Scribd
Upload
158 views

CFD Validation Procedure

The document describes the validation procedure for the 3rd Workshop on CFD Uncertainty Analysis being held in Lisbon in October 2008. It will involve validating computational fluid dynamics (CFD) simulations of the ERCOFTAC C-30 test case against experimental data. Required activities include verifying the CFD code, verifying the simulation solution, estimating numerical uncertainty, evaluating the validation error and uncertainty, and interpreting the validation results. Optionally, participants can evaluate parameter uncertainties like those from boundary conditions. The goal is to assess modeling errors and determine whether errors are from the model or from numerical and input uncertainties.

Uploaded by

Mohammed Gamal
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
158 views

CFD Validation Procedure

The document describes the validation procedure for the 3rd Workshop on CFD Uncertainty Analysis being held in Lisbon in October 2008. It will involve validating computational fluid dynamics (CFD) simulations of the ERCOFTAC C-30 test case against experimental data. Required activities include verifying the CFD code, verifying the simulation solution, estimating numerical uncertainty, evaluating the validation error and uncertainty, and interpreting the validation results. Optionally, participants can evaluate parameter uncertainties like those from boundary conditions. The goal is to assess modeling errors and determine whether errors are from the model or from numerical and input uncertainties.

Uploaded by

Mohammed Gamal
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 5

3rd Workshop on CFD Uncertainty Analysis, Lisbon, October 2008

Validation Procedure
for
rd
3 Workshop on CFD Uncertainty Analysis
Lisbon, 23rd and 24th of October 2008
Instituto Superior Tcnico
Lisbon
Introduction
The required validation procedure is adapted from the following
ASME/ANSI Standard document, cited herein as V&V 20.
ASME V&V 20-2008: Standard for Verification and Validation in
Computational Fluid Dynamics and Heat Transfer, ASME Committee PTC
61, (expected) 2008
This document uses internationally accepted concepts of uncertainty
defined in the following document on experimental procedures, also an
ASME/ANSI Standard document.
ANSI/ASME PTC 19.1-2005. 2006. Test Uncertainty. ASME, New York,
NY.
The validation process is preceded by Code Verification and
Calculation (Solution) Verification. Code Verification (required only for
those code options involved in the validation problem) shall be achieved
using the Method of Manufactured Solutions. Calculation (Solution)
Verification shall be achieved using the GCI or Least Squares GCI, or any
other method for uncertainty estimation. If non-GCI methods are used, note
that (1) any methods for error estimation (e.g. single-grid error estimation
methods) must provide error estimates for the validation quantities of
engineering interest, and (2) these error estimates must be used to obtain
uncertainty estimates, e.g. by being multiplied by a Factor of Safety Fs (as in
the GCI procedure).

3rd Workshop on CFD Uncertainty Analysis, Lisbon, October 2008

The Workshop Validation Procedure


The validation procedure for the Workshop, to be applied to the
ERCOFTAC C-30 test case, is a simple version of the more general and
elaborate V&V 20 procedure. It assumes independence of error sources and
fixed parameters (strong model concept), and uses a target uncertainty
level of 95% confidence1, consistent with the GCI. The objective is to
evaluate the validation comparison error E and the validation uncertainty
Uval and to interpret the result, according to the following prescriptions.
The validation comparison error E is defined as the difference
between the Simulation value and the experimental Data value.
(1)

E =S D

The validation uncertainty Uval at 95% confidence level is estimated as


2
2
U val = U num
+ U input
+ U D2 .

(2)

Unum is the 95% confidence level estimate of numerical uncertainty; if the


GCI or Least Squares GCI is used, Unum = GCI. UD is the experimental Data
uncertainty, also at the 95% confidence level. Uinput is due to parameter
uncertainty, = 0 for the strong model concept (but see Parameter
Uncertainty below).
Interpretation of Validation Results Using E and Uval
The validation comparison error E = S - D can also be written in terms
of component errors as
E = mod el + num + input D

(3)

The term model is the modeling error which we intend to assess. It is


composed of the errors in the governing continuum equations of the model
(e.g. the RANS model used) and errors due to any other non-ordered
approximations such as inflow and outflow boundary conditions; these
1

The more general V&V20 procedure (a) covers in detail the important cases wherein error sources are not
independent, (b) treats input parameter uncertainties, and (c) uses standard uncertainties in all derivations
so that any confidence level can be chosen by the user.

