Verification and Validation in Computational Solid Mechanics and The ASME Standards Committee
Verification and Validation in Computational Solid Mechanics and The ASME Standards Committee
L. E. Schwer
Schwer Engineering and Consulting Services
Abstract
WIT Transactions on The Built Environment, Vol 84, © 2005 WIT Press
www.witpress.com, ISSN 1743-3509 (on-line)
110 Fluid Structure Interaction and Moving Boundary Problems
Somewhere between the above succinct and elegant definitions I offer the
following layman’s definitions:
• Verification - getting (most of) the bugs out of the code.
• Validation - demonstrating the numerical model is capable of making
(appropriate) predictions.
WIT Transactions on The Built Environment, Vol 84, © 2005 WIT Press
www.witpress.com, ISSN 1743-3509 (on-line)
Fluid Structure Interaction and Moving Boundary Problems 111
Understanding this diagram start with the box labelled ‘Reality’ which
represents the physical system to be numerically modelled, e.g. a wing on an
aircraft. From this box we move clockwise and being the model building activity
which leads to a mathematical model. The mathematical model should be
thought of as a system of partial differential equations, arrived at by considering
what aspects of the reality need to be described, i.e. analysis. Continuing the
airplane wing example, suppose we are attempting to predict the tip deflection
under static loads, our analysis may indicate that a simple Bernoulli-Euler beam
theory should be adequate.
Next we convert, via programming, the system of differential equations
(mathematical model) into a numerical algorithm (code), which along with initial
and boundary conditions, material properties and a description of the geometry
form the “Computational Model.” “Verification Activities” are the checks and
sample problems we use to exercise the Computational Model to provide
confidence that we have converted the mathematical model into a correct
computational model. For the aircraft wing example, it is easy to imagine the
complex cross sectional properties of an aircraft wing, Thus as part of the
verification activity know analytical or benchmark problems, associated with
WIT Transactions on The Built Environment, Vol 84, © 2005 WIT Press
www.witpress.com, ISSN 1743-3509 (on-line)
112 Fluid Structure Interaction and Moving Boundary Problems
WIT Transactions on The Built Environment, Vol 84, © 2005 WIT Press
www.witpress.com, ISSN 1743-3509 (on-line)
Fluid Structure Interaction and Moving Boundary Problems 113
WIT Transactions on The Built Environment, Vol 84, © 2005 WIT Press
www.witpress.com, ISSN 1743-3509 (on-line)
114 Fluid Structure Interaction and Moving Boundary Problems
fault tolerant operation in vivo. For economic reasons, perhaps not all
required test configurations are tested, and, in the latter example,
limited access to testing ‘facilities’ for biomedical devices, both place
an emphasis on “just do some calculations.”
• From Developer/Analyst to Black-Box/User – Less than 20 years ago,
most analyses were performed by the same person that wrote the
analysis code, or worked in the same group with the code developer.
Today, almost all analyses are performed using commercial software,
where even a trusted user is not allowed to access the source code, for
verification, or any other, purpose. The most ardent user is forced to
‘know’ the analysis package through its user documentation, which
ranges from at best good to all too often nonexistent.
As more emphasis is placed on using computational mechanics to replace
testing, and the corresponding computational results have greater economic, and
safety, consequences, management will more frequently ask “How valid are
those results?” and enlightened management will require an informed and
quantitative answer.
As of this writing, it seems the path forward to V&V standards will involve
dissecting computational mechanics into its basic parts and then writing Best
Practices, and hopefully standards, for these basic parts. The immediate
challenge is how to dissect such a topic where many of the parts are interrelated;
the surgeon’s task of separating Siamese twins comes to mind. A logical
approach would be to select those areas of computational mechanics that can be
isolated relatively easily, e.g. error estimation and constitutive modelling, and
attempt Best Practices documents for those areas. But the emphasis in this
approach will of necessity be on verification, as validation requires the whole of
computational mechanics to perform adequately.
The most pressing need in V&V is validation, this is not to ignore
verification, but for the reasons cited above, practical guidance, in nearly any
form, on validation must be given a priority; certainly verification and validation
can proceed in parallel. There seems to be three key areas in validation that
guidance needs to be communicated to analysts, and their managers,
immediately:
1. Precision Testing – what is required to specify and perform experiments
that will be meaningful and useful for validation.
2. Validation Metrics – how do we quantify the comparisons between
measurement and simulation results.
3. Role of Non-Determinism – neither the observed nor simulated
behaviour is known with certainty, e.g. due to randomness in the
physical system and modelling idealizations, this further complicates
the validation process.
It is my belief that guidance in these three areas of validation needs to be
presented in a form that provides for both education and guidance. Validation
Metrics are an area of research and development and thus require education to
disseminate the results. Most analysts will have some training in the area of Non-
WIT Transactions on The Built Environment, Vol 84, © 2005 WIT Press
www.witpress.com, ISSN 1743-3509 (on-line)
Fluid Structure Interaction and Moving Boundary Problems 115
WIT Transactions on The Built Environment, Vol 84, © 2005 WIT Press
www.witpress.com, ISSN 1743-3509 (on-line)
116 Fluid Structure Interaction and Moving Boundary Problems
7 Conclusions
Significant progress has been made in verification and validation in
computational mechanics. Increasing recognition of the importance of V&V is
evidenced by such items as the formation of another ASME V&V Standards
Committee (PTC #61) with a focus on ‘procedures for quantifying the accuracy
of modelling and simulation in computational fluid dynamics and heat transfer,’
and adoption by two respected journals in computational fluid dynamics of V&V
related requirements for publication of computational and related experimental
results.
While the publication of verification and validation guides by the AIAA CFD
Standards Committee, and soon the ASME Committee on Verification and
Validation in Computational Solid Mechanics, are necessary first steps, the next
step of attempting to write best practice documents appears to be a greater
challenge. These committees need the support of the computational mechanics
community. The best type of support is involvement in the committee’s
activities. The members of these committees are volunteers who give generously
of their time for the larger good of the community. The committee work is very
WIT Transactions on The Built Environment, Vol 84, © 2005 WIT Press
www.witpress.com, ISSN 1743-3509 (on-line)
Fluid Structure Interaction and Moving Boundary Problems 117
rewarding on a personal and professional level, you will not regret becoming
involved in this worthwhile effort.
References
[1] Roache, P., Verification and Validation in Computational Science and
Engineering, Hermosa Publishers, ISBN 0-913478-08-3, 1998.
[2] Computational Fluid Dynamics Committee on Standards, Guide for
Verification and Validation of Computational Fluid Dynamics
Simulations, American Institute of Aeronautics and Astronautics, AIAA
G-077-1998, ISBN 1-56347-285-6, January 1998.
[3] Popper, K., The Logic of Scientific Discovery, English translation,
Routledge Classics, ISBN 0-415-27844-9, 1959.
WIT Transactions on The Built Environment, Vol 84, © 2005 WIT Press
www.witpress.com, ISSN 1743-3509 (on-line)