Rule-based expert systems operate through a set of "IF....THEN" rules processed by an underlying "inference engine". A typical rule-based expert system is composed of four major elements: the Inference Engine, a Knowledge Base, a User Interface and an Explanation Facility.
The Inference Engine is that part of the expert system that performs the reasoning. It is analogous to the raw intelligence of a human expert. Many different forms of inference engines exist, but all are designed to perform the same task, i.e., to examine the current facts and use available rules to generate new facts.
The Knowledge Base is where the information resides within the expert system. It consists of two distinct parts: the rule base "IF <condition> THEN", and the fact base containing simple statements about the condition of the world, as it is applicable to the problem under study.
The User Interface enables the expert system and the user to communicate. The exact form of this interface depends on the intended audience for the expert system.
The Explanation Facility presents the user with the expert system's justification for its conclusions, i.e., an audit trail, as necessary.
A typical expert system initially partitions the
problem by applying a broad set of inference rules to an initial set of data
describing the problem or the symptoms. Each of these inference rules will
take the inference engine to a further data-acquisition stage (typically
another, more directed, questionnaire) or the establishment of a new fact.
This process of a directed search with additional data gathering continues
until the expert system has reached a leaf node in the resulting decision
tree. Some inference engines may resolve an ambiguity, when several inference
rules evaluate as TRUE to a given data set, by selecting the one with the
highest associated weighting or confidence factor; others may use a different
approach (e.g., fuzzy logic --see 22.214.171.124.2
The rules in the knowledge base, that portion which drives any expert system, are painstakingly constructed by an expert systems specialist interrogating the knowledge expert and subsequently codifying the often imprecise descriptions of their thinking processes into inference rules, possibly with numerical limits. For example a rule for a medical diagnostic expert system may state:
"IF heart rate > 100 beats per minute AND body
temperature > 101°F
THEN recommend that patient be placed in an ice bath".
The fact portion of the knowledge base would simply record the patient's heart rate and temperature.
A general approach for the physical development of a maintainability expert system is shown in Figure 9.
It may be difficult to capture all of an expert's knowledge in an expert system knowledge base because the expertise is encoded as a causal relationship. "Rational" knowledge, where the solution can be described analytically, is comparatively straightforward to codify into inference rules. "Semi-rational" knowledge, where the expert can specify suitable ranges for conditions, but cannot (easily) defend the choice of these ranges are more
difficult. This process may take some detective work by the expert system specialist. Unfortunately, however, much of what "makes an expert" occurs at an intuitional or visceral level, where even the expert is unaware of the underlying mechanism behind their decisions and may even be unable to quantify appropriate ranges. This area presents the major challenge and limitation in the design of a rule-based expert system. The following three sections will address some alternative solutions to this problem.
Figure 9 -
Steps in a General Approach for the Physical Development of a Maintainability