3rd Workshop on CFD Uncertainty Analysis, Lisbon, October 2008

errors do not 0 as 0 (where is a representative measure for the grid


cell size). The error num is composed of the ordered numerical errors; these
errors do 0 as 0. The error input is composed of the (non-ordered)
errors arising from the use of incorrect parameter values in the model
equations. Here we use a strong model concept so that input = 0. The error
D is composed of the total experimental errors.
A validation standard uncertainty uval (referred to in the V&V 20
procedure) is defined as an estimate of the standard deviation of the parent
population of the combination of all errors (num + input - D) except the
modeling error. Once a validation exercise has determined values S and D of
a validation variable, then the sign and magnitude of the validation
comparison error E = S-D are known. (E uval) then defines an interval
within which model falls (with some as-yet unspecified degree of
confidence). Thus E is an estimate of model and uval is the standard
uncertainty of that estimate and can properly be designated as u of model or
uval. For the simple validation procedure of the Workshop, we use the more
specific and common engineering target Uval with 95% confidence, rather
than standard uncertainty uval.
Thus, (E Uval) defines an interval within which model falls, with
~95% confidence, or

mod el [ E U VAL , E + U VAL ], Confidence ~ 95% .

(4)

Application
(1) If
E>> Uval
then probably model E.

(5)

(2) If
E Uval

(6)

then probably model is of the same order as, or less than, (num + input - d).

3rd Workshop on CFD Uncertainty Analysis, Lisbon, October 2008

In the first case one has information that can possibly be used to
improve the model (i.e., reduce the modeling error). In the second case,
however, the modeling error is within the noise level imposed by the
numerical, input, and experimental uncertainties, and formulating model
improvements is more problematic.
Parameter Uncertainty
Although evaluation of parameter uncertainty Uinput (or, as in V&V
20, uinput ) is highly recommended for thorough validation exercises, it is not
strictly necessary, since the decision of which parameter values to include in
Uinput and which to consider hard-wired is somewhat arbitrary. The limit
situation of all parameter values fixed has been termed the strong model
concept, in which Uinput = 0 by definition. Regardless of which parameters (if
any) are included in Uinput , nothing is ignored in the validation procedure.
The errors and uncertainties are still present and produce error and
uncertainty in the final model validation; E = S - D is not changed.
To obtain an estimate of Uinput, the sensitivity of the simulation results
(in the quantities of interest) to parameter variation must be determined
numerically, and an estimate of the distribution of these input parameters
must be made. The two methods described in V&V 20 are based on an
uncertainty propagation method (local) and a sampling (Monte Carlo)
method. Note that in situations (not uncommon) in which the same
parameter uncertainty affects both the simulation and the experimental
uncertainty (via experimental data reduction dependencies) it is not possible
to separate the three contributions to uncertainty as in Eq. 2. Instead, Uval
must be estimated in a tightly coupled and complicated procedure to avoid
problems. (One particular problem is the case where the effects of a
parameter uncertainty should approximately cancel between Uinput and UD if
treated correctly, but will contribute two terms if Eq. 2 is used.) The
estimation of Uval (or uval) in such coupled cases is at the core of the
methodology presented in V&V 20. For the Workshop exercises, we will
assume that Uinput and UD may be determined independently, and use Eq. 2.
It is not required to investigate parameter sensitivity for the Workshop
participation, but such results would be of great interest. At this stage in the
development of RANS turbulence models, it is probably not of interest to
further investigate numerical parameters of these models. The most
significant candidates for parameter uncertainty are the inflow boundary
4

3rd Workshop on CFD Uncertainty Analysis, Lisbon, October 2008

conditions, and to a lesser extent the outflow conditions. (The latter can in
principle be made negligible by using a long enough computational domain
L and possibly by extrapolation to L = .)
Summary of Workshop Activities
Required:
1. Code Verification by MMS
2. Calculation (Solution) Verification for the ERCOFTAC C-30 test case
(Local and Integral flow quantities identical to the previous editions of
the Workshop)
3. Validation exercise for the ERCOFTAC C-30 test case:
a) Estimation of Unum = GCI or Least Squares GCI or other 95%
uncertainty estimator (not merely an error estimator).
b) Evaluation of:
the validation comparison error E from Eq. (1)
the validation 95% uncertainty Uval from Eq. (2)
c) Interpretation of the validation results using Eqs. (4,5,6)
Optional:
Evaluation of parameter sensitivity and/or uncertainty [using Uinput in
Eq. (2)], preferably to assess the significance of the inflow boundary
conditions.

You might also like