Upload
tadittti
View
220
Download
0
Embed Size (px)
Citation preview
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 1/40
Adama Science and Technology University
School of Engineering and Information TechnologiesDepartment of Mechanical and Vehicle Engineering
Post Graduate program
Manufacturing Engineering
Artificial Intelligence in Mechatronics System
Presented by: Tadele Belay
November 2011
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 2/40
Description Definition of Artificial Intelligence
Features and Characteristics of AI Fundamental technologies of AI
Applications of AI in manufacturing
system
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 3/40
Objectives
be aware of the distinction between strong and weak AI;
be aware of the concepts of human-compu ter syste ms and human-compu ter interaction;
know the benefits of building more intelligence into machines;
realize that there are currently limits to how much intelligence can bebuilt into machines;
know the five critical features of AI and the ten enduringcharacteristics suggested by Schank;
be aware of how AI can be implemented in manufacturing systems.
Understand the technologies of artificial intelligence Know the characteristics of intelligence of manufacturing system
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 4/40
Introduction The massive use of digital technology is assured by its
continuous improvement in performance/price ratio.
=== > every eighteen to twenty four months chip density
and hence computing power doubles while cost remains
constant.
We can expect therefore physical objects, including
living systems, to be tagged and thus endowed with the
ability to communicate with each other, opening
extraordinary opportunities for advanced Mechatronics. Artificial Intelligence (AI) has matured and is now
capable of providing innovative solutions to many
practical problems where there is a need to replace
automation with intelligence.
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 5/40
What is Articial Intelligence?
Some definitions that have been proposed:
1. Systems that think like humans
2. Systems that think rationally3. Systems that act like humans
4. Systems that act rationally
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 6/40
STRONG AI, WEAK AI AND COGNITIVE SCIENCE
AI has two main camps:
The proponents of strong AI believe that it will bepossible to build machines with human-like intelligence.
By contrast, the proponents of weak AI believe that
machines can exhibit what might be called intelligentbehavior, but that there are limits which mean that
machines will always be intellectually inferior to humans.
Within the weak AI camp there is an active research
area called c ognitiv e sc i enc e which uses computers tomodel human behavior with the intention of learning
more about human beings.
A very important aspect of this is the relationship between
humans and computers.
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 7/40
HOW MUCH INTELLIGENCE CAN BE BUILT INTO MACHINES
The simple answer to this question is that no one
knows.
N o mac h ine works wit h ou t some level of human s upervision.
In other words, all mac h ines are part of a human-compu ter syste m.
Suppose the intelligent part played by machines in any
particular system can be quantified between 0% (nomachine intelligence) and 100% (no human intelligence,
which we claim is impossible).
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 8/40
The cost of increasing the machine component of the
system's intelligence can be sketched as shown
The running costs of systems c ombining human and machine
int ell igenc e d ec r ease w it h t he introduc tion of greater
machine int ell igenc e, but t he c apit al c osts inc r ease
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 9/40
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 10/40
Alan Turing (1912-54), one of the founding fathers
of machine intelligence, realized right from the
outset that for a machine to become 'intelligent' it
would have to:
l earn, and to generalize from specific
information in order to acquire new knowledge.
know what is possible and choose between the
options.(Prediction)
detect that mistake and recover from it
constantly seeking to know 'what's out there
(curiosity)Although there is a substantial body of applicable
knowledge arising from AI, the subject remains
young with many more questions than answers.
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 11/40
ARTIFICIAL INTELLIGENCE & INTELLIGENT MANUFACTURING SYSTEMS
Intelligent manufacturing systems are those performing themanufacturing functions as if the human operators are doing the
job
However, none of the intelligent systems is capable of replacing their
human counterparts.
They are designed to support the operators, not to replace them.
They can provide knowledge and respective facts filtered out from
the huge amount of stored knowledge and increase the speed of
decision making capability of the operators.
They can utilize decision-making capability of computers and
make it available to the operators.
Since it is necessary to equip the systems with intelligent capabilities in
order to accomplish this, it would be wise to briefly review AI
technologies at this point.
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 12/40
TECHNOLOGIES OF ARTIFICIAL INTELLIGENCE
Some AI techniques, that are popularly known are: expert systems;
artificial neural networks;
genetic algorithms;
fuzzy logic; and
intelligent agents.
Intelligence is to perceive the environment in which
the system is operating, to relate events taking placearound the system, to make decisions about those events,
to perform problem solving and generate the respective
actions and control them.
In order to achieve these, the computers or systemsshould be equipped with the capability of planning,
learning, reasoning, monitoring, control, etc.
