Download Paper in Word ()

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Ecological interface design wikipedia , lookup

Human-Computer Interaction Institute wikipedia , lookup

Artificial intelligence in video games wikipedia , lookup

Computer vision wikipedia , lookup

Human–computer interaction wikipedia , lookup

Expert system wikipedia , lookup

Collaborative information seeking wikipedia , lookup

Embodied cognitive science wikipedia , lookup

Intelligence explosion wikipedia , lookup

Personal knowledge base wikipedia , lookup

Ethics of artificial intelligence wikipedia , lookup

AI winter wikipedia , lookup

Philosophy of artificial intelligence wikipedia , lookup

Existential risk from artificial general intelligence wikipedia , lookup

Computer Go wikipedia , lookup

History of artificial intelligence wikipedia , lookup

Knowledge representation and reasoning wikipedia , lookup

Transcript
ARTIFICIAL INTELLIGENCE
AND CIM
Samir Sawant
Masters of Science Graduate Student
Submitted in Partial Completion of the Requirements of
INDEN 5303
Advanced Manufacturing Systems Design
This paper was developed to assist students in partial fulfillment of course
requirements. No warranty of any kind is expressed or implied. Readers of this
document bear sole responsibility for verification of its contents and assume any/all
liability for any/all damage or loss resulting from its use.
Table of Contents
Artificial Intelligence and CIM ......................................................................................... 1
Abstract ............................................................................................................................... 1
Introduction........................................................................................................................ 1
The Rationale Behind the Use of AI ................................................................................. 2
Evolution of AI ................................................................................................................... 3
Basic Elements of AI.......................................................................................................... 5
Knowledge Representation ....................................................................................................... 5
Rule based knowledge representation ..................................................................................... 6
Search Methods ......................................................................................................................... 7
Blind Search ................................................................................................................ 8
Heuristically informed search methods ...................................................................... 9
Reasoning and Planning ......................................................................................................... 10
Machine Learning ................................................................................................................... 10
Role of AI in CIM ............................................................................................................ 14
AI aids Concurrent Engineering............................................................................................ 15
AI in Planning and Scheduling .............................................................................................. 15
Machine Learning in Planning and Scheduling ................................................................... 16
Impacts of AI .................................................................................................................... 17
Conclusion........................................................................................................................ 17
Bibliography ..................................................................................................................... 18
List of Figures
FIG. 1 BASIC SEMANTIC NETWORK [1]…………………………………………………………………….. 5
FIG.2 DEVELOPED SEMANTIC NETWORK [2] ……………………………………………………………….6
FIG.3 GRAPHICAL REPRESENTATION OF “ANTECEDENT-CONSEQUENT RULES” [1]…………………….. 7
Page i
FIG.4 SEARCH TREE [1]……………………………………………………………………………………..8
FIG. 5 DEPTH-FIRST SEARCH [1]…………………………………………………………………………....9
FIG.6 BREADTH-FIRST SEARCH …………………………………………………………………………….9
FIG.7 NETWORK FRAGMENT [1]………………………………………………………………………… .10
FIG.8 THE ARCH [2] ……………………………………………………………………………………....11
FIG.9 INITIAL KNOWLEDGE REPRESENTATION [2] ………………………………………………………11
FIG.10 NEAR-MISS #1 [2] ………………………………………………………………………………....12
FIG.11 [2] …………………………………………………………………………………………………12
FIG.12 NEAR-MISS #2 [2]………………………………………………………………………………… 13
FIG.13 [2]………………………………………………………………………………………………….13
FIG.14 NEAR-MISS #3 [2] ………………………………………………………………………………....13
FIG.15 COMPLETE REPRESENTATION OF THE ARCH [2]…………………………………………………14
Page ii
ABSTRACT
Recently, complex Artificial Intelligence (AI) systems have found new applications in the
engineering industry, an area dominated by traditional analytic techniques. This paper
provides a brief introduction to the concept of Artificial Intelligence and discusses its role
in Computer Integrated Manufacturing (CIM). The paper is divided into two major parts.
The first part deals with the field of AI and explains the components of AI. The second
part contains a discussion on the applications of AI in manufacturing and its role in
Computer Integrated Manufacturing.
INTRODUCTION
Artificial Intelligence is an emerging technology that has recently attracted considerable
publicity. Many applications are now under development. One simple view of AI is that
it is concerned with devising computer programs to make computers smarter. Thus
research in AI is focussed on developing computational approaches to intelligent
behavior.
Artificial Intelligence can be defined as
The study of the computations that make it possible to perceive, reason and act [1].
From the perspective of this definition, artificial intelligence differs from psychology
because of the greater emphasis on computation and AI differs from most of the
computer sciences because of the emphasis on perception, reason and action. From
perspective of goals, artificial intelligence can be viewed as part engineering and part
science. The engineering goal of AI is to solve real-world problems using AI as an
armamentarium of ideas about representing knowledge, using knowledge and assembling
systems. The scientific goal of AI is to determine which ideas about representing
knowledge, using knowledge, and assembling systems explain various sorts of
intelligence [1].
Page 1
THE RATIONALE BEHIND THE USE OF AI
The computer programs with which AI is concerned are primarily symbolic processes
involving complexity, uncertainty and ambiguity. These processes are usually those for
which algorithmic solutions do not exist and a search is required. Thus, AI deals with the
types of problem solving and decision making that humans continually face in dealing
with the world.
This form of problem solving differs markedly from scientific and engineering
calculations that are primarily numeric in nature and for which solutions are known that
produce satisfactory answers. In contrast, AI programs deal with words and concepts and
often do not guarantee a correct solution – some incorrect answers being tolerable as in
human problem solving.
Another aspect of AI programs is the extensive use of “domain knowledge”. Intelligence
is heavily dependent on knowledge. This knowledge must be available for use when
needed during the search. It is common in AI programs to separate this knowledge from
the mechanism that controls the search. In this way, changes in knowledge only require
changes in the knowledge base. In contrast, domain knowledge and control in
conventional computer programs are integrated together.
As a result, conventional
computer programs are difficult to modify, as the implications of the changes made in
one part of the program must be carefully examined for the impacts and the changes
required in other parts of the program.
AI also complements the traditional perspectives of psychology, linguistics and
philosophy. This is justified by the following reasons

