12
ELSEVIER J. Mater. Process. Technol. 43 (1994) 165-176 Journal of Materials Processing Technology An investigation of an expert system for sheet-metal bending design Zone-Ching Lin*, Gow-Jeng Peing Department of Mechanical Engineering, National Taiwan Institute of Technology, 43, Keelung Rd., Section 4, Taipei, Taiwan, 10772, ROC (Received March 24, 1993;accepted October 23, 1993) Industrial summary The major objective of this paper is to study the possible application of an expert system in sheet-metal bending, and to build a prototype of a sheet-metal bending expert system on a PC/AT. On this basis, a large expert system could then be developed to enhance the design capability of sheet-metal bending. This sheet-metal bending expert system utilizes the qualitative data in a knowledge base and the quantitative data in a database, together with empirical design data, to aid the user in the design of sheet-metal bending. By means of the heuristic concept, the user can use data with the implicit expert experience and can alter the design whenever the design result is not satisfactory. Further to the above, this system has learning capability for the design of sheet-metal bending. When the inference engine cannot infer a result or a large quantity of data is going to be added into the rule base, the learning system can be started to convert data into qualitative rules and then save them in the rule base. Finally, the deduced drawing of the sheet-bending dies is demonstrated using AutoCAD graphic software. Key words." Sheet metal; Bending; Design; Expert system 1. Introduction Traditional sheet-metal processors care only whether the products comply with required standards and ignore the ability of automatic bending machines to design and regulate the products. Therefore, if they encounter the situation where the products do not meet the required standards, they are at a loss as to the solutions to the problem: this situation is the major focus of the present study. Needed to be * Corresponding author. 0924-0136/94/$07.00 © 1994Elsevier Science B.V. All rights reserved. SSDI 0924-0136(94)E0125-Z

An investigation of an expert system for sheet-metal bending design

Embed Size (px)

Citation preview

Page 1: An investigation of an expert system for sheet-metal bending design

ELSEVIER J. Mater. Process. Technol. 43 (1994) 165-176

Journal of Materials Processing Technology

An investigation of an expert system for sheet-metal bending design

Zone-Ching Lin*, Gow-Jeng Peing Department of Mechanical Engineering, National Taiwan Institute of Technology, 43, Keelung Rd.,

Section 4, Taipei, Taiwan, 10772, ROC

(Received March 24, 1993; accepted October 23, 1993)

Industrial summary

The major objective of this paper is to study the possible application of an expert system in sheet-metal bending, and to build a prototype of a sheet-metal bending expert system on a PC/AT. On this basis, a large expert system could then be developed to enhance the design capability of sheet-metal bending.

This sheet-metal bending expert system utilizes the qualitative data in a knowledge base and the quantitative data in a database, together with empirical design data, to aid the user in the design of sheet-metal bending. By means of the heuristic concept, the user can use data with the implicit expert experience and can alter the design whenever the design result is not satisfactory.

Further to the above, this system has learning capability for the design of sheet-metal bending. When the inference engine cannot infer a result or a large quantity of data is going to be added into the rule base, the learning system can be started to convert data into qualitative rules and then save them in the rule base.

Finally, the deduced drawing of the sheet-bending dies is demonstrated using AutoCAD graphic software.

Key words." Sheet metal; Bending; Design; Expert system

1. Introduction

Traditional sheet-metal processors care only whether the products comply with required standards and ignore the ability of automatic bending machines to design and regulate the products. Therefore, if they encounter the situation where the products do not meet the required standards, they are at a loss as to the solutions to the problem: this situation is the major focus of the present study. Needed to be

* Corresponding author.

0924-0136/94/$07.00 © 1994 Elsevier Science B.V. All rights reserved. SSDI 0924-0136(94)E0125-Z

Page 2: An investigation of an expert system for sheet-metal bending design

166 Z.-C. Lin, G.-J. Peing/Journal of Materials Processing Technology 43 (1994) 165 176

realized is that a processing technology has to combine operation skills with design to merit recognition.

Lin [1,2] has published studies on expert systems concerning shearing presses and deep-drawing presses. However, few people have researched expert systems in sheet- metal bending; the necessary design of punches and dies appropriate for bending machines for sheet bending and forming; and the selection of machine tools. This study is therefore devoted to researching in this area, whilst also incorporating the newly developed artificial intelligence-learning. On one hand this paper combines the experiences and techniques of experts, whilst on the other hand it also uses the data in design manuals [3-5], and in ordinary publications [1,6] to achieve an expert system for sheet-metal bending. This study has thus achieved originality in both application and academic research and, furthermore, conforms to today's trend towards automa- tion.

