1 - Computer and Information Science

advertisement
1
Research on hybrid expert system application to blanking technology
Xiaoyi Chi
,
, Ma Haojun, Zhao Zhen and Peng Yinghong
Department of Plasticity Technology, National Die and Mold CAD Engineering Research
Center, Shanghai Jiao Tong University, Shanghai 200030, PR China
Received 6 July 1999. Available online 11 October 2001.
Abstract
This paper introduces the application of a second generation expert system in blanking
technology. Depending on the empirical knowledge of domain experts and field
literature, an expert system can play an important role in the field of blanking technology.
An object-oriented blackboard model is given which can realize hybrid knowledge
representation, such as production rules, frameworks and procedures. This paper mainly
discusses the structure of the blackboard model and hybrid knowledge representation in
blanking technology expert system. Since strip layout is a crucial and determining part in
blanking technology, given here is an example of a blackboard model supporting hybrid
knowledge representation about the strip layout process.
Author Keywords: Blanking technology; Hybrid knowledge representation; Blackboard
model; Expert system
Article Outline
1. Introduction
2. Traditional knowledge representation formalism
3. Hierarchical knowledge representation
4. Blackboard model supporting hybrid knowledge representation
4.1. O-o technology
4.2. Blackboard model
4.3. An example
5. Conclusions
References
1. Introduction
Standardization and computerization of die design knowledge is of prime importance
owing to the trends in automation and intelligence of blanking die design. As an
important sub-field of artificial intelligence, an expert system has already been introduced
into the field of die CAD. Since expert system technology has evolved from first
generation to second generation [1], the new technology should be applied to the die
CAD system, and researchers should use these advanced technologies in their work.
2
First generation expert systems usually employed a single knowledge representation
formalism and they explained their recommendations by rephrasing the rules involved in
the inference chain that led to the solution. However, blanking technology design is a
complicated, experience-based process including many procedures such as blanking,
bending and local forming, as shown in Fig. 1. These procedures involve an amount of
complicated domain knowledge, and so a first generation expert system is difficult to
meet the requirements of blanking technology design.
(11K)
Fig. 1. Blanking technology classification.
Second generation expert systems are distinguished from their counterparts of the first
generation in the following ways [1]: (1) they organize their knowledge base in modules
and multi-level structures; (2) they employ hybrid representations by combining multiple
representation schemes; (3) they supply sophisticated semi-automated and automated
tools to help with knowledge acquisition; (4) they can provide more satisfactory and
comprehensive explanations of their actions. As a feature of second generation expert
systems, they separate control knowledge from domain-specific knowledge and organize
the knowledge base in multi-level structures. These advantages of a second generation
expert system can provide a mechanism reflecting practical design process, and will
improve the system’s intelligence and automation when applied in blanking die CAD.
2. Traditional knowledge representation formalism
Simply put, the phrase "knowledge representation" means that knowledge about facts and
relations is encoded in a manner that allows computers to recognize it. Knowledge
representation is very important in building an expert system because the success of the
expert system depends largely on the effectiveness of knowledge representation. Most
blanking technology design expert systems employ single knowledge representation, the
following being the two most commonly used types [2]:
1. Production-rule format. The format can encode knowledge as follows:
IF (prerequisite/condition) THEN (action)
The following rule is an example:
IF the hole-to-hole distance has precision requirement
THEN holes should be pierced at the same step in order to decrease accumulated dimensional tolerances
caused by location errors
2. Frame representation. The basic structure can be described as follows:
3
Frame representation is widely used in expert systems because its structure is suited to
the description of a group of basic concepts and facts.
3. Hierarchical knowledge representation
Hierarchical knowledge representation means that concepts can be classified into
hierarchical organizations in which the top knowledge is abstract and general while the
descendant knowledge is specific and is used to solve practical problems [3].
There are two types of knowledge contained in the knowledge base of blanking
technology: domain knowledge and control knowledge. Domain knowledge is specific to
the problem at hand. Control knowledge is more of abstract nature and describes the
overall objectives and strategies of how the system is to pursue a goal. In first generation
expert systems, no distinction was made between the two types of knowledge and control
knowledge would either be absent or implicit and hidden within the domain-specific
knowledge. Consider the following example:
IF there exists two holes whose differences in diameter are great, and the distance between two
holes<minimal hole-to-hole distance that will prevent a weak die
THEN pierce the large hole first, then the small one
There is an important control strategy here which is implicit in the design of the rules.
