman) approach. In this approach, one would focus all the resources on the potential candidates such that he or she can acquire as much experience or knowledge as possible. With the experience and knowledge, comes the leadership. Then the rest of the development is straight forward. (7). So, in principle, the bridge between the industries and the academic institutions is a good idea. But, in reality, the transient nature of the world make it very in-efficient. By the time, one is ready to challenge the world, he is likely to find out that his old sponsor is now out looking for his next job. Such examples are everywhere. (8). I have been following the superman principle to stay in the CFD and computer graphics, regardless of my sources of income. I think, this is one practical way to understand the whole technical issues of CFD world. And the critical step of over-all optimization requires such knowledge and experience base. (9). It is hard to make the distributed expert system work, because of the transient nature of the states, the communication lines, the priorities, ... and the missing of the nodes due to reorganization. (10). If one look at the CFD-online as a system, then it is a system for superman principle. In this system, the person who asks good questions, decodes messages posted is actually accumulating his knowledge and experience base. As this knowledge and experience base mature, he is ready to challenge the real world problem. (11). To share the distributed expert knowledge and experience among experts? and then solve the real problem? here on CFD-online, most of the time, it is hard to figure out the contents of the posted message. (to get the National Science Foundation money first then ask a CFD question here may take a long time. Same is true for me answering questions here, because in real life, very few come to me for answers at work places. Unless, their project has this item in it. Just a joke. )

At the start of my post-graduate studies involving a lot of CFD-work I read through many papers dealing with topics ranging from discretization methods, turbulence models, grid generation etc. My impression was that there are many issues that are still unresolved in the field of CFD, especially with regards to turbulence modeling. After my literature survey I visited a number of commercial CFD-code websites, mainly to decide which code to use for my investigations. The projected impression from these websites was that there is nothing we cannot solve with CFD.

I realize that in order to sell a product you need to emphasize what it can do rather than the opposite, however discussions with people who use the commercial codes for consultation work leaves me with the uneasy feeling that at times, very little consideration is given to the danger areas of CFD. Whether this is deliberate or due to ignorance I do not know, in some cases I believe it is the latter but I fear that most of the times it has to do with time limitations. My point is this: When considering the difficulties associated with dealing with phenomena such as turbulence etc. one can easily arrive at the conclusion that the way in which CFD is used in industry is somewhat premature. I am NOT saying that no useful solutions are obtained but rather that the use of CFD is running ahead of the capabilities of commercial codes. Maybe we are selling a product that still needs more time to develop. In this sense CFD is more art than science, the product of a simulation is after all a host of colorful pictures.

(1). In terms of the degree of difficulty, CFD is on the top of the listing. (2). In terms of the number of problems solved, it is on the bottom of the listing. (3). How many problems are actually solved? I would say that the number is very small. (4). For the simplest flow over a cylinder,can we say that the solution is known and the procedure to obtain such solution is repeatable? (5). The flow over a flat plate is cosidered known, not because of CFD but because of the boundary layer theory. (6). Then is the flow over the backward facing step solved? Are we still getting different sizes of separating bubble when different turbulence models are used? Do you consider those as the same solution or different solutions? (7). For flow in a curved pipe, such as the IC engine inlet manifold, can we say that we know how to predict the flow profiles entering the engine inlet port? (8). Can we say that we know how to predict the flow over an airfoil? How about the real wing on the aircraft with multiple elements? (9). The four strokes IC engine has four valves per cylinder now, so it is more difficult than the two strokes engine, right? Wrong! There have been attempts to simulate the flow through two strokes engine, but no luck, it is too difficult. (10). That is the state of art about the CFD. (11). It seems to me that there is a "quantum jump" in the degree of difficulty related to the CFD solutions. If someone is still debating on the "upwind method" and "numerical viscosity", it is going to take forever to solve a CFD problem. (12). Perhaps, no one is interesting in getting the CFD problems solved at all. As long as one is running a CFD code, the product should be all right, or at least, it is better than doing nothing at all. (if you are paid to run a code, do you care?)