The paper focuses on the minimum pressure required when sheet metals undergo bending, the problem of the amount of springback, and the design and selection of punches and dies. The U-type, V-type and R-type dies in sheet-metal bending are the main areas included in this study. U-type bending can be divided further into the types: wedge; roller; and counter holder; whilst the V-type can be divided into the types: ordinary; straight sword; gooseneck; and sash. The R-type can be divided into the types: 1-V-die; cam; and general dies. During the selection of punches and dies, commercial sizes were considered first, designed press tools being chosen only under uncontrollable conditions. Due to the consideration of commercial cost-effectiveness, experimental data from the industries concerned was used as much as possible so as to be able to achieve the accuracy required in industrial sheet bending.

The ultimate purpose of this study is to promote actual application in the industry. Therefore, the PC-AT was chosen as the computer hardware, and the artificial intelligence language LISP, which possesses powerful symbol processing ability, was chosen as the system construction language. At the same time, a natural language dialogue program was added to perform knowledge base consultation. As to the acquisition of knowledge sources, the experiences and patterns of the human experts in problem solving were selected, using the most common knowledge demonstration method and production rule to save and acquire knowledge. Through the commu- nication of the man-machine interface, the modular concepts were used to read and design the related required data. This data can be of the calculation-type knowledge group or of the experience-type knowledge group: it can also be of the mixed-type knowledge group. After conversing with the man-machine interface, the acquired data is saved in the blackboard mode, and then entered into the inference engine to infer the required punches and dies with forward inference: these punches and dies are shown with drawing numbers. Then the graphic mechanism is used to infer the press die, which is shown with a drawing. If the inference engine cannot use the rules in the knowledge base to infer the drawing number, then the learning program is initiated. AutoCAD 3D wire-frame software (i.e., AutoCAD R10) developed by the Autodesk Co. was used as the software for the graphic mechanism, whilst Golden Common LISP LM version 2.0 [7] provided by the Gold Hill Computers Co. was used as the working environment for the expert system.

Page 3: An investigation of an expert system for sheet-metal bending design

Z.-C. Lin, G.-J. Peing/Journal of Materials Processino Technology 43 (1994) 165-176 167

2. The construction of the sheet-metal bending expert system

This study divided the construction of the knowledge of the expert system know- ledge into three parts: (1) the sheet-metal bending knowledge base; (2) the sheet-metal bending data base; and (3) the design experience knowledge base of sheet-metal bending. The sheet-metal bending knowledge base uses the rules of experience to control the system flow chart. The sheet-metal bending data base serves as the place for saving qualitative data such as the pressure measure, the amount of spring-back, etc., which is to be applied to the rules in the knowledge base. The design experience knowledge base of sheet-metal bending is for saving previous design experiences to be used in future design and learning. The construction of the sheet-metal bending knowledge-base system, of the data-base system and of the design-experience know- ledge-base, and the learning and the design of sheet-metal bending, are discussed separately in greater detail in the following sections.

First, the user enters the LISP environment, runs the expert system program, and chooses to enter the learning system or the expert system. Through the dialogue with the user, the program enters the bending model and its sub-model. Then the appropri- ate data is chosen by the strategy mechanism to start the operation of the expert system. After that, the data acquired from the man-machine interface is used by the expert system to infer the punches and dies required in the press. If no results can be inferred, then the learning system program is initiated. Then, through man-machine interface dialogue learning, the blackboard mode data is converted automatically to rules and input into the expert system knowledge base to be deduced one more time. After this the drawing number inferred by the expert system is combined with Auto LISP to demonstrate the punches and dies required in sheet-metal bending with drawing. This study uses the forward chaining inference, the construction being shown in Fig. 1, the following being the explanations of the function of various parts in this figure.

2.1. Man-machine interface

Generally speaking, the design of the man-machine interface consists of various forms, these being: "yes or no", "multiple choices", and "question and answer". This study used a mixture of all three forms, which varies according to the different requirements of the questions.

2.2. Control strategy system

The control strategy system in this study serves to converse with the user through the man-machine interface to obtain related data. Afterwards, it activates the data base and the knowledge base to call in required knowledge groups for supplying necessary data during the design of sheet-metal bending.

