Towards Data Democracy Beyond Fukushima - Principia Designae: Pre-Design, Design, and Post-Design - page 99

object obtained by observations and/or measurements by analytic prolongation and
singularity treatments to cover unknowns by knowns.
As the value space of users reflects many factors associated with users, it implies
human dimensions anyway. Thus it has been discussed as general and popular
subjects in humanities, philosophy, economics, engineering and so on. However the
value space on materials can be almost equivalently described in terms of properties
by correlating them to a value or function in a system where materials are used. And
the fact space on materials can or should be described in terms of data, logics and/or
physics following constituent principle: the basic constituent of matter are various
kinds of identical particles, with causality, covariance, invariance and equi-
probability principle (Ni 2014 ). Moreover in case of materials, the fact space can
be associated with the structural space of structural data of the constituents with
finite patterns and their derivatives endorsed by geometry, for example, crystallog-
raphy, we can rewrite materials design procedures to find mappings between the
fact space and the structural space which can be associated with the value space by
so-called structure-property correlation.
On the basis of these frames, materials design under development can be
summarized as procedures by data-intensive ways including three loops for learn-
ing, namely, learning how to learn (discovery approach to find better mappings
through verification and evaluation on each model derived systematically
from data (Villars et al. 2008 )), changing the rules (LPF approach by linking
data associated with structural data including first principles models and cluster
variation methods) (Chen et al. 1996 ) and following the established rules
(total quality control of data by such established rules as geometry, symmetry,
valence and so on), and we call the strategic combination of three loops as
data-driven design. As far as the patterns to link associated data are finite, we
can reach design solutions anyway by adding data step by step confining singular-
ities at boundaries of solution spaces as shown in Fig. 8.1 . Combinations and
arrangement of constituents are classified by finite patterns (yellow dots and blue
dots in Fig. 8.1 for example) which are associated with derived numerical values
computed systematically from attribute values of constituents. Finding the best
computation corresponds to finding the best mapping between the value space
yellow dot dominant or blue dot dominant regions here and the fact space
described by the attribute values of constituents. This method of discovery is
explained as inverse problem solution for models, which is equivalent to finding
design methods. In short it is a process of learning how to learn something from
data systematically based on a creation of quasi-universal set prepared as one
window of high quality almost exhaustive data.
Further design procedures of materials to meet engineering requirements are
carried out step by step by taking into account property changes due to microstruc-
tural variations, defects, aging, service conditions and so on adaptively. And they
are processed in design procedures of engineering products.