4.2 Requirements for the Perfect Modelling Approach 153
problem solving process – goes then iteratively from left (the customer demands)
to right (the process variables), trying to formulate requirements that would satisfy
the needs, then to find parameters (or parameter values) that would satisfy the
requirements, and finally to find process variables (or their values) that affect the
corresponding design parameters in a way to achieve the desired results. But if we
discuss the modelling methodology and the respective tools in general, and not in a
specific case, it is a bit problematic to determine particular customer needs. This
means that even if we carry out a survey about the needs and desires of the
(potential) customers, the answers would be so different that they would have to be
reorganized and probably reformulated and grouped in an observable list of
generalized needs. Therefore, a quicker and cheaper possibility is to conduct a
mental experiment and to consider what would these needs and desires be. Suppose
we start with just one customer need, which is
CN: get/have perfect models.
To satisfy this need it is sufficient to define just one requirement:
FR: use/apply a perfect modelling approach.
It seems simple but does not change much in the initial situation, because no
approach until now is perfect and it is still unknown what would be a perfect
approach. Therefore, we have to do some decomposition and try to find other
relations. Although according to the axiomatic design CNs do not always need
decomposition, in this case it would help us clarify the requirements. So, let us
consider what it means for a model to be perfect. Of course, everybody would have
his own preferences, but we shall say that perfect models have the following five
main inherences.
The most important one is that each model should
resemble all important
modellee traits
as closely as possible - CN
0
. This is necessary in order to ensure
that any
use of a model instead of its modellee would be sensible at least for the
purpose for which the model is created:
CN
1
. The purpose could be the creation or
the improvement of the modellee itself, the planning of the modellee's production
or simply the taking of a (perhaps more general) decision, related to the modellee.
The next inherence of a perfect model is its
unrestrictedness, i.e., the possibility to
model or represent anything without exception:
CN
2
. Next important trait to be
expected can be
ease and comfort when dealing with the models – model handling
should be effortless:
CN
3
. Of course, the models should also be robust and
reliable
: CN
4
. And last but not least – the perfect models should live forever: CN
5
.
Now let us see what requirements should be imposed on the perfect modelling
in order to cover these needs? We shall give the same indices to the corresponding
requirements in order to keep the correlations clear and easily maintainable.
Therefore,
CN
0
leads to the first requirement – allow representation of arbitrary
attributes
: FR
1
. CN
1
, respectively, leads to the next requirement – ensure the model
adequacy
: FR
2
. The requirement correlated to CN
2
still cannot be formulated
clearly at the moment, but we can try again after formulating its sub-requirements,
which correspond to parts of the decomposed
CN
2
. Thus we say for now that FR
3
is underspecified. The same holds for the requirement, correlated to
CN
3
– it can be
neither specified nor named at this stage and just shows once again why the
axiomatic design postulates zigzagging among the domains during the design
process. Similarly to
FR
2
, FR
4
is underspecified and can be determined/named only
after working out the details at the lower levels. The requirement correlated to
CN
4
could be formulated as
use of reliable modelling methods and tools: FR
5
.