|
[Sponsors] |
June 21, 2001, 10:10 |
Dimensionless Calculation and Convergence
|
#1 |
Guest
Posts: n/a
|
Hi I am using a commercial CFD code. Just out of curiosity I was simulating a developing turbulent pipe flow of an incompressible fluid. According to dimensional analysis, this problem can be described by a single dimensionless number, say the Reynolds number. So I kept the Reynolds number constant, but was varying the physical parameters, velocity, density, length and viscosity (but keeping the Reynolds number constant). So say for a Reynolds number of 1E6 I gave the viscosity the value 1E-6 and the other variables 1.0 and so on. Understandingly I obtained the same results for all combinations (in dimensionless graphs). The only difference was, that some combinations needed more iterations to get a converged solution than others. This is also understandable. The remarkable thing was that the most number of iterations I needed for a combination of the parameters, which is closed to a combination in the "real world". Does anyone know which combination of parameters is best, e.g. keeping all but one parameter 1, but one, or keeping all parameters the same?
|
|
June 22, 2001, 07:33 |
Re: Dimensionless Calculation and Convergence
|
#2 |
Guest
Posts: n/a
|
Unfortunately nothing can be said "a priori" ... it depends on an high number of factors ... first of all we'd like to know which kind of code you're using ... and some more details of your problem.
|
|
June 22, 2001, 07:55 |
Re: Dimensionless Calculation and Convergence
|
#3 |
Guest
Posts: n/a
|
First of all I do not consider it as a problem, I was just curious. I am using CFX4.3. I was simulation a turbulent pipe flow having a Re=40000. I did four clases of simulations: first:u=1m/s,d=1m,mu=1/40000kg/ms,rho=1kg/m^3 second:u=40000m/s,d=1m,mu=1kg/ms,rho=1kg/m^3 third:u=200m/s,d=200m,mu=200kg/ms,rho=200kg/m^3 fourth:u=6m/s,d=0.1m,mu=1.8E-5kg/ms,rho=1.2kg/m^3 After 200 Iterations the ratio of the mass source residual to the residual after the second iteration for the fourth simulation was one magnitute higher than for the other three iterations. Hope this is detailed enough.
|
|
June 22, 2001, 08:40 |
Re: Dimensionless Calculation and Convergence
|
#4 |
Guest
Posts: n/a
|
I think CFX uses a false time stepping approach to solving to steady state. If I'm right then they'd have to calculate a physical time scale for the model based on a residency time, this time is more a function of pipe length than diameter. More than that I can't say...??
Fred. |
|
June 22, 2001, 13:35 |
Re: Dimensionless Calculation and Convergence
|
#5 |
Guest
Posts: n/a
|
(1). All I can say is that it is a stupid code. (2). I don't know whether it is also UNITS sensitive. Can you convert the case-four into English unit and run a case? this should make the number larger for the diameter.
|
|
June 23, 2001, 03:57 |
Re: Dimensionless Calculation and Convergence
|
#6 |
Guest
Posts: n/a
|
Stupid Code. How exactly can you justify that John?
|
|
June 23, 2001, 17:31 |
Re: Dimensionless Calculation and Convergence
|
#7 |
Guest
Posts: n/a
|
(1). That's very simple. (2). We are answering questions. And since there is no such thing as a "stupid question", then something else must be stupid. (3). We can't say that the person who ask the question is "stupid", because the forum is here to answer the reader's question. They are here because they are smart. So, we can't say that the reader is stupid. A person asking question is learning, so he is not stupid. (4). The only thing I can think of is the code. It is creating all sort of questions, and it can't answer its own question, so it must be stupid.
|
|
June 25, 2001, 03:27 |
Re: Dimensionless Calculation and Convergence
|
#8 |
Guest
Posts: n/a
|
I'm not really sure that help but your problem reminds me 2 thinks :
1st : when using a computer : a + b(a-c) + c could be not equal to a(1+b) + c(1-b) it depends on the order of magnitude of b with respect to 1 and of (a-c) with respect to a or c. 2nd : the distribution of real number is not continuous in a computer (see for example IEEE floating point execption messages). Hope that help, Sylvain |
|
June 25, 2001, 07:41 |
Re: Dimensionless Calculation and Convergence
|
#9 |
Guest
Posts: n/a
|
Ok, so the assumption is that everyone here is 'smart' and so the only thing left, when unable to answer a question, is the code. OK. I'm just left with a sense of bewilderment that all those clever developers at those big commercial CFD software houses have themselevs been so stupid as to even consider using a pseudo time stepping method for steady state simulation (or maybe stupid not to have implemented it generically).
