Title: The Importance of Adequate Verification and Validation Strategies in Risk Management
1The Importance of Adequate Verification and
Validation Strategies in Risk Management
- Prof. Ch. Hirsch
- Vrije Universiteit Brussel
- President, NUMECA International
2Content
- Introduction
- Error identification and management
- Verification requirements
- Validation requirements
- Beyond VV
- The new generation of simulation tools
- Non-deterministic simulations
- Robust design methodologies
- Conclusions
3Introduction
- Whats new in simulation and Computer Assisted
Engineering (CAE) - CAE covers all areas of physical simulation,
particularly - Computational Structural Mechanics (CSM)
- Computational Fluid dynamics (CFD)
- Computational Electromagnetics (CEM)
- Growth of computer power leads to unprecedented
levels of CAE simulations - Million of points or degree of freedoms (DoF)
- Complex geometries
- Multiphysics (coupling fluid-thermal
fluid-structure aero-acoustics,) - Leading to Multidisciplinary design and
optimization software systems (MDO)
4Objectives
- This leads to a growing request for quality
assurance (QA) on the simulation tools - In order to respond to this request, joint
efforts are required to define - uncertainty bounds on simulation results
- limits on models
- error control methodologies
- appropriate test cases and QA strategies
5Validation and verification
- Validation The process of determining the degree
to which a model is an accurate representation of
the real world from the perspective of the
intended uses of the model. - Verification The process of determining that a
model implementation accurately represents the
developer's conceptual description of the model
and the solution to the model. - Verification is a prerequisite, upstream of the
validation process and requires a dedicated
effort towards adequate methodology
6VV
7Identification of Error Sources
- Model errors and uncertainties
- Difference between exact solution of the
equations and real flow - Turbulence model uncertainties predominate
- Application uncertainties
- Uncertain boundary conditions or geometry
- Discretization or numerical errors
- Difference between exact solution and solution on
a finite grid - In industrial CFD, solutions are usually not grid
independent - Iteration or convergence error
- Difference between converged solution and result
after n steps - User errors
- Mistakes, blunders, carelessness, optimism, etc.,
- Programming or code errors (bugs)
8Verification process
9Verification process
- No matter how many tests and empirical
experiments we perform, we can never prove that a
software implementation is error free, or has
zero software defects. - Experience suggests that one can never totally
eliminate bugs from complex CFD software,
certainly not simply by performing algorithm
testing alone. - An important approach to algorithmic testing in
verification is the method of manufactured
solutions
From W.L. Oberkampf, T. G. Trucano, Ch. Hirsch
(2002), Verification, Validation, and Predictive
Capability in Computational Engineering and
Physics. Paper presented at FOUNDATIONS 02,
Foundations for Verification and Validation in
the 21st Century Workshop, October 2002
10Validation process
11Validation hierarchy
12Validation Application domain
- In case a) we have high confidence that the
relevant physics is understood and modeled at a
level that is commensurate with the needs of the
application. - In case b) the boundary of the domain indicates
that outside this region there is a degradation
in confidence in the quantitative predictive
capability of the model. - Outside the validation domain the model is
credible, but its quantitative capability has not
be demonstrated.
From W.L. Oberkampf, T. G. Trucano, Ch. Hirsch
(2002), Verification, Validation, and Predictive
Capability in Computational Engineering and
Physics. Paper presented at FOUNDATIONS 02,
Foundations for Verification and Validation in
the 21st Century Workshop, October 2002
13Validation Application domain
- Figure 5a depicts the prevalent situation in
engineering which shows the complete overlap of
the validation domain with the application
domain. The vast majority of modern engineering
system design is represented in Fig. 5a. - Figure 5b represents the occasional engineering
situation where there is significant overlap
between the validation domain and the application
domain. Some examples are prediction of crash
response of new automobile structures, entry of
spacecraft probes into the atmosphere of another
planet, and the structural response of new
designs for deep-water offshore oil platforms. - Figure 5c depicts the situation where there is no
overlap between the validation domain and the
application domain. - Predictions for many high-consequence systems are
in this realm because we are not able to perform
experiments for closely related conditions. - The inference from the validation domain can only
be made using both physics-based models and
statistical methods. Model calibration, which
employs explicit tuning or updating of model
parameters to achieve some degree of agreement
with existing validation experiments, does not
fully assess uncertainty in the predictive use of
the model. - The requirement for predictive code use far from
the validation database necessitates
extrapolation beyond the understanding gained
strictly from experimental validation data. - For example, the modeling of specific types of
interactions of physical processes may not have
been validated together in the given validation
database. There is then uncertainty in the
accuracy of model predictions describing such
interactions.
