N. A. Technologies Company
has developed a new type of expert system, as a result of
our nearly 14 years of experience developing expert systems
for industrial and government customers. We call it a Dynamic
Expert System, and the software is called DYNEX. We have
found that many expert systems for industrial application,
in order to be useful, consisted of several hundred, or
even several thousand rules. These systems are very useful,
but can become very complex to maintain and to update. Our
approach in developing the DYNEX system was to substantially
simplify the overall structure of the software, and make
it very easy for the end user, or a designated individual,
to maintain and update the system.
DYNEX has an internal
structure that stores all of the rules in a common format,
so that the rules themselves are easily organized into a
database. The rule format is also simplified so that each
rule has a simple yes/no or on/off truth value to the antecedent,
and the consequent then is easily nested into a tree structure
that can be visualized on screen both by the user and by
the individual responsible for maintaining the system. Standard
Microsoft conventions are used, so the tree structure is
shown in a Microsoft viewer, and both the system user and
the maintainer can easily visualize the structure - simple
to use and easy to maintain.
Attached to the expert
system is a multi-media database that is tied to the rule
system by a key word structure. The database can store images,
graphics, video, audio, and other media. Additionally, the
database is easily linked to all of the standard Microsoft
data types and application software. Thus, it is easy to
automatically provide the user with access to a rich environment
of information resources. Thus, each rule can have information
resources attached to the antecedent - this makes it easy
for the user to get helpful information in order to provide
the system with answers to determine the truth value of
the antecedent. Additionally, the consequent of each rule
also is linked to the database, and can provide the user
with an additional information resource about the conclusions
made by the expert system.
Background - Expert System Technology
The representation of
knowledge, or the study in the field of Artificial Intelligence,
has provided several useful technologies. One of the most
successful is the Expert System. In it's commonest form,
the expert system consists of a series of rules and "inference
engine" software to operate on those rules in a logical
manner. Using this technology, it is possible to capture
the knowledge of one or more individual "experts" in a subject
matter and provide that expertise to others in the form
of a computer system. The knowledge of the experts is contained
in the rules, which are written in the form of "If … [an
antecedent], then … [a consequent]". The antecedent is a
condition which must be true if the result, known as the
consequent, is to be determined to be true. Generally this
is an easy format for capturing knowledge from a human expert.
An example of such knowledge
which could be applicable to the maintenance of welding
systems in a manufacturing plant, might be: "If the weld
arc is showing signs of increased instability and it has
been more than four hours of operating time since the wire
contact tip was changed … Then the wire contact tip may
need replacement". Thus, by querying the user of the system,
to determine the level of stability of the arc plasma (or
an on-line monitoring system) the knowledge of the human
expert, who knows from experience that tip wear causes arc
instability, is provided to the non-expert. The software
would attempt to determine the truth quality of the antecedent
(… arc has indications of instability and more than fours
hours since the tip was changed …). If the antecedent is
true, then the wire contact tip may need to be changed.
The truth value of the antecedent is determined from the
consequent or result from other rules, or may be a combination
of inputs from an on-line monitoring system, the human user,
and the result of other rules. The inference engine has
the task of determining which rules are applicable to a
specific session, and what their truth value is.
Through a set of logical
manipulations in the software, the non-expert user is informed
about the software's "best estimate" of the situation. Through
links to the multi-media database, video, audio, graphics,
and text information can be supplied to assist with understanding
the questions from the system, or the actions needed to
correct the problem.