Since the main subject of this study is a design-oriented system, and due to the designing job usually having certain fixed procedures, and being easily subject to becoming stages, the concepts of related control can be applied to the knowledge base

Page 4: An investigation of an expert system for sheet-metal bending design

168 Z.-C. Lin, G.-J. Peing/Journal of Materials Processing Technology 43 (1994) 165-176

I graphic control unit I

I graphic data base 1

Auto CAD

inference engine

working memory system (black- board mode)

learning system

l design experience knowledge base

user t - -

. man-machine interface dialogue and

explanation system inquiringthe mater- ial and t~ickness Qf sheeeL metal,.and t h e b e n d m g an~le...

~ control strategy / system

/

knowledge base system I experience-type calculation-type knowledge group / knowledge group J

data base system J mixed-type

knowledge group

combinin~ with the graphic mechanism ] to show t~e data .about ben~ling.~rm- ation and required comb naUon draw- j ing of punches and dies

Fig. 1. System construction.

for solving design problems. According to the characteristics of becoming stages, the modular concept can be used in grouping the knowledge base, i.e. dividing a large knowledge base into sub-system modules specifically for every design stage, the advantage being that design does not require every type of knowledge. The input of related knowledge can be done according to the data given by the user, in which case the speed of program processing can be increased considerably. The problem of insufficient memory can, hopefully, be improved. Particularly for a PC/AT, the function of modulization skills proves to be unusually efficient, as shown in the modulization diagram in Fig. 2.

Page 5: An investigation of an expert system for sheet-metal bending design

Z.-C. Lin, G.-J. Peino/Journal of Materials Processing Technology 43 (1994) 165-176 169

V-type bending

m a n - machine interface dialogue

U-type bending

R-type bending

input ~ _ U-type

p rogram I

U-type Wedge

bending

U-type Roller

bending T

input Roller

knowledge base

U-type Counter holder

bending

_• s ta r t des igning

1 input

re la ted da t a

Fig. 2. Modulization diagram.

2.3. Inference engine

The rule of the inference engine used in this study is the forward chaining inference, where if the presentation of the rule is one-order, then no other rule is applied in the command "IF". As to the searching method, the degree of depth is the priority.

2.4. The construction of the sheet-metal bending knowledge-base

The LISP was used as the developing tool in the sheet-metal bending expert system in this study, using the production rule representation as the structure for knowledge representation. The forms were uniformly the "IF-THEN" two-way rule. The know- ledge of the sheet-metal bending knowledge-base system in this study can be divided into two types of knowledge group: a calculation-type and an experience-type know- ledge group.

The calculation-type knowledge group constructed in this study consists of die shoulder-width design, radius of die shoulder design, radius of roller design, left-over crack design, U-type bending depth design, U-type punch head radius design, and the calculation of pressure, etc. The experience-type knowledge group consists of die gutter angle design, punch tip angle design, selection of V-type punch tip radius, and

Page 6: An investigation of an expert system for sheet-metal bending design

170 Z.-C. Lin, G.-J. Peing/Journal of Materials Processing Technology 43 (1994) 165-176

selection of the punches and dies, etc. So far there are 107 rules in the experience-type knowledge group constructed by the production rule in this study.

As to the lists constructed by the data base, there are V-type bending pressure list, R-type bending pressure list, U-type bending pressure list, minimum flange length and minimum inner radius of V-type bending list, amount of spring-back of R-type bending list, selection list of the Urethane pad, selection list of R-type punch radi- us, etc.

3. Sheet-metal bending design-learning system

Rote-learning and concept learning are the methods used in this learning system. The learning method is divided into two parts. In the first part, related modules are classified according to the concept learning, and then rote-learning is used to memor- ize the knowledge. In the other part, the rote-learning memorizes all the input knowledge, and changes it into production rules. In addition, this system follows "Martin's Law": unless one almost knows the subject, one cannot learn it. In other words, learning means using the knowledge already acquired as the basis and supplementing it with the knowledge that has been gradually accumulated with added data. Therefore, the new rules acquired through the learning program are useful only when they are combined with the existing knowledge base.