14Dramatic illustration
- A significant example is to be found in the
recent report from NASA about the failure of the
scramjet hypersonic vehicle experiment X-43A. - The report, issued recently at http//www.space.co
m, mentionsSPACE.com has learned that the
failure of the NASA X-43A hypersonic aircraft in
June 2001 was the result of inaccuracies in
computer and wind-tunnel tests that were based on
insufficient design information about the vehicle
itself. According to the NASA-MIB documents, the
X-43A Hyper X launch vehicle "failed because the
vehicle control system design was deficient for
the trajectory flown due to inaccurate analytical
models, which overestimated the system margins."
15VERIFICATION AND VALIDATION EFFORTS IN EUROPE
16European activities
- Various collective efforts in Europe for CFD
validation - ERCOFTAC privately funded action on CFD Quality
and Trust - QNET-CFD Thematic Network on QT
- FLOWNET Collection of data for validation test
cases - ECORA QT for CFD in the nuclear industry
17PHASE 1ERCOFTAC BEST PRACTICE GUIDELINES
18Objectives
- Guide to avoid the most common pitfalls
- Guidelines not specific to individual codes,
methods or applications - Guidelines that have very wide support
- Not exhaustive 20 of rules to cover 80 of
aspects - Modular form for each topic
- Discussion section
- Problem description and discussion, including
references to important books, articles and
reviews with relevant examples for the
user. - Short simple statements of advice which provide
clear guidance, are generally accepted, and are
easily understandable without elaborate
mathematics.
19Follow-up
Expert Knowledge-base Expert knowledge on limits
of physical models in specific flow regimes
Extension Extension of scope of BPG to more
difficult flow regimes
RevisionUpdates, corrections andimprovements
within current scope
Application ProceduresAdvice on how to best do
CFD and statements on accuracy for relevant
industrial cases
20PHASE II THE QNET-CFD NETWORK
21Objectives
- To assemble, structure and collate existing
knowledge on the industrial application of CFD
and to make these available to European industry - To improve the quality of the industrial
application of CFD - To improve the level of trust that can be placed
in industrial CFD calculations by assembling,
structuring and collating existing know-ledge
encapsulating the performance of models
underlying the current generation of CFD codes - To establish a shared database of computational
and experimental results to support industrial
applications - To provide a regular state-of-the-art review on
quality and trust - To promote technology transfer between industries
through workshops, regular meetings and
electronic communication
22Membership
- 44 participating organisations with
representatives from - 8 European Union member states
- 1 EFTA member state
- 2 Pre-accession states
- The membership is principally built from
industries, although there is a strong
representation from universities and research
organisations - A diverse range of industrial sectors is
represented, including industries as diverse as
civil construction and textiles - Also included representatives from the major
European CFD code vendors - The organisations involved range from large
industrial concerns to SMEs
23Research Centres
Industries
Q N E T - C F D
DERA Health and Safety Executive NCSRD The
Meteorological Office CEA CIRA CIMNE Vattenfall
Utveckling AB Czech Academy of Science
- WS Atkins Science Technology
- CFS Engineering SA
- Renault
- Sulzer Innotec
- NUMECA Int.
- Electricite de France
- SNECMA Moteurs
- MTU
- AEA Technology
- BMW Rolls-Royce
- MAN Turbo
- ABB ALSTOM Technology
- Rolls-Royce Power Engineering
- Fluent Europe
- Magnox Electric
- Mott MacDonald
- SNPE Propulsion
- HR Wallingford
- Arup RD
Universities
University of Brussels (VUB) University of
Karlsruhe University of Surrey University of
Poitiers University of Southampton University of
Rome Martin-Luther-Universitat NTUA FH
Niederrhein Cranfield University University of
Florence University of Czestochowa
24Work Procedure
- Development of knowledge base
- Develop Quality Trust knowledge base that
consist of a library of application challenges
(AC) within each of the industry sectors and
information on a series of well-documented flow
regimes that underlie these industrial
applications - Identification and documentation of application
challenges - Quality checks of application challenges
- Identification of underlying flow regimes (UFR)
- Documenting underlying flow regimes and
development of best practice advice - Quality checks on underlying flow regimes and
review of best practice advice - Development of application challenge best
practice advice - Review of best practice advice
- Exploitation of knowledge base after end of
project
25Quality Control
- Quality Control is a major element through the
QNET-CFD operation - Managed by a Quality Coordinator (A. Hutton)
and a Scientific Coordinator (W. Rodi) - A Quality procedure document has been generated,
providing guidelines for acceptance or rejection
of submitted ACs - All ACs have been screened and cross-reviewed by
2 partners - After acceptance, they are recorded in the
knowledge and data base.