The order of the two conditions of the rule is not arbitrary. The condition "the distance
between two holes<minimal hole-to-hole distance that will prevent a weak die" should be
considered only if "there exists two holes whose differences in diameter are great". So,
there is an implicit blanking technology hierarchy which determines the rule structure.
The control strategy of this rule may be described as follows. Before checking the
location, check the diameters. The strategy remains unstated in the knowledge base.
However, making such control knowledge explicit can make it more convenient to
represent blanking technology and can supply more comprehensive explanations of the
system decisions, which also attributes to building and maintaining the knowledge base.
Take the above rule for example: two conditions locate at different hierarchies. When the
top domain knowledge (i.e. there exists two holes whose differences in diameter are
great) is activated, the down domain knowledge (the distance between two
holes<minimal hole-to-hole distance that will prevent a weak die) is activated through
control knowledge; thus the domain knowledge is separated from the control knowledge.
The above example is one rule about blanking which is easy to understand. However, the
process of blanking is complicated and involves many factors that have to be considered.
4
For example, when piercing holes in the process of strip layout design, many related
factors must be taken into account. The related rules are as follows:
Rule 1: If the hole-to-hole distance has precision requirement, then holes should be
pierced at the same step in order to decrease accumulated dimensional tolerances caused
by location errors.
Rule 2: If the hole-to-contour distance has precision requirement, then the hole should be
pierced at one step before the blanking step as far as possible so that it is approximately
close to the blanking step.
Rule 3: Pilot holes should be pierced at the first step in order that piloting could work at
the following steps.
Rule 4: If a hole has high-dimensional precision and contour requirements (especially to
thin strip materials), then the hole should be pierced at the back steps in order to avoid the
influences of other steps.
Rule 5: If a hole interferes with many other holes (i.e. the minimal hole-to-hole
distance>permitted minimal value), then the hole should be pierced at the next step.
Rule 6: If there exists two holes whose differences in diameter are great, and the distance
between two holes<minimal hole-to-hole distance that will prevent weak die, then pierce
the large hole first, then the small one.
Rule 7: With the strip material rigidity requirement, pierce the small hole first, then the
large one.
In the rule-based expert systems, the rules are stored in the knowledge base as text and
have no clear and definite classification. In this paper, the authors put these rules into
hierarchical organizations. Although it has been noted that every person has a different
hierarchy of concept which is influenced by one’s personal experiences and beliefs, it is
still feasible to employ a hierarchical organization of knowledge in the field of blanking
technology which has been developed maturely for many years. Fig. 2 gives an
illustration of hierarchical knowledge representation. The illustration is a general thought
which ignores some details. For example, in Rule 6, more details discussed above here
are ignored.
(14K)
Fig. 2. Hierarchical knowledge representation illustration.
5
4. Blackboard model supporting hybrid knowledge
representation
Blanking technology includes many types of knowledge each of which employs different
knowledge representation format. Now, it is possible to employ hybrid knowledge
representation by applying hierarchical concept to the organization of blanking
technology knowledge base and realizing the separation of domain knowledge and
control knowledge. This paper gives an object-oriented blackboard model which employs
many types of knowledge representation formats and forms hybrid knowledge
representation under the administration of blackboard.
4.1. O-o technology
O-o (object-oriented) technology is an essential field of software in recent years. The
main building block of an o-o program is object which simulates real-world concept. O-o
technology describes the data structure of objects to build the system, which matches the
human’s cognitive process. It realizes data abstraction, information hidden and class
inheritance between objects. The concept of class reflects the human’s cognitive process
from the general to the specific. So, it is feasible to apply o-o technology to the design of
expert systems.
4.2. Blackboard model
The blackboard model provides a very flexible control structure for judging and solving
problems. In the blackboard model, information in different processes (or at different
levels) is described on the blackboard, then recommending/reference information is
output. So, the characteristics of the blackboard is that it is a shared data area, or it is an
interactive medium of stored information (data) and knowledge sources. Blackboard
model usually consists of three major components [4]:
1. Knowledge source. Knowledge used to solve problems is divided into many
knowledge sources which are stored individually and independently of each other.
2. Blackboard data structure. The blackboard is referred to as a global database that
records and organizes intermediate problem-solving data. Knowledge sources produce
changes to the blackboard that lead incrementally to a solution to the problem.
Communication and interaction among knowledge sources take place solely through
changes on the blackboard.