Since the knowledge base in this study is constructed with the modular concept, the acquisition of knowledge or data concerning the aspect of learning is also established by this way. Learning is conducted through the man-machine interface inquiry. The inquired knowledge or data is input into the related module. A data matching program is also established to match the inquired data with the sheet-metal bending knowledge designed in the past. If the match is successful, it means that the sheet- metal, the user intends to bend, has been designed before. Therefore, the user can enter the exchange program directly to convert the inquired data into the data the designer needs. If the match is unsuccessful, the exchange program is entered first to convert parts of the data of the successful matches into the data necessary for inference. Then the man-machine interface dialogue program is entered selectively to complete the unsuccessful part of the match. After the man-machine interface is completed, the inference engine is entered to deduce the blackboard mode data. If the inference is unsuccessful, the learning system program is started and the automatic rule addition program is entered, which automatically arranges the knowledge or data stored with the blackboard mode through the array system into the "IF-THEN" model, and combines it with the existing knowledge base to become a new knowledge base with even more wisdom.

In addition, when there is a huge quantity of empirical knowledge or data to be converted into production rules and entered into the knowledge base to be stored, it will take a lot of effort and time to key in by hand and repetition with existing rules may also happen: the new rule addition program improved in this study can be used to solve this problem. This program adopts concept learning and, through the man-machine interface inquiry, inputs empirical knowledge or data into related

Page 7: An investigation of an expert system for sheet-metal bending design

Z.-C. Lin, G.-J. Peing/Journal of Materials Processing Technology 43 (1994) 165-176 171

modules, converts the acquired knowledge or data into the form of production rules, and then matches these rules with the existing knowledge base. After deleting the repeated portion, the remaining rules are combined with existing rules, thus forming a system with even more wisdom. The details are explained in the following.

3.1. The acquisition of knowledge or data concerning learning

With the necessary knowledge or data for learning acquired from the user through the man-machine interface, this study uses itemized assistance plus instruction to let the user reply and obtain knowledge or data. If no more knowledge is stored, then the formation of new rules in part 2 is activated.

3.2. The formation of new rules

This part uses the learning software established in this study to arrange data and form new rules. That is, the acquired knowledge or data passes through the array system to be arranged into the form of production rules and matched with the existing knowledge in order to delete any repeated rules.

3.3. The combination of knowledge bases

After the repeated rules are deleted, there are two ways to combine the new with the existing knowledge:

(1) Replacing existing rules: when there is repetition between a new rule and an existing rule, the existing rule is replaced by the new rule.

(2) Adding new rules: when there is no repetition between the new rules and the existing rules, the new rules are added to the knowledge base of existing rules in order to increase the system capabilities.

3.4. The recording of learning

In order to let the user distinguish between the existing rules, those rules which have been inferred unsuccessfully by the inference engine and then added automatically, and the numerous new rules input by the user, the authors implemented special distinction features whilst creating the program. Thus the record of every learning can be stored. For example, the rules added automatically use the "Rule Added-Rule" as the rule catalogue. The following are two more examples.

In Example 1 the representation numbers of punches and dies are PRVX and DRVY respectively, these being the drawing numbers of punches and dies given automatically by the system, where P represents the punch, R represents R-type bending, V represents the 1-V die of R-type bending, D represents the die, and X and Y represents that the size of this drawing element is in the form of variables. This kind of design will result in a difference in the die shoulder width or the punch radius on the surface. Although there will be contradicting rules about the same drawing number,

Page 8: An investigation of an expert system for sheet-metal bending design

172 Z.-C. Lin, G.-J. Peing/Journal of Materials Processin 9 Technology 43 (1994) 165-176

the size on the blackboard mode in a design record is different from that in all the other records, and the size in the drawing number is represented by a variable, so that differences in the size can then represent the differences in the design experience knowledge. New design experiences can thus be learned. Generally speaking, this kind of learning of the new design experience rules usually occurs during the re-designing of dies and punches with high accuracy requirement. In other words, the adopted punches and dies do not conform with commercial sizes.

3.5. The initiation of learning

This learning program can be regarded simply as the practice of the combined application of concept learning and rote learning, which means to record the data through the acquisition of data. The computer is very capable of such tasks, some- times surpassing the performance of humans. The data then goes through the matching program and is entered into the original knowledge base in the form of production rules in order to increase the capabilities of the knowledge base. When to initiate the learning program depends on the following situations:

(1) When the knowledge or data acquired through the blackboard mode is entered into the inference engine for inference, if no results can be inferred from existing rules, then the learning program is initiated.

(2) If the automatic rule addition system is initiated, then the learning program is initiated, Fig. 3 being the flow chart of the learning program.