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 13/40
EXPERT SYSTEMS
are widely used in most of the industrial applications.
make it possible to create intelligent software systems capable
of solving problems in the same way as human experts would do
when facing the same problems.
Similarly, to human experts, they utilize their knowledge,
experience and talents. This is possible when respective knowledge
is stored in a specific format understandable to the computer.
The knowledge, which is represented in machine-readable format,is stored in so-called knowledge bases. It is utilized by expert
systems for a decision-making process similar to that of a human
being in order to produce solutions to the problems
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 14/40
Basic components and architecture of an expert system
Expert systems are proven to be useful especially in:-
planning; control; maintenance; repair; diagnosis; monitoring;
and inspection, etc.
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 15/40
K nowledge acquisition: acquire and elicit respective
knowledge from the expert and represent that to the
computer in a machine-readable format. This process iscalled knowledge acquisition.
K nowledge base is the place where acquired
knowledge is stored. Traditional ways of knowledge
representation may include using IFTHEN rules,knowledge frames, classes and procedures. The
knowledge base is populated with the domain knowledge
using these formats.
Inference engines search and scan the knowledge
base, filter the required knowledge out and reason about
those in order to be able to solve the problem posed by
the user through the user interface.
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 16/40
There are two general approaches for inference:
Forward chaining: this starts with the facts of the domain and tries to achieve a solution
through available knowledge in the knowledge
base.
Backward chaining: this starts with a
solution and tries to find out respective facts
supporting that particular solution.
User interface handles the communication
between the expert system and the users.
It may explain how and why certain
conclusions are reached.
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 17/40
ARTIFICIAL NEURAL NETWORKS
are mainly designed to perform learning.
generate new knowledge through learning the
manufacturing events using the examples.
are formed by hierarchically connected process elements
capable of information processing. Each connection
between the processing elements has a weight value
indicating the effect of one processing element on the
others.
These weight values, once the network is trained, are
believed to indicate the knowledge of the network. The main purpose of the learning is therefore to find out
the right weight values of the connections. Since the weight
values are distributed to the overall network, the neural
networks are assumed to have a distributed memory.
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 18/40
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 19/40
They have various characteristics some of which may
include the following:
They can only work with numeric information.
They can learn using examples.
They have the capability to process incomplete information.
They can perform graceful degradation and they are faulttolerant.
They have distributed memory.
They automatically generate information for unseen
examples.
They are mainly used for perceptual information processing.They perform learning well on pattern recognition and
classification.
They have a self-organizing capability.
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 20/40
Several examples of successful manufacturing applications can be
realized especially in the following areas:
probabilistic function estimation;
pattern recognition and classification;
pattern matching;
time series analysis;
signal filtering;
data fusion;
non-linear signal processing;
non-linear system modelling;
optimization;
intelligent and non-linear control;
data mining;
optical character recognition;
optimum path planning for robotics;
fingerprint recognition;
robot motions;
life cycle estimation of mechanical parts;
sonar signal classification;
production planning and scheduling.
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 21/40
GENETIC ALGORITHMS
Genetic algorithms are very popular in solving especiallyvery complex optimization problems. A set of some initial
solutions to a specific manufacturing problem is randomly
selected. These initial solutions are paired as parents and
new solutions are produced (children) out of them. Ineach production like this, the solutions are aimed to be
improved.
Once new solutions are generated, some of them are
carried out to the next generation in order to producenew solutions. This process continues until no better
solution is achieved. The better solutions are sought
through genetic operators and a fitness function.
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 22/40
Basic elements and procedure of genetic algorithms
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 23/40
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 24/40
Components of fuzzy logic
Fuzzification is to generate
fuzzy propositions and identify
linguistic variables and theirmembership functions
Defuzzification is theprocess of finding a crisp
value for the solution.
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 25/40
Generally the value corresponding to the highestmembership value is taken as the solution to the
problem.
This may be difficult for some solution spaces as theremay be more than one value having the same
membership value.
In this case, the average of those, or a valuecorresponding to the geometric average of the solution
space is taken as the net solution to the problem.
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 26/40
An example of a membership function
An example of a solution space
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 27/40
INTELLIGENT AGENTS
Intelligent agents are independent and autonomous
systems in performing their intended functions.
They can be both hardware and software systems and they
may incorporate more than one AI technology. They can
learn and work concurrently.
The general architecture of an agent is given in Figure
below:
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 28/40
There are three main components such as perception, cognition
and action
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 29/40
Perception is to receive the inputs from the environment
through sensors and convey them to the cognition module
to be processed. This process may include filtering, and
prioritization according to the order of importance.