Fred. |
|
June 25, 2001, 10:54 |
Re: Dimensionless Calculation and Convergence
|
#10 |
Guest
Posts: n/a
|
(1). If the code vendor or developer is writing a code for himself, then there is no problem. (2). If the code developer is writing a code for the user without considering the user's ability to understand his original intention (not possible because of the nature of the black box), then the users will run into problems (which might be normal for the developer's scheme ) and can not understand the behavior. (3). I have heard that in business "the customer is always right". Because the definition of "right" is relative. (4). But, it seems to me that many vendors are trying to say that "they" are the best, and their codes are the best. (5). My thinking is: if the code can not help the user to get his solution, then the code is stupid and useless. (you can't say that the user select the code because he is stupid. )
|
|
June 25, 2001, 11:28 |
Re: Dimensionless Calculation and Convergence
|
#11 |
Guest
Posts: n/a
|
True, all too true. There's still an assumption that the black box is intended to work with no prior training or experience at all. I don't think any vendor is brave enough to stand up and say that their blox is black, maybe medium grey but not black. So, not a stupid user or code but an as yet uneducated user and a grey code.
Fred. |
|
June 25, 2001, 14:32 |
Re: Dimensionless Calculation and Convergence
|
#12 |
Guest
Posts: n/a
|
(1). I think, I like your answers. (2). I also agree with your analysis. (3). The fact is: If a couple of accidents happened and killed some user's of Ford SUV, then some 6 to 12 millions of tires were recalled and replaced. (4). My goal here werea). get everyone to understand the importance of "the mesh independent solution", (b). get everyone to understand that the wall function is a "black art", (c). get everyone to understand the proper way to use a "commercial code". (5). My last goal is: get everyone understand that to do business in CFD, his solution must be "repeatable", whether it comes from a commercial code, or from his own code. (6). How to create such code? It is the business of the code vendor. Since CFD is rather young, so, it is attractive to a kid like a toy. For this reason, any toy with safety problem must be re-called right away. (7). The other way to do is to set up a law and ask the user of a commercial code to take a test and get a license first, before he is certified to use the code. (Is that what MS and other software vendors are doing these days? I heard that it is easier to get a job when he pass the exam and becomes a certified engineer) (8). Without these basic rules, the future is unknown. And if a code is really good, with the speed of Internet today, that good news will be everywhere in less than one day world wide. And everyone will jump on it without any Ads.
|
|
June 26, 2001, 00:01 |
Re: Yes, some kind of regulation should be there
|
#13 |
Guest
Posts: n/a
|
I agree 100% with you, Dr. Chien on your item (7). There should be an international evaluation board too. The publishers of journals should be asked to publish the license number of the author(s). Otherwise, with the presence of these 'unlicensed drivers on the roads driving the latest model hi-tech automobiles without knowing what they are doing', the lives of many people will be under severe threat.
|
|
July 1, 2001, 09:52 |
Re: Dimensionless Calculation and Convergence
|
#14 |
Guest
Posts: n/a
|
Hi Fabian,
the subject is interesting, non-dimensionalization is used frequently in publications, it's neat, helps to gain insight in global behaviour of physics, that's the reason that it's there, another is to impress the community but that's 'outside the scope of this posting'. You might search the internet, using appropriate keywords, ask someone known to you as being scientifically minded, post again here(probably with some tips resulting next time, from list members practical about the argument) - after reading a intro bit about the subject's mechanism. don't take offence, Frank |
|
|
|
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
Convergence | Centurion2011 | FLUENT | 48 | June 14, 2022 23:29 |
Reference Velocity Calculation & Convergence Problem | hasan_hzl | Main CFD Forum | 0 | January 29, 2011 13:42 |
Standard K-Epsilon TO Reynolds Stress calculation (for convergence) | wjarrah | Main CFD Forum | 2 | May 27, 2010 05:32 |
CONVERGENCE (transient calculation) | mehrnaz | FLUENT | 2 | March 8, 2004 02:55 |
convergence in trantient calculation | Mohamed | CFX | 2 | May 14, 2003 08:15 |