Establishmodel hypothesis for the description of the experiment- physics
- distribution(s)
- parameters
Identify free and controlled parameters
Establish experiment setup in accordance with the hypothesis - free parameters are free
- teh controlled variables of the experiment are identical with the
controlled variables of the hypothesis
Perfom pilot experiment - about 10 experiments in all considered free parameter
domains in order to investigate the variability of the results - analyse results and perform a costbenefit evalutation for the
planning of additional experiments in selected domains
Test hypothesis- probability paper
- distribution tests
- variance analysis
- correlation analysis
Perform final experiments- note results
- note experiment conditions
- note unexpected results and
possible causes
Figure A1: Practical approach to experiment planning, experiment execution and experiment
validation.
The first step when planning experiments is therefore to establish all relevant hypotheses,
which may be adequate to describe the model at hand. As the hypothesis, i.e. the a-priori
information in the end may be given even more weight in the a-posteriori model than the
experimental results it is of utmost importance that these are explained and justified in detail.
The hypothesis shall include assumptions in regard to modes of failure, dependencies between
free and dependent variables, physical phenomena, which may prevail for different value
ranges for the free variables etc.
The second step in the experiment planning is to identify the free (controlled) variables and
the dependant variables in the postulated models.
As a third step the experimental set-up and equipment shall be designed such that the free
variables may be adequately controlled and such that the dependent variables are dependent
on only the free variables.
As no experimental set-up is perfect it is important to assess and describe the effects related to
the experimental set-up, which may lead to undesired systematic and un-systematic errors in
the experiment results. This discussion shall also lead up to possible modifications of the set-
up.
In some cases the observed values (dependent variables) at the experiments may not directly
be the variables, which are searched for. In such cases it is necessary to develop and document
the appropriate (probabilistic) models for the conversion of the observed values to the desired
values.
An example of the above mentioned case is when the shear capacity of glued connections in
timber structures is considered. A practical experimental set-up, however, implies that each
test specimen has two failure modes. If it is assumed that the shear strength of a glued
connection has a probability distribution function given by FX()x then this distribution
Annex B.3