“bl -
from this examplei For- a factual data base -- e.g. on. me-
tals properties - this work has to be done in advance. So
pairs of values of alloy designation and heat conductivity
must be extracted from the documents for data input. But
this may be not sufficient for the following reason:
In a document the whole context for which a value is valid
can be reported in free text form. In a factual data base
the documentation of the context must be provided for in
the data base structure. The heat conductivity e.g. de-
pends on the temperature and the chemical composition (fi-
gure 1), and an individual value may be influenced by the
measuring procedure.
From this the question arises what is the correct value of
the heat conductivity of say pure copper (how pure? There
is no absolutely pure MLS Different values will be
found in the literature. Shall all values be documented or
one validated value only. Who is responsible for the vali-
dation?
And what about an alloy designated by its name? Such an
alloy has an allowed range for its chemical composition.
Therefore samples of the "same" alloy with a different
chemical composition must have different values of the
heat conductivity in addition to measuring uncertainties.
The result of all these questions is that in the design
phase of a factual data base a thorough analysis has to
be performed and some decisions have to be made according
to the purpose of the planned information system:
- Which types of objects and which properties of them
shall be containd in the data base?
- Which influencing factors on a property have to be
documented?
- Shall the data base contain raw data or validated
data?
- If validiation: Which type of value is needed for
each property ("true" value, upper and/or lower
limit, mean value); how is it performed; how is
it adapted to future developments?
- Is there a need for further processing of retrieved
data and which kind of processing?
If a comparison is drawn with literature data bases the
following is valid in general for factual data bases:
- The scope is relatively narrow.
- The input is relatively expensive.
- The output is more specific to certain tasks.
- The system can easier be extended to new types of
objects or to new methods of data evaluation.
Dahte 2