4. Case study

After turning on the computer, "FF" is keyed in, when the LISP environment is entered automatically. Input "Work" will initiate this system. First, the man-machine interface inquires if the user is familiar with this system. If the answer is no, then the following will appear on the screen:

* * * * * * * * * * * * * * * * * * * * * * * * * * * * *

* SHEET-METAL EXPERT SYSTEM * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *

(1) RUN THE EXPERT SYSTEM. (2) DISPLAY THE RULES BASE. (3) UPDATE THE RULES BASE. (4) EXIT THE SYSTEM.

PLEASE INPUT YOUR CHOICE ? 1

If (1) is selected, the user enters the sheet-metal expert system and the consultation begins. If (2) is selected, the 13 knowledge bases are displayed to the user. If (3) is selected, this enables the user to add more knowledge. If (4) is selected, the user exits the system and enter the graphic mechanism.

Page 9: An investigation of an expert system for sheet-metal bending design

Z.-C. Lin, G.-J. Peing/Journal of Materials Processing Technology 43 (1994) 165-176 173

call the learning program ]

use the modular concept I to select the bending classification to enter

concept learning, enter inquiry] the knowledge or data

i

input the required sub-program and proceed Ij into related modules

with the sub-program I through the inquiry of I inquiry Ii the man-machineinterface

I input design experieence I type knowledge base

. . . . . . . . . i . . . . . . . . . . . . . . . complete directly convert

the knowledge or

match _ data by means of the exchange

partial match I successful

I enter the exchange program I to convert part of the data

enter the man-machine ] interface to complete

unfinished inquiry

I save the design conclusion in the design experience

type knowledge base

infer the blackboard mode knowledge or data

program without going through the inference engine

_ _ ~ enter the graphic mechanism

i

add new rule l

repeat the rote-learning

input the array [ I rule program

I I replace the

existing rule

automatic rule addition system

Fig. 3. The flow chart of the learning program.

Page 10: An investigation of an expert system for sheet-metal bending design

174 Z.-C. Lin, G.-J. Peing/Journal of Materials Processin9 Technology 43 (1994) 165-176

The complete design of sheet-metal bending is explained with V-type bending as the example.

Suppose the V-type bending is processed with the form of bottoming bending, where the bent type is the ordinary type of 90 °, the material to be bent is stainless steel SUS304, and the bent sheet has a thickness of 2 mm.

First, let the above data go through the man-machine interface dialogue. Since this study adopts the modular concept, related data is input whilst selecting the bending type and shape. After the input of data is completed, the analysis of the calculation- type knowledge group is run to obtain pressure, width of die shoulder, minimum flange length, and product inner radius, this data being saved in the working memory. The inference engine is then entered to conduct inference. The necessary drawing numbers of the punches and dies are inferred from the data obtained in the working memory, and then these inferred drawing numbers enter the graphic mechanism to produce draw- ings of the punches and the dies. The following is the contents of the working memory, the inferred results and the drawing demonstrated by the graphic mechanism.

The contents of the working memory are as follows:

(BENDING-TYPE IS V-TYPE-BENDING) (BENDING-KIND IS BOTTOMING) (THE ANGLE OF BENDING IS 90) (THE SHAPE OF 90-BENDING IS ORDINARY) (THE MATERIAL IS SUS304.)

* * * PLEASE TYPE ANY CHARACTER TO CONTINUE * * *

(THE (THE (THE (THE (THE

THICKNESS OF MATERIAL IS 2 MM) V-WIDTH IS 12 MM) MINIMUM FLANGE LENGTH IS 8.5 MM) INNER RADIUS IS 2 MM) REQUIRED PRESSURE IS 29.3333 TON/M)

THE PUNCH NO. IS 0 1 6 1 5 H R C 4 3 - 4 8

° 97

I. 67 . % - 90*

THE DIE NO. IS 1 2 4 0 0 H R C 4 3 - 4 8

~ 7.s--t I ' -

8 X 12

I. 5o 1

Fig. 4. The inferred drawings of the punches and the dies.

Page 11: An investigation of an expert system for sheet-metal bending design

Z.-C. Lin, G.-J. Peing/Journal of Materials Processing Technology 43 (1994) 165-176

The results inferred by the inference engine are as follows:

THE RESULT IS (RULE COMBINE6 SAYS THE PUNCH NO. IS 1615) THE RