679
Technical Committee 103 /
Comité technique 103
Spatial variation of soil properties
Loading conditions during an earthquake
Future developments around the project to be designed
Design versus the actual construction
In order to deal with uncertainties, various methods are
available, such as:
Global safety factor approach
Partial factor approach
Probabilistic analysis
Parametric analysis
Users of finite element models in which such methods have
been implemented need to be aware of the possibilities and
limitations of these methods.
3.5
Software and hardware issues
Although some models or processes may seem to be uniquely
described by their mathematical model, the outcome of these
models or processes, when implemented in computer software,
might deviate from their original formulation; either
deliberately or by accident. Here, the focus is on specific
software and hardware issues that might lead to discrepancies in
the outcome of a computer model which cannot immediately be
influenced by users because they are
result of specific implementations made by the
developers of the software
result of the used operating system
result of the used computer configuration.
Examples of such software or hardware issues are:
‘Bugs’ (programming flaws in the application software)
Specific implementations of models (for example
rounding-off the corners of the Mohr-Coulomb failure
criterion)
Iterative solvers and their numerical solution tolerances
Parallel solvers (solution differences depending on the
number of threads or cores being used)
3.6
Misinterpretation of results
If the modelling process has been completed, the calculation has
finished successfully and results have been obtained, it is not
the end of the story. It should be realised that the computer
model does not directly provide the answer to the original
engineering problem. Therefore, a translation needs to be made
from the results of the computer model towards the engineering
and design issues. The translation and (mis)interpretation of
results may also lead to discrepancies between the real situation
and the computer model. Examples where misinterpretation of
results might occur are:
Interpretation of safety factors
Structural behaviour (if the structure is too much
simplified)
Overlooking essential details (in particular complex 3D
models)
In general: Insufficient knowledge and understanding of
the modelling software being used.
All this is subject of the validation process. In the next
chapter, various methods of validation and other procedures are
described in order to (im)prove the quality of finite element
models and the modelling results.
4 METHODS OF VALIDATION
In the previous sections several sources of discrepancies
between a real project and its finite element model have been
identified. In order for a particular project to manage the
uncertainties and to reduce the discrepancies, the finite element
model must be validated.
Before considering validation of a computer model for a
practical application, it is relevant to verify that the models and
methods implemented in a software package are reliable. In the
first place this is a responsibility of the software developers, but
also users should consider performing a verification of models
and methods that are relevant for the solution of their
engineering problem. Verification is done by comparing the
results of computer models for typical situations with known
solutions. Examples of such solutions are:
Analytical solutions of elasticity problems, plasticity
problems, constitutive models, dynamic problems,
bearing capacity solutions, solutions of flow and
coupled problems.
Limit equilibrium solutions (approximations) for global
safety factors or bearing capacities.
Upper and lower bound solutions (limit analysis).
Benchmarks (see Section 4.5).
After a proper verification of the models and methods to be
used in a finite element model, the finite element model itself
needs to be validated. Validation of the model as a whole will
not be enough to make plausible that the results that are
obtained from the model are representative for the real situation.
In fact, discrepancies in individual components may
accidentally cancel each other out if they are not validated
individually. The validation process should therefore also
comprise the individual components of the modelling process in
addition to validation of the integral model. This also gives
insight in the accuracy of the individual modelling components.
The following sections briefly describe the validation of
individual components of a finite element model.
4.1
Validation of constitutive models and parameters
The selection of a constitutive model should be based on an
evaluation of the capabilities (and limitations) that the model
has to describe the essential features of soil behaviour for the
application at hand. In that respect, the constitutive model
provides the qualitative description of soil behaviour, whereas
the parameters in the model are used to quantify the behaviour.
The composition of the model plus parameters can be regarded
as the ‘artificial soil’ that is used in the finite element model,
which should be representative for the real soil behaviour in the
application. Before considering the numerical model in full
detail, it makes sense to evaluate the behaviour of the ‘artificial
soil’ (= model + parameters) separately in particular stress
paths. Therefore it is useful to check the behaviour of the soil in
simplified soil lab tests simulations in element tests or using a
single stress-point algorithm.
The results of the lab test simulations can be compared with
real test data. This provides insight in the possibilities and
limitations of the model to describe particular features of soil
behaviour and the accuracy at which it does so. Moreover,
parameters could be optimised to make a ‘best fit’ to the test
data. However, it should be noted that the stress paths, stress
levels and strain levels in the real application can be
significantly different than those in the soil lab tests. Hence, a
good fit between the results of a simulated test and the real test
data is not a guarantee that the artificial soil is a good
representation of the real soil in the practical application.
Nevertheless, the numerical simulation of soil lab tests is, in any
case, relevant to qualitatively understand the behaviour of the
‘artificial soil’ and should therefore be considered in the
validation process.
In contrast to soil lab tests, in-situ tests cannot be simplified
to a single stress point model. However, some in-situ tests can
still be modelled as a simplified boundary value problem in the
finite element method. The simplified modelling of in-situ tests
can be used to optimise stiffness and strength properties, and
they could be useful as part of the validation process. An
example of such a model test is the pressuremeter test, modelled
as a cavity expansion problem.
The validation of the selected soil model and parameters on
the basis of soil lab tests is not sufficient to make plausible that
the ‘artificial soil’ will sufficiently resemble the real soil in the