data base, so it can be accessed by the expert 
system. The gathering of the information is 
typically done by a knowledge engineer. It is 
his job to collect the information from the 
human expert through a series of interviews. 
These heuristics (rules of thumb) are 
translated into "if then" statements (knowledge 
representations). The rules are used by the 
expert system's inference engine. The 
inference engine consists of algorithms 
(solution methods) that correlate rules and 
facts and return information to the user. 
Figure 1 shows a block diagram of an expert 
system. 
  
  
  
THE THE INPUT 
INFERENCE KNOWLEDGE FACTS & RULES 
ENGINE REPRESENTATION 
THE USER 
INTERFACE 
USER 
  
  
  
  
Figure 1 
An Expert System 
The user interface is typically through a 
computer terminal (CRT and keyboard/mouse 
combination). It is designed to be easy to 
interpret by non-expert users. One way this 
can be achieved is with the use of menus. 
Menus present the user with questions and 
multiple choice answers. The user then selects 
one or more (where allowed) of the given 
choices. Numerical value data can be entered 
directly on the computer keyboard. 
Depending on the nature of the expert system, 
the output can be in the form of conclusions or 
directions to solve a given problem. The 
output can be in a form of easy to understand 
instructions with further explanations and 
illustrations through computer graphics. 
Expert systems are typically applied to solve 
complex problems through non-expert users. 
This tends to put a heavy burden on the expert 
system user interface. This is true both in 
the way the questions are presented to the user 
for proper interpretation and response, and 
later to enable the user to understand the 
system's final conclusion or instructions. 
THE TOOLS FOR EXPERT SYSTEMS 
SOFTWARE TOOLS: There are at least two ways 
one could set out to create an expert system 
program. One way is to utilize one of the AI 
computer languages such as LISP, PROLOG, or 
even C (a non-AI language). He would write 
from scratch the knowledge representation, 
inference algorithms, and user interface 
portions of the program. This method would 
provide flexibility but could become expensive 
due to the cost of engineering development. 
The second method is to start with a "shell." 
A shell is an existing computer program with 
the building blocks of an expert system, the 
knowledge representation tools, the inference 
engine, and tools for user interface, built in. 
But, it lacks the knowledge. The knowledge is 
then programmed in by the knowledge engineer. 
The use of existing shells can prove more cost 
effective in the long term since the 
development cost of the shell is spread over 
many users of that shell. 
HARDWARE TOOLS: One reason for slow widespread 
acceptance of artificial intelligence has been 
the cost of software and hardware tools. Until 
recently, most AI packages required powerful 
main frame computers to operate. In the past 
five years, more software tools, such as expert 
system shells that can run on personal computer 
work stations, are becoming available. This 
will make the expert system a much more viable 
entity as an industrial tool since the cost 
will be far less prohibitive. 
EXPERT SYSTEMS AND MACHINE VISION 
The field of machine vision also falls under 
the topic of artificial intelligence. Most 
vision systems, however, utilize conventional 
computer programs. Therefore, whether they can 
be categorized under AI is questionable. AI or 
not, applying machine vision can produce some 
great challenges in applications engineering. 
One area of machine vision applications where 
there are often engineering difficulties is in 
the front end, that is, the lighting, optics, 
and sensor selection. Many good applications 
have failed due to improper front end design. 
Therefore, machine vision lighting and optics 
was selected as the topic of an expert system 
development. The following pages describe the 
creation of an expert system and how it is used 
to solve lighting and optics problems for 
machine vision applications. 
THE LIGHTING AND OPTICS ADVISOR 
THE OBJECTIVE: The Lighting Advisor expert 
system was created to help solve lighting and 
optics problems for a specific group of machine 
vision applications, namely, small parts 
assembly verification. The field of machine 
vision encompasses a large area of possible 
applications, such as, assembly verification, 
electronics, packaging, non-contact gauging, 
etc. It was decided that, at least initially, 
the Lighting Advisor should be limited to one 
group of applications. This expert system has 
been designed so that end users with little or 
no lighting and optics background can use it. 
HOW IT WORKS:  Transferring human knowledge 
into a form acceptable by the expert system 
shell proved challenging. The process became 
easier as the human expert could see how his 
knowledge was being transformed into facts and 
rules of logic for machine use. There are over 
three 
expert 
is bei 
  
  
  
The Li: 
as its 
driven 
user t1 
applica 
and mak 
2. The 
  
  
[7