26Thematic Areas
- The Network is organised around 6 Thematic Areas
(TA) aligned with the following industrial
sectors - TA 1 EXTERNAL AERODYNAMICS
- TA 2 COMBUSTION AND HEAT TRANSFER
- TA 3 CHEMICAL PROCESS, THERMAL HYDRAULICS AND
NUCLEAR SAFETY - TA 4 CIVIL CONSTRUCTION HVAC
- TA 5 ENVIRONMENT
- TA 6 TURBOMACHINERY INTERNAL FLOWS
- Each Thematic Area has a TA coordinator that
assumes responsibility for the activities of the
group
27Underlying flow regimes UFR -
- Subdivided in 4 categories
- UFR 1 -- Free flows
- UFR 2 -- Flow around bodies
- UFR 3 -- Semi-confined flows
- UFR 4 -- Confined flow
- Each UFR can be attributed to several ACs
- Contains experimental data and a variety of CFD
simulations, from which Best Practice Advice
(BPA) guidelines are derived - From the collection of BPAs of all the UFRs
associated to an AC, a BPA for the AC is to be
established
28UFR 1 -- Free flows
- 1-01 Underexpanded jet Partner-07
- 1-02 Blade tip and tip clearance vortex flow
Partner-09 - 1-03 Buoyant plumes Partner-12
- 1-04 Annular coaxial jets, flow and mixing
- Partner-19
- 1-05 Jet in a cross flow Partner-36
29UFR 2 -- Flow around bodies
- 2-01 Blunt base flow (streamwise flow past
cylinder with cut-off end) - 2-02 Flow past cylinder
- 2-03 Flow around oscillating airfoil
- 2-04 Flow around (airfoils and) blades
(subsonic) - 2-05 Flow around airfoils (and blades) A-airfoil
(Ma0.15, Re/m2x106) - 2-06 Flow around (airfoils and) blades
(transonic) - 2-07 3D flow around blades
- 2-08 Flow around grid bars
- 2-09 Rotor/stator interaction
30UFR 3 -- Semi-confined flows
- 3-01 Boundary layer interacting with wakes
under adverse pressure gradient - NLR 7301 high
lift configuration - 3-02 Atmospheric boundary layer with rough wall
- 3-03 2D Boundary layers with pressure gradients
- 3-04 Laminar-turbulent boundary layer
transition - 3-05 Shock/boundary-layer interaction (on
airplanes) - 3-07 Natural and mixed convection boundary
layers on vertical heated walls - 3-08 3D boundary layers under various pressure
gradients, including adverse pressure gradient
causing separation - 3-09 Impinging jet
- 3-10 Plane wall jet
- 3-11 Plane wall jet in counter current flow
- 3-12 Stagnation point flow
- 3-13 Flow over isolated hills/valleys (without
dispersion) - 3-14 Flow over surface-mounted cube/rectangular
obstacles - 3-15 2D flow over backward facing step
- 3-16 Wave-driven flow in a basin
31UFR 4 -- Confined flow
- 4-01 Secondary flow in rotating and non-rotating
channels - 4-02 Confined coaxial swirling jets
- 4-03 Pipe flow - rotating
- 4-04 Pipe flow - non rotating
- 4-05 Curved duct/pipe flow (accelerating)
- 4-06 Swirling diffuser flow
- 4-07 Developing channel flow with mass injection
through wall - 4-08 Orifice/deflector flow
- 4-09 Confined buoyant plume
- 4-10 Natural convection in simple closed cavity
- 4-11 Simple room flow
- 4-12 Flows in chambers with multiple
inlet/outlets - 4-13 Compression of vortex in cavity
32Best Practice Advices
- To be derived from the BPA of individual UFR
- BPA to be strongly supported by the evidence
examined in the UFR document Section on
Comparison of CFD Calculations with Experiments - Recommendations for Future Work
- Such as new experiments to be undertaken for
which the values of key parameters are much
closer to those encountered in real application
challenges - Visit http//www.qnet-cfd.net
33 THE FUTURE OF VV
34New approach to VV
- It is generally recognized that, although actions
towards the reduction of simulation uncertainties
are still required, numerous sources of
uncertainties will always remain. - Therefore new methodologies are required in order
to incorporate the presence of uncertainties in
the simulation process and in order to introduce
the existence of these uncertain simulation
results in the decision process related to
industrial design. - This implies responses to the following questions
- How to manage uncertainties in simulations, that
is how to quantify, in a rational way, the impact
of the different sources of uncertainties on the
simulation results. - Given the existence of uncertain simulation
results, how are multidisciplinary design
optimizations (MDO) techniques to be developed
and applied in order to support the industrial
design process?
