CFD Online Discussion Forums

CFD Online Discussion Forums (
-   Main CFD Forum (
-   -   Acceptable Accuracy or Error? (

Erich June 7, 2001 16:44

Acceptable Accuracy or Error?
If one were to model a turbomachine in order to calculate the pressure ratio, what would the common acceptable error in a 3D multiple frame of reference anaylsis? I realize this figure will vary from industry to industry; however, I am curious as to what most users of commercial codes deem as a 'worthy' solution.

John C. Chien June 8, 2001 04:31

Re: Acceptable Accuracy or Error?
(1). 3-D blade passage flow is highly complex. (2). you normally have leading edge vortices, flow separations, and the formation of secondary flow, which take off from the blade surface into the wake.(3). Th approaching boundary layer thickness also has a great impact on this secondary flow development. (4). Currently, it is very difficult to predict the loss. (off by several hundred percent from test data is not uncommon).(5). For low speed approach, you are facing the difficulty of turbulence modelling. For transonic flows, you are facing the existence of shocks. (6). I do not think that the solution to the single blade row is settled yet. So, stage or multi-stage calculations are just calculations. There is still a long way to go.(if the solution oscilates, if the solution changes with mesh size, if the solution does not converge) (7). All I can say is: commercial codes and in-house codes have been used for turbomachinery calculations, but I don't know whether anyone can say that his solution is converged and mesh indpendent. (8). Remember that the Euler solution (multi-stage also) is still commonly used in today's analysis, which does not include viscosity effect at all. Also modern blade design is fairly complex (aiming to improve various inefficiency factors), that also complicate the reliability of the solution. (9). So, you are really not talking about the accuracy of solutions, but rather a system of codes to follow, which can give you meaningful results only you have validate against the test data for that particular design.(whether in-house codes or commercial codes)

Fred Uckfield June 8, 2001 06:51

Re: Acceptable Accuracy or Error?
Maybe the question should be "what is an acceptable accuracy such that the model is still useful". One person's usefulness is another person's pretty picture.


John C. Chien June 10, 2001 00:52

Re: Acceptable Accuracy or Error?
(1). In real life, it is rarely asked about what is acceptable. (2). Meeting the project schedule is perhaps the most critical issue. "can we meet the schedule?" perhaps is the most frequently asked question. (3). Well, if one submit a paper for publication, then there will be different questions asked by the Journal reviewers. (4). By ISO9000, the engineer who performed the analysis is solely responsible for the tools he used in the analysis.(validation is required but rarely performed).

Erich June 11, 2001 11:00

Re: Acceptable Accuracy or Error?
Thanks for your comments. I was looking for an industry 'standard' more or less I suppose. As in the trends we see are within a certain percentage X...

I would imagine a standard error might be documented in ISO, but alas I fail to see how ISO is of much use. Appears to require a great deal of overhead... Or put bluntly, how to make small business as inefficient as big government, i.e. paperwork. If you build junk, yet document it as you build it, you too can be ISO certified... No guarantee of quality unless you check their drawings, etc...

clifford bradford June 16, 2001 04:53

Re: Acceptable Accuracy or Error?
Do you mean what's the acceptable difference between the CFD and the experiment?

frank July 2, 2001 13:30

Re: Acceptable Accuracy or Error?
Hi Erich,

late but I saw your question now (unfrequent visitor to this forum).

In general CAE simulation with FEM (or FEA in english terms) FDM, FVM, BEM are considered within engineering precision if they are within e.g. 5% error, in earlier days e.g. 10% were accepted (methods less developed, computing resources).

In cases you'll need to screw down to 1% or some, depending on product, component. If e.g. quantity in some servo valve or MEMS structure is of concern you might need tighter precision (and model definition), confronting two designs wrt their basic behaviour, you might not need that precision, depends on target.

Depending on which (CFD in this context) code u use, FLUENT, STAR-CD, FLOTRAN etc, you have different tools, means available to check residuals, convergence behaviour, plus mesh independence checks (can be costly but still recommendable) this can give insight in the stability of results.

I work with FLOTRAN so I can't comment on other tools, but it seems that CFD has still some intrinsic stablity issues (i.e. is the solution converged towards a real solution). Might have to do with how some solvers and wall models function.

Anyway there are means to check things out, so if a knowledgable CFD person and hotline attendant IS avail, a few things can well be analysed(check e.g. FLUENT site for loads of examples).

Small test setups can help much to check and build up experience and convidence in CFD numerical tools.


All times are GMT -4. The time now is 10:11.