Computer metaphors aid thinking: Work with computers has led to a rich new
language for talking about how to do things and how to describe things.
Metaphorical and analogical use of the concepts involved enables more
powerful thinking about thinking.

Computer models force precision: Implementing a theory uncovers conceptual
mistakes and oversights that ordinarily escape even the most meticulous
researchers.

Computer implementations quantify task requirements: Once a program
performs a task, upper bound statements can be made about how much
information processing the task requires.
Page 2

It is simple to deprive a computer program of some piece of knowledge to test
how important that piece of knowledge really is. It is difficult to do this with
human brain [1].
Thus intelligent systems with AI incorporated in them can accomplish several complex
real life problems, a few of which are as follows

Solving difficult analysis problems involving calculus, non-linear dynamics,
differential equations and other complex mathematical concepts

Synthesis of
various factors to design devices and improving it my
performing iterative tests

Improving decision making by incorporating learning programs (based on
experience or data) that reason about new experiences in the light of common
sense knowledge (incorporated in the program) [3]
As a result of these capabilities of AI and its attainment of matured state in the recent past
the development of applications based on AI is motivated by increasingly by the business
people to achieve strategic business goals [7].
EVOLUTION OF AI
The term “artificial intelligence” was coined by John McCarthy in 1956 for a conference
called “The Dartmouth Summer Research Project on Artificial Intelligence” [2]. In early
stages of AI it was assumed that intelligent behavior was primarily based on smart
reasoning techniques and that bright people could readily devise ad hoc techniques to
produce intelligent computer programs. The major initial activity involved attempts at
machine translation. However this approach failed miserably because of factors such as
multiple word senses, idioms and syntactic ambiguities [3].
In 1961, Slagle at M.I.T. devised a heuristic computer program to do symbolic
integration. This proved to be the forerunner of a successful series of symbolic
mathematical programs used today at M.I.T. and by other AI researchers. The pioneering
work in computer vision was Robert’s (1965) program designed to understand block
scenes. It was based on the use of spatial derivatives of image density and utilization of
simple features such as the numbers of vertices, to relate the objects in the line drawing to
stored 3D models of blocks [3]. Inspite of these important breakthroughs the first 15
Page 3
years of AI remained restricted to game playing (helped understand machine learning)
and puzzles solving (lead to the development of problem solving techniques based on
search and reducing difficult problems into easier subproblems). The real problems
remained beyond the scope of the research then.
However the AI efforts of the 1950’s and 1960’s were not without merit. The research
revealed some important insights in to the development of AI concepts, which could be
stated as follows

It was found that expectation is a human characteristics of intelligence

Perception both visual and in language, is based upon knowledge, models and
expectations of the receiver

It was found that in communication and problem solving, lack of contextual
knowledge severely limit capability to communicate

Heuristics are necessary to guide a search to overcome combinatorial
explosion of possible solutions that pervade complex problems [3]
The research in the next decade capitalized on the above facts and new knowledge
representations techniques were developed. The search techniques began to mature.
There were a lot of interactions with other fields such as medicine, electronics and
chemistry. Feasible approaches were demonstrated for language processing, speech
understanding, computer vision and computer programs that could perform like experts.
Thus, the 1970’s found the AI research community developing the basic tools and
techniques needed, and demonstrating their applicability in prototype systems. Future
complex systems were proved feasible. The emphasis on knowledge, as essential to
intelligence, led to the subfield of
“Knowledge Engineering” associated with the
building of expert systems.
Based on the framework formed in 1970’s AI proliferated in the form of expert systems
and are increasingly used for commercial applications. In the expert systems area, DEC
reports that RI – a system designed to configure VAX computer systems – is already
saving millions of dollars [3]. Big players in respective industries have started using the
AI tools and investing funds in its research and development.
Page 4
BASIC ELEMENTS OF AI
KNOWLEDGE REPRESENTATION