35Next steps
36Required steps 1
- Quantification and management of simulation
uncertainties - Quantification and management of uncertainties is
required at the individual discipline level (CFD,
CSM, CHT,) as well as at the integrated system
level. - The current practice of CAE simulations is based
on deterministic parameters such as fixed
boundary or initial conditions, fixed geometry
and physical properties, fixed physical model
parameters, etc. - However, these conditions are not generally known
precisely and are attached with unavoidable error
levels, as listed above. Therefore one should be
able to evaluate simulation results by
incorporating these uncertainties in the
simulation process, in order to approach a
rational quantification of uncertainties,
including the establishment of a confidence
interval for the simulation-based predictions.
37Required steps 1a
- Quantification of uncertainties
- Methods have to be developed to achieve the
following objectives - model and quantify the parameter uncertainties
and stochastic inputs (such as initial
conditions, boundary conditions, geometrical
uncertainties) - model the modeling uncertainties associated
with physical models, transport properties,
source terms - specification of the nature of the parameter and
model uncertainties (interval limits, probability
density distributions, fuzzy logic sets,) - quantification of the parameter (and model
uncertainties, based on all possible sources of
information, including experimental data,
analytical estimates, results from computational
processes, .
38- Quantification of uncertain variables
39Required steps 1b
- Management of uncertainties
- Implement probabilistic simulations, for which
innovative mathematical and algorithmic methods
have to be developed for the treatment of
differential equations containing stochastic
input parameters and model parameters. Several
methods are to be developed, assessed and
validated, such as - perturbation techniques whereby the variables of
the problem are expanded in terms of Taylor
series around their mean value (mainly for
Gaussian or weakly non-Gaussian processes) - Monte Carlo methods for complete statistics, but
generally excessively expensive - Fuzzy logic methods. Fuzzy logic allows to create
models based on inexact, incomplete, or
unreliable knowledge or data, and, moreover, to
infer approximate behavior of the system from
such models - Polynomial Chaos methods, whereby the randomness
of the solution uncertainties is represented and
the PDF of the different solution components is
defined at every point and time.
40Management of uncertainties
- Examples of probabilistic output for a model
equation, with a relative viscosity variation
41Probabilistic Design and Risk management
- Introduction of probabilistic simulations into
the design and decision process
42Probabilistic Design approach
- The objectives are therefore
- To develop aerodynamic and structural
optimization algorithms that provides designs,
which are robust with respect to uncertainties in
geometry, operating conditions, and code
simulation uncertainties. - To control and reduce risks by providing designs
with performances insensitive to intrinsically
uncertain quantities - To reduce system risks by enabling uncertainty
quantification and design strategies at the
conceptual design stage
43Probabilistic Design approach
- The two major classes of uncertainty-based design
problems are robust design problems and
reliability-based design problems. - A robust design problem seeks a design that is
relatively insensitive to small changes in the
uncertain quantities. - A reliability-based design seeks a design that
has a probability of failure that is less than
some acceptable value.
44Conclusions
- Classical verification and validation require a
large scale effort, which has to continue,
although we will never eliminate uncertainties. - Risk management needs new generation methods to
allow for - The quantitative assessment and management of
simulation uncertainties by Non-Deterministic
methodologies - Efficient probabilistic simulation outputs to
quantify reliability bounds of the predictions
(mean and standard deviations of relevant design
quantities) in a rational way, - The development and application of design
methodologies incorporating probabilistic based
simulations
45Conclusions
- The non-deterministic approach is critical in
order to - Reduce system risks by enabling uncertainty
quantification. - Increase design confidence and safety by
measuring and controlling uncertainty in
performance predictions and to enable estimates
of probabilities of failure
46- Thank you for your attention !
47Verification procedures
- From
- W.L. Oberkampf, T. G. Trucano, Ch. Hirsch (2002),
Verification, Validation, and Predictive
Capability in Computational Engineering and
Physics. Paper presented at FOUNDATIONS 02,
Foundations for Verification and Validation in
the 21st Century Workshop, October 2002
48The KNOWLEDGE BASE
- Created and managed by Univ. Surrey and Atkins
company - Collects all the data from the ACs, their
relevant UFRs - Covers experimental and the CFD data
- Highly interactive and user-friendly environment
- http//www.qnet-cfd.net
49Required steps 2
- Applications to subsystems and systems
- The objective at this step is to provide
efficient uncertainty quantification and control
strategies and algorithms for multidisciplinary
uncertainty quantification. - The general methodologies are to be applied,
validated and tested for the different
disciplines CFD, CSM, CHT,for various aerospace
and other representative applications. - In a second phase, the validated subsystem
methods are to be applied for integrated
components and systems, to be defined by the
different application areas (aerospace,
automotive, environmental,).