Cognition is to process the information perceived and
making decision accordingly. This process may require
various methods of intelligence systems such as learning to be implemented. The cognition mechanism of an agent may
also deal with unexpected situations and adapt itself to new
situations as quickly as possible.
Action is to perform the command received from the
cognition using the respective effectors. This could for
example be to perform walking for a robot or stopping
when an immediate barrier in front is encountered.
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 30/40
INTELLIGENT MANUFACTURING SYSTEMS
Basic characteristics of intelligent manufacturing include the
following: -
ªminimize human involvement in manufacturing
activities.
ªarrange material and production compositions
automatically wherever possible.
ªmonitor and control production processes and
manufacturing operations.
ªrecommend and take immediate actions to prevent
faulty production.ªperform maintenance activities.
ªmake sure that the processes are working properly and
monitor their performance.
ªdiagnose machines and sustain manufacturing integrity.
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 31/40
APPLICATIONS OF INTELLIGENT MANUFACTURING SYSTEMS
Pham and Oztemel (1996) reported their work on intelligent
quality systems in their book. They have reported an expert system
called XPC capable of performing statistical process control.
Norrie et al. (1990) recommended an integrated manufacturing
management planning system, which takes process planning,group technology, scheduling, and simulation into account.
Injection moulding design system developed by Mok et al. (2008).
In this study, a Java-based intelligent design system equipped with
AI technologies is introduced
Tsourveloudis et al. (2000) reported a neural network and fuzzy-
logic-based robot gripping system
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 32/40
Intelligent control
is most useful in situations where classical linearcontrol is not suitable.
There are basically three ways that intelligent control
overcomes the limitations or the lack
of a model:1. learn to control a system using methods such as neural
networks and genetic algorithms which do not explicitly
require a model.
2. it can make do with very simple models, such asdescriptions of a system in words, and takes this
description to produce a controller using fuzzy logic.
3. it can use incomplete and imprecise models
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 33/40
Controlling systems can be represented by mathematical
formulae, or hierarchical.
For example, a motor car engine may be modeled by
relationships between numerical variables, but a road
system containing many motor cars cannot adequately bemodeled by formulae alone.
To illustrate the application of different methods of
intelligent control to a system which can be modeled usingequations, a problem called the b room- b alancer will be
investigated.
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 34/40
In the two-dimensional problem of the broom-
balancer, the trolley is moved by applying ahorizontal force in either direction- left or right.
When the broom starts to fall to the right, the trolley
is moved to the right and the broom should moveback to a vertical position.
One more feature is added to the system, namely that
the track has end-stops, so that not only has thebroom got to be balanced but the trolley must stay
near the centre of the track.
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 35/40
The objective is to keep the broom upright and the trolley in the
middle of the track.The angle U must be kept as close as possible to zero and the
position x from a fixed reference point must be kept as close as
possible to zero.
It will be assumed that the system has sensors which allow it to
measure U and x directly.
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 36/40
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 37/40
In general:
An intelligent mec h atronic syste m adapts the controller to
the mostly nonlinear behavior (adaptation), and stores
its controller parameters in dependence on the position
and load (learning), supervises all relevant elements,and performs a fault diagnosis (supervision) to request
maintenance or, if a failure occurs, to request a fail safe
action (decisions on actions).
In the case of multiple components, supervision mayhelp to switch off the faulty component and to perform
a reconfiguration of the controlled process.
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 38/40
CONCLUSIONS
Intelligent manufacturing systems are evolving with an
increasing speed. It is now apparent that intelligentmanufacturing systems will be more dominant in
industrial and manufacturing areas and:
They will make it possible to reduce the size of the
product but increase their functionality as well ascomplexity.
They will be equipped with highly effective sensors for
perceiving inputs and acting accordingly.
They will be capable of collecting/filtering andreasoning about the knowledge and data without any
extra burden on the systems.
They will reconfigure themselves in the light of the
progress in manufacturing systems and technologies.
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 39/40
They will have an extensive capability for learning and
planning as well as up-dating their knowledge
accordingly.
They will be able to adapt themselves to any kind of
manufacturing environment.
They will be able to define the best manufacturing
methods for their intended purposes.
They will increase their performance continuously and
reduce the human involvement.
It is now obvious that the next decade will be the decade
of unmanned factories and new changes will continue to
surprise both the academic and the industrial community
8/3/2019 Mech... AI
http://slidepdf.com/reader/full/mech-ai 40/40