Knowledge Representation is a set of conventions about how to describe a class of things.
A description makes use of the conventions of a representation to describe some
particular thing. A good representation of knowledge is the key to problem solving.
Semantic Networks
One common form of representation involves semantic networks. Suppose we want to
BIRD
is a
ROBIN
FIG. 1 BASIC SEMANTIC NETWORK [1]
represent the fact that “all robins are birds” in a semantic network. We can do this by
creating a simple graph in which the nodes (here ovals) represent the objects and the links
the “is-a” relation between them.
If “Clyde” is a particular robin and if we want to add to our knowledge that birds have
wings we can represent this information my adding nodes and relations between them.
This representation enables us to deduce facts that are not explicitly stated in the network
e.g. that robin have wings and so does Clyde since he is a robin. This feature is called
“property inheritance”. Knowledge in this form is further used for programming and is
called “object oriented programming”. Further if we want to show that “Clyde owns a
nest” it is represented as a node called “OWNERSHIP” which is a type of a
“SITUATION”. The ownership of Clyde “OWN-1” is an instance of “OWNERSHIP”
just as Clyde is an instance of robin.
Page 5
has part
WINGS
BIRD
SITUATION
is a
is a
ROBIN
OWNERSHIP
is a
NEST
is a
is a
owner
CLYDE
ownee
OWN -1
NEST-1
FIG.2 DEVELOPED SEMANTIC NETWORK [2]
RULE BASED KNOWLEDGE REPRESENTATION
Knowledge can also be stored in the form of rules like the following, each of which may
contain several if patterns and one or more then patterns:
Rn
If
If1
If2
.
.
then
then1
then2
.
.
A statement that something is true, such as “Stretch has long legs,” or “Stretch is a
giraffe,” is an “assertion”. In all rule-based systems, each if pattern is a pattern that may
match one or more of the assertions in a collection of assertions. The collection of
assertions is called “working memory”.
Page 6
In many rule-based systems, the then patterns specify new assertions to be placed into
working memory and the rule-based system is said to be a deduction system. In deduction
systems the convention is to refer to each if pattern as an “antecedent” and to each then
pattern as a “consequent”.
Antecedents
Consequents
FIG.3 GRAPHICAL REPRESENTATION OF “ANTECEDENT-CONSEQUENT RULES” [1].
Sometimes, however the then patterns specify actions, rather than assertions – for
example “Go to stores”- in which case the rule based system is said to be a “reaction
system”.
SEARCH METHODS
Search is a very important tool for problem solving, inference, planning and reasoning.
But before trying to understand about various search methods it is important to
understand the search tree.
Page 7
current state
A
B
C
D
H
E
I
J
F
K
L
G
M
N
O
target state
FIG.4 SEARCH TREE [1]
A search tree is a representation that is semantic in nature where the nodes denote the
states and the connection between the nodes denotes transition between states. If a node
has b children, it is said to have a branching factor of b. If the number of children is
always b for a non-leaf node, then the tree is said to have a branching factor of b. The
search is complete when the state of a child node matches the target state.
The search could be a “blind search” or guided by heuristic quality estimates.
Blind Search
In a blind search it is implicit hat one path at a node is as good as any other. There are
two types of blind searches such as
 Depth-first search
 Breadth-first search
In Depth-first search the search works as shown in Fig.5. In this search alternatives at the
same level are ignored completely and lower levels are explored for the target state. In
the Breadth-first search depicted in Fig.6 the nodes at a level are checked before going to
nodes at a lower level.
Page 8
current state
A
B
C
D
H
E
I
F
J
K
G
L
M
N
O
target state
FIG. 5 DEPTH-FIRST SEARCH [1]
current state
A
B
C
D
H
E
I
J
F
K
L
G
M
N
O
target state
FIG.6 BREADTH-FIRST SEARCH
The Depth-first search is a good idea when one is confident that all partial paths either
reach dead ends or become complete paths after a reasonable number of steps. Breadthfirst search works even in trees that are infinitely deep or effectively infinitely deep.
Both the above methods evaluate every node for the target node and hence are inherently
time consuming. But Breadth-first search is wasteful when all paths lead to the goal node
at more or less the same depth. Moreover if the branching factor is large there is an
exponential explosion which again consumes a lot of time and may be intractable.
Heuristically informed search methods
Search efficiency may improve spectacularly if there is a way to order the choices so that
the most promising is explored first. This gives rise to “Heuristically informed search
methods”. There are 2 types of searches in this category
Page 9