3. Controller. The KSs (knowledge sources) respond opportunistically to changes on the
blackboard. The control shells, which lie in KSs, blackboard and isolated modules or the
combination of the above three items, monitor the changes on the blackboard and decide
which action to take next.
A blanking technology design process is characterized by the following requirements: (1)
it is a creative process; (2) it contains a variety of information; (3) multiple design results
will be given. In order to meet these requirements, expert systems should provide a
variety of knowledge representation formats to build KSs base. An object-oriented
6
blackboard model can realize hybrid knowledge representation, such as production rules,
frameworks and procedures. According to the characteristics of blanking technology,
different types of knowledge are represented in different ways, which are embedded in
blackboard model to produce an o-o blackboard model supporting hybrid knowledge
representation. In this way, high efficient administration of blackboard combined with the
advantages of hybrid knowledge representation and o-o technology can be incorporated
to the complicated blanking technology expert system design. Fig. 3 shows a blanking
technology expert system blackboard model.
(14K)
Fig. 3. Blackboard model.
As shown in Fig. 3, the blackboard is a global database which stores necessary
conditional information and solutions recommended by knowledge sources. Interaction
among independent KSs must take place through blackboard. When a rule of blanking
technology is activated, a message is sent to the blackboard which processes the message
and then sends another message to knowledge base. The knowledge source with the
highest priority is executed, it will complete one or more of the following procedures: (1)
add a hypothesized element to the dynamic blackboard; (2) revise existing elements; (3)
produce a link object of some hypothesized element. Then, the control shell in the
blackboard will monitor the changed element, activate the correspondent KSs and
produce further changes on the blackboard until these changes lead to a satisfactory
blanking technology decision.
4.3. An example
The processing feature is the main criteria in blanking technology design. Thus, this
paper employs hybrid knowledge representation based on hierarchical knowledge
representation and takes processing features as objects. In this method, frameworks are
used to describe object which includes object name, attributes and operations on object.
Production rules and procedures can be combined with frameworks to represent complete
and detailed blanking knowledge. Object name represents processing feature, attributes
include the super-class of the object, knowledge type and object class, method means
dynamic actions of objects and procedure knowledge processing. Blackboard is a
dynamic monitor which processes all types of knowledge and at the same time, its
functions include receiving and sending messages and searching matching rules, etc. Fig.
4 gives an example of a blackboard model supporting hybrid knowledge representation
about the strip layout process. Operation sequence arrangement is one of the sub-classes
of strip layout, being derived from class strip layout. The knowledge is stored at different
7
levels of knowledge sources and is activated by the blackboard event driven by messages
[5 and 6].
(17K)
Fig. 4. An example of the blackboard model supporting hybrid knowledge representation about the strip
layout process.
5. Conclusions
An expert system for blanking technology is significant in improving die CAD
intelligence. However, blanking technology design is a complicated and flexible process
which makes it challenging to build an expert system. In this paper, the authors describe
the characteristics of a second generation expert system, and give an o-o blackboard
supporting hybrid knowledge representation. In addition, how knowledge and experience
are represented in knowledge base is also discussed. Since there are many functions for
further development, it is possible to enhance the research in the future.
References
1. C. Nikolopoulos, Expert System, Marcel Dekker, New York, 1997.
2. X. Xiangzhi, et al., Blanking Technology and Die CAD, National Defence Industrial
Press, Beijing, 1996 (in Chinese).
3. M. Kamel, Y. Quintana, Generation of Concept Hierarchies for Engineering
Knowledge-based Systems, Application of Artificial Intelligence in Engineering VII,
Computational Mechanics Publications, 1992, pp. 107–124.
4. H. Bo, Z.K. Ni, Structural design of an intelligent decision system generator based on
blackboard model, Comput. Res. Dev. 34 (5) (1997) 382–386 (in Chinese).
5. B.N.K. Ann and C.C. Kai, A knowledge-based system for strip layout design. Comput.
Ind. 25 (1994), pp. 31–44. Abstract-INSPEC
6. J.C. Choi, B.M. Kim, Y. Cho, C. Kim and J.H. Kim, An integrated CAD system for the
blanking of irregular-shaped sheet metal products. J. Mater. Process. Technol. 83 (1998),
pp. 84–97. Abstract-Compendex | Abstract-INSPEC | Abstract + References in Scopus |
Cited By in Scopus
8
Corresponding author. Present address: Department of Mechanical Engineering, State
University of New York at Stony Brook, Stony Brook, NY 11794-2300, USA; email:
xchi@ic.sunysb.edu
Download