I am working on estimation of the uncertainty with in the prediction of a fluid dynamic problem by a numerical code. There are some input parameters (thermal conductivity, heat capacity and..) and also, some target parameters (evaporation rate) in numerical code which would be interesting to be investigated by the sensitivity analysis and finally estimating the margins of uncertainty in prediction of the target parameters. I was wondering if could someone tell me about a standard procedure to analysis the uncertainties in predictions of a numerical code ?
There are many papers of G. Iaccarino (Stanford) about this topic, see his work on RG
Please look at ASME V&V20 standard. Chapters 3 and 4 are detailed discussion of this issue and it has some numerically solved examples to nail the concept.
Basically you have three type of uncertainty:
1- Uncertainty in your model structure (Is this model is mathematical representation of the physics?)
2- Uncertainty in the solution of that mathematical model (which is usually done by numerical methods like FVM, FDM, FEM, ... so you do not have exact solution)
3- Uncertainty induced by uncertainty in numerical scheme parameters, material related parameters, and geometry of the problem(I.C., and B.C. and roughness, ...)
In the standard you will find how to quantify those sources one by one and show you how you may find the total uncertainty in the modeling activity.
That standard also discuss how to find the uncertainty in experimentation (not modeling activity, that is the fourth type) and incorporate that in your universal uncertainty analysis which has those four factors.
Mohamad ,
Please refer to chapter in Computational Error and Uncertainty Quantification in Special Topics in CFD.
Ideen
You need to calculate a kind of non linear approximation of maximum and minimum and put a check on this reaching to the steady state situation...
Hi. If you want to evaluate the parameter uncertainty in a high dimensional calibration problem the DDS-AU algorithm is a must. In this kind of uncertainty analysis the biggest problem is the inability of sampling enough values near the optimous point of the objective function when uniform random sampling is used. An implementation of this algorithm in the widely used R language is described in this paper. The code is freely available in GitHub. Article Using R to easily and efficiently predict the uncertainty in...
Regards
- Badges
- Science topic