Hill Climbing Search

Beam Search
The “Hill Climbing Search” is similar to Depth-first search, except that the choices are
ordered according to some heuristic measure of the remaining distance to the goal. The
“Beam search” is like Breadth-first search but unlike the later the beam search moves
downward only through the best w nodes at each level; the other nodes are ignored.
Consequently the number of nodes explored remains manageable and prevents
exponential explosion.
REASONING AND PLANNING
Reasoning and planning deals with the issue of finding the solution. Consider the
example discussed in section on semantic network. Suppose we want to answer the
question “What does Clyde own?” using the knowledge representation in the form of
OWNERSHIP
is a
owner
CLYDE
ownee
OWN -?
?
FIG.7 NETWORK FRAGMENT [1]
semantic network developed earlier. The question is restructured into a network fragment
as shown in Fig.7
The next step is to look for a similar structure (of nodes and links) in the semantic
network. This process is performed by a pattern matcher. Once the pattern matcher
performs the task successfully the solution is determined and in this case it is “NEST1”.
MACHINE LEARNING
This section deals with making the computer “understand”. Suppose that we want the
computer to learn to understand the concept of an arch shown in Fig.8:
Page 10
FIG.8 THE ARCH [2]
One method of teaching the concept of arch involves first presenting to the computer an
accurate realization of the concept, followed by a number of near misses – realizations
such that each deviate in one particular aspect from the arch in a “didactic” sequence.
This sequence of presentations enables the program to improve its representation in a
stepwise fashion. It is assumed from the prior knowledge of the fact that an arch consists
of three elements (X, Y, Z) all of which are blocks:
is a
is a
X
BLOCK
Y
is a
Z
FIG.9 INITIAL KNOWLEDGE REPRESENTATION [2]
Further more it is assumed the program knows various categories for which it must look,
such as support, touch and orientation, which it can then add to its representation by
corresponding links. If we now present to the program this near-miss representation:
Page 11
FIG.10 NEAR-MISS #1 [2]
and inform the program that this is not an arch, it will realize from comparison between
this and the original that in an arch two blocks are standing (vertical) and one is lying
(horizontal), and improve its representation accordingly:
STANDING
has orientation
is a
is a
X
has orientation
BLOCK
is a
Z
has orientation
LYING
FIG.11 [2]
If the next near-miss is as follows:
Page 12
Y
FIG.12 NEAR-MISS #2 [2]
and inform the program that this is also not an arch, it recognizes the difference related to
“support” and adds two more links to its representation:
STANDING
has orientation
is a
is a
X
BLOCK
Y
is a
supports
Z
has orientation
LYING
FIG.13 [2]
Finally, the next near-miss is presented:
FIG.14 NEAR-MISS #3 [2]
Page 13
it will realize that, in an arch, the two standing blocks must not touch. We thus arrive at
the final representation of an arch, which would correspond to the arch originally
depicted above [2].
STANDING
has orientation
is a
is a
X
has orientation
BLOCK
Y
is a
Z
has orientation
LYING
must not touch each other
FIG.15 COMPLETE REPRESENTATION OF THE ARCH [2]
ROLE OF AI IN CIM
The idea of Computer Integrated Manufacturing (CIM) stresses its global strategic goal:
to integrate all company activities into a unified management structure exploring a largescale hierarchy of computers. Recent innovations in the field of CAD, CAM, CAPP,
CAQ enabled due to the development in software technology and hardware have
improved the productivity of the processes considerably. But these developments have
enabled solving partial tasks of the entire goal. The ambitious targets of CIM emphasize a
complete integration of computer-aided activities in the factory as well as the use of the
knowledge-based technology within the entire information processing. Unified behavior
and intelligent access of particular subsystems imply the need for integration of the
technical, managerial, and business activities [4].
Page 14
Reasonable functionality of CIM requires development of sophisticated AI - based
techniques because the complexity of the goals usually faces the designer with
computationally intractable problems or decision making [6]. Within the frame artificial
