T
CD RO «€
fv we
ships between these network elements can be represented
through the definition of class->frame and frame->sub-
frame relations.
Firstly, classes (denoted by O) are defined for each of the
four categories of elements. Associated with each class is
a set of slots which described it’s characteristics. Impor-
tantly, each time an instance (denoted by A) of a class is
created, that instance (frame) inherits the class slots. In
Figure 8a, as a simple example, the class OPT (for object
points) possesses the slots X, Y, and Z. Initially, frames
Pt, and Pt;,; are not attached to a class. By making Pt,
and Pt,, ; instances of OPT they inherit the slots of the
class (Figure 8b). Of course, the values of these slots can
be uniquely set for each individual frame.
O OPT O OPT
(X, Y, Z) (X, Y, Z)
A Pt, AP, À Pt; A Pt,
X=1 X=4
Y=2 Yz5
Z=3 Z=6
(a) (b)
Figure 8 Property inheritance with frames.
Secondly, frame-subframe relations are defined to repre-
sent the relations between network elements belonging to
different categories. In such cases, the subframe is recog-
nised as a component of the frame, but does not inherit
it's slots. For example, in Figure 9 image point Imgpt, is
an instance of the class /MGPT (for image points) and a
sub-frame of object point Pt, representing the status of
the image point as an observation of the object point.
O IMGPT O OPT
(x. y) ena
+ tj
d eR
x=0.1
yz02
Nx»
uuu ^
WN ~
Figure 9 Frame-subframe relations.
The combination of both types of frame relationships
permits the network structure (Figure 7) to be accurately
represented by frames, as illustrated in Figure 10. Note
here that it is necessary that the ES only have permanent
knowledge about the classes and possible frame->sub-
frame relations. As a result the same ES can design any
network simply by dynamically creating the necessary
network elements and their relations in the frame repre-
sentation as each design proceeds. Permitting this flexi-
*This point is one of the major reasons for in-
cluding a CAD component in CONSENS.
bility is pattern matching, which allows the structure of a
frame representation to be reasoned with in rules. In pat-
tern matching, all instances of a class, or components
(sub-frames) of a frame, are referenced in the condition
or action of a rule. As is exemplified in the next section,
this is very useful in a task such as network design where
not only the number of elements and relations varies
from design to design, but many of the reasoning steps
need to be applied over classes or groups of elements.
O STATION O IMAGE O IPT O OPT
A Stn;
Almgpt; Almgpt, 1
Figure 10 Frame-representation of network data. Dashed
lines indicate class membership; full lines in-
dicate object-sub-object relationships.
4.2 Example: Representing Heuristic Design
Knowledge with Rules
A rule is a chunk of knowledge that represents a situation
and its immediate consequences. Rules are expressed as
condition-action, (i.e. IF-THEN) statements. IF all the
conditions of the rule are true, THEN the rule’s hypothe-
sis is confirmed and any actions associated with the rule
are triggered by the ES’s inference engine. When at least
one of the conditions is not true, the hypothesis is false.
Rules are often appropriate for the representation of heu-
ristic knowledge. Consider, for example, the network di-
agnosis heuristics discussed in Section 3.4. In
formalizing the decision tree (in Figure 6) derived from
these heuristics, the rules listed below might be written.
Note that Rule 1 only provides for control by directing
the ES to diagnosis as soon as new performance meas-
ures for the network have been computed.
Rule 1:
IF new performance measures have been computed
THEN start diagnosis
Rule 2:
IF the hypothesis start. diagnosis is TRUE &
lOPTI».Num, rays »—- 4
THEN all points sufficient rays
Rule 3:
IF the hypothesis start diagnosis is TRUE &
IOPTi.Num, rays « 4
THEN some points insufficient rays
AND Add these IOPTI to class IUNRELIABLE PTI
