92
In comparative tests the refined synthesis concerns the broadness
and consistency. The optimisation is based on the gained experience
and on the current state of the knowledge.
5. Reliability
An important property in the functioning of physical systems is
their reliability. When conducting or optimising a testing process,
the occurrence of failures should be considered. For this purpose,
the operations, where failures are probable, should first be defined.
For instance, components containing electronic switches and cir
cuits are liable to failures. Since failures have a random character,
they can be tested statistically. One approach is the s.c. Monte
Carlo methodbased on repeated runs with randomly selected
(equally probable) inputs. However, a practical drawback of this
method is the great effort required.
6. Representation of test results
The final tests data, comprising both the achieved and desired
performance levels, should be compiled or edited in a consistent,
clear and uniform manner. For this purpose the following re
presentations are in use
a) Graphical representationwhich facilitates a rapid flow of
information from the record to the human's brain (by sweeping
it by eye). The information is condensed and the inspection is
easy, but the accuracy is limited.
Graphically represented results are usually error patterns, error
curves, transfer functions, error density distributions, etc.
b) Semi-graphical representationin which some data are repre
sented numericallywith an arbitrary accuracy. Inspection
requires more care than in the case of pure graphs. This method
is frequently applied for the representation of residual errors
which are distributed in two dimensions.
c) Numerical representationwhich applies to singele numbers such
as standard deviations, r.m.s. errors, confidence limits, etc., to
groups of numbers as, for example, weight or covariance matrices,
and to arbitrary tables which are used for digital processing. The
accuracy is unlimited.
d) Mathematical formulaewhich represent the most condensed
information, (e.g. the error functions, transfer functions,.
Their inspection requires more effort. They are applicable as
correction formulae in digital data processing.
e) Verbal descriptionswhich are convenient for representing phe
nomena defined inaccurately, or of less important information.