intelligence, the central role of problem specific (often heuristic as well) knowledge for
problem solving has been discovered. Besides the expert systems (as software products)
AI provides a wide spectrum of methods and techniques (state space search, space
pruning by heuristic knowledge, theorem proving, uncertainty processing etc.) which
may be embodied into various software products, thus influencing their structure and
enhancing the performance.
AI AIDS CONCURRENT ENGINEERING
AI deeply influences the progress in CIM area more than it was expected. CIM
necessitates a lot of interaction between various functions of an organization and hence a
great amount of information flow and usage is unavoidable. AI provides an important
concept called “feature based techniques” as a methodology to integrate product design
with other stages of the product life cycle. The features are defined as form features
associated with the semantics of their engineering design. These features integrate
geometry, technology and function aspects and are treated as objects in CAD / CAM
engineering tasks.
AI also provides a solution to computerized concurrent engineering. Each application of
the concurrent engineering requires the exact formulation of subtasks to be solved
parallel. The AI inspired constraint-handling approach is used for this purpose [4]. The
constraint demarcating the local problem solving spaces may be considered as objects in
the Object Oriented Programming environment and organized in networks, which ensure
the appropriate constraint-propagation [4]. Usually the individual subsystems in
concurrent engineering are handled in a computerized modular way but the
communication among them is based on multi- agent techniques of distributed AI [4].
AI IN PLANNING AND SCHEDULING
Planning and scheduling play an important role within CIM and has great impact on the
overall corporate efficiency. The decision making involved in these functions can be
found in the task of estimation of delivery dates, in resource allocation, circumvention of
machinery failures and in many other cases. The task is extremely difficult even from the
Page 15
mathematical viewpoint. Most of these tasks include problems, which are NP-complete
and thus intractable. The main difficulty with the application of the OR methods for NPhard problems is in the necessity of explicit specification of all the constraints and utility
functions. In real life tasks, the specification represents many thousands of equations and
inequalities on hundreds of discrete variables. The AI approach to this problem is based
on state space search based on knowledge-based and logic-based reasoning and provides
a possible solution to this problem [8].
MACHINE LEARNING IN PLANNING AND SCHEDULING
The concept of machine learning is based on the biological concepts of evolution and
adaptability. Recent research in the AI field has given rise to the development of genetic
algorithm, which is based on the concepts of evolution of chromosomes. Genetic
algorithms are used to find optimal schedules. Besides these algorithms Case-based
reasoning is the intrinsic method used in CIM [4].
In addition to this AI is extensively used in machine tools and other plant facilities thus
enabling efficient use of resources. An example of this is the tool-monitoring system used
by aircraft industries. The main concern here is preventing rejection of the expensive part
due to tool-breakage. The above objective is accomplished by established by obtaining a
tool signature, which includes a graph indicating the variations in the torque during the
entire cycle time for the part with a new tool. This tool signature is then compared with
other tools in operation and the machine is stopped on reaching a predetermined torque
and vibration level [8].
Thus it can be concluded that AI has a major role to play in CIM as it not only provides
isolated methods/solutions but also provides completely new way of handling and
processing of information required for the integration of various functions.
Page 16
IMPACTS OF AI
AI is a new field, which has radically changed the way we thought about technology. The
power of AI is so great that it has even affected our day to day life. We are now using
artificial intelligence devices like household washing machines with fuzzy logic. The
impacts of the AI can be enumerated as follows

