Learning process at development of intelligent environment for computer-aided design using case-base reasoning.
Portjanski, L. ; Nekrassov, G.
Abstract: The aim is to develop an integrated design space for CAD
of technological equipments for SMEs. The model of the design
environment is represented as a two-level hierarchical decision-making
system. The proposed environment provides support for creating and
manipulating 3D models of technological equipment and their components,
calculating functional and structural properties and the evaluation
criteria, and for determining the rules to control the design process
for optimising the portfolio of interdependent projects of technological
equipments. As examples the design of workholder is considered. The
computer aided workholder design system is created on case based
reasoning (CBR), in which the attributes of the workpiece and structure
of workholder as case index code are designed for the retrieve of the
similar cases, and the structure and hierarchical relation of case
library are stored.
Keywords: CAD, integrated design environment, case-based reasoning,
neural network
1. INTRODUCTION
The collaborative filtering and intelligence layer of the knowledge
management system builds on several possible combinations and
permutations of technologies: artificial intelligence tools, intelligent
data warehouses, genetic algorithm tools, neural networks, expert
systems, case-based reasoning (CBR) applications, rule bases, and etc.
(Tivana, 2002). A critical differentiator among the tools listed so far
is the level of knowledge needed to successfully use and apply a
particular technology or tool. Some tools require a high level of domain
knowledge from the user, whereas others assume that the end user is a
relatively passive observer in the process, and a time that is needed to
find a solution with a knowledge management tool in the specific
business application domain of interest.
In this report we propose a new approach for product's
configuration that integrates a constraint satisfaction problem with
putting together two methods: neural networks and CBR, and this
framework is applied to workholder design system.
2. THE MODEL OF INTEGRATED DESIGN SPACE
Based on the case based designing (CBD) methodology, the object
similarity is in two respects: the function and the structure
information. (Dieter, G. 2000). Then, the computer aided object design
system is created on case based reasoning (CBR), in which the attributes
of the main features of workpiece and structure of object as case index
code are designed for the retrieve of the similar cases, and the
structure and hierarchical relation of case library are setup for store.
Meanwhile, the algorithm based on the knowledge guided in the retrieve
of the similar cases, the strategy of case adaptation and case storage
in which the case identification number is used to distinguish from
similar cases are presented.
There are two ways of products' engineering:
* to realize engineering only using CAD system resources. Such of
designing is implemented for original designs, required much time and
great qualification of a designer;
* to realize engineering using the resources of preliminarily
environment of engineering for this CAD system. In this case the great
qualification of a designer is required at the stage of designing the
object-oriented environment, when the staff of a part is being
determined, its geometrical adjectives and assembling rules. It takes
less time and energy to execute the designing of specific parts of this
group. It may be used methodology when creating environment of
designing.
In the design space development the problem arises how to decompose the design process so that individual design tasks are communicating
only over the supervisory subsystem (Kuttner et al., 2003).
3 MASTER WIZARD AND NEURAL NETWORKS
Wizard is the main element of environment, supporting to design
objects of this group. However it is necessary to create its own wizard
for each group.
This report shows us in details the possibility of creating the
wizard master, which is invariant to designing objects. The scheme of
wizard master's work is corresponded with CBR
methodology--description of a problem, 4-RE cycle performance (retrieve,
reuse, revise, retain) (Aamodt & Plaza 1994), with the need of the
problem's decomposition with its specification performance and etc.
Creating environment of designing with using wizard master is
implemented on supervisor- level.
Tooling wizard master to the given group of tools is made with the
help of worked out language of high level as well as CAD system means.
In the latter case a designer puts together goals from parts, pointing
out conditions of impurities or a dialogue's call for choosing the
properly element of a design. In that way the script of a designing
process is created. This script is and/or a graph. There are procedures
at the top of a graph which is necessary for designing of this type of
parts, there are terms of procedures' choosing on arches.
A last user inserts additions to this script while designing parts.
The system is instructed on the basis of information (knowledge) got as
a result of designing of specific parts. The basic procedure of
designing is the procedure of decisions' forecasting.
The method of neural network is implemented for its realization
(program Neuro Pro 0.25).
Neural networks is the group of analytical methods, based on
(hypothetical) principles of teaching intellectual beings and brain
functioning and premising to forecast values of some variables in new
observations according to other data of observations after passing the
teaching stage according to existing data (Wasserman, 1989).
While using these methods first of all the question of choosing
specific architecture of network is aroused.
Dimension and structure of a network must correspond to a creature
of an investigated event. Then the set network is under the process so
called "teaching". On this stage the network's neurons
work up the input date interactively and correct its influences thus
that the network forecasts data in the best way, on which
"teaching" is implemented.
[FIGURE 1 OMITTED]
Learning together with a teacher supposes that for every input
vector there is a target vector representing a required exit. Together
they are called learning pair. Usually network is learnt on some numbers
of learning pairs. It is presented a output vector, calculated the exit
of network and compared to a corresponding target vector difference with
the help of return connection forwards to the network and weights are
being changed according to algorithm, aiming to minimize an error.
Learning vectors are presented in succession, errors are founded
out and weights are staved for each vector until an error reaches
acceptably low level throughout learning file.
5 CONCLUSION
The report considers the problem of design case representation and
retrieval phase of case-based reasoning. The design problems are often
difficult to represent as a well-structured list of features; the
representation requires various models because design content involves
different groups of decomposition hierarchy of design space such as:
workpiece, operation and workholder domains and different properties:
topological, geometrical, and physical. The work focuses on universal
way of representation of complex problems and the integration of
representation and retrieval of the cases in one step.
After teaching according to existing data the network is ready to
work and may be used for forecasting. "Network" got as a
result of "teaching", expresses the regularity existing in
data.
Using this method the network is the functional equivalent some of
the model between dependent variable, like those which is under
constructing in traditional modeling. However in contrast to traditional
models in case of "network" these depending cannot be written
in an explicit form like it is made in statistics.
The new paradigm is in the test phase, planned to be used in
different SME.
Acknowledgements:
Hereby we would like to than the Estonian Science Foundation, for
the grant G6183 "Methodology of Case-Based Reasoning (CBR) for
development design environment of technological equipment."
enabling us to carry out this work.
6. REFERENCES
Aamodt A. & Plaza E. (1994). Case-based reasoning: Foundational
issues, methodological variations, and system approaches. AI
Com-Artificial Intelligence Communications, 7(1), March 1994, pp. 39-59.
Dieter, G. (2000). Engineering Design, McGraw-Hill International
Editions, Mechanical Engineering Series, ISBN: 0-07-366136-8, USA
Kuttner,R.; Nekrassov,G.; & Sutt, A. (2003) Design Space for
Collaborative Concurrent CAD of Technological Equipment. DAAAM
International Scientific Book 2003, ed. Branko Katalinic, pp. 347-358.
Vienna 2003
Amrit Tivana (2002) Knowledge Management Toolkit. Second edition,
Dimensions K; ISBN: 01300922X
Wasserman, P. D. (1989). Neural Computing: Theory & Practice.
Van Nostrand Reinhold: New York. ISBN 0-442-20743-3