The monotonous work requiring some reasoning can now be relegated to
smart computers leaving human beings for more creative work. This will lead
to a increase in the “third generation” jobs involving creativity.

AI has brought along with it the concept of “knowledge” which is something
more than just information. This has lead to the concept of “knowledge
trading”. The exchange of knowledge has now enabled innovative practices in
the business, which have lead to cost reductions and shorter lead-times.

The use of AI by the Defense departments has enabled development of
efficient but more vulnerable weapons system.

AI has also improved the computer surveillance

AI can also be used in the field of education for computer based training.
CONCLUSION
Artificial intelligence is of great value where decision making in symbolic processes is
involved. Artificial intelligence is also a very effective tool to perform highly
complicated tasks involving a lot of analysis and search. The use of AI has assumed
widespread proportions although it is less conspicuous. Artificial Intelligence will
continue to be a part of our lives for quite sometime.
Page 17
BIBLIOGRAPHY
1. Winston H. Patrick, Artificial Intelligence, Addison Wesley, Massachusetts, 1992.
2. Trappl R., Impacts of Artificial Intelligence, North-Holland, Amsterdam, 1991
3. Gevarter B. William, Artificial Intelligence, Expert Systems, Computer Vision
and Natural Language Processing, Noyes Publication, New Jersey, 1984
4. Marik V. et al "The AI impacts on CIM", IEEE Journal, CH2881-1-1/90/0000-1284
1990
5. Boden A. Margaret, Artificial Intelligence, Academic Press, London, 1996
6. Walters M.J.Helena and Schtaklef G. Roger, “Commercial Benefits of AI
applications in CIM: Value Analysis Approach”, IEEE Journal, 087942-6004/90/1100-0767,1990
7. Nilsson. J. Nils, Principals of artificial Intelligence, Tioga Publishing, Palo Alto
California, 1980
8. Famili A. Fazel, Nau and Kim H. Steven, Artificial Intelligence Applications In
Manufacturing, AAAI Press/ The MIT Press, California, 1992
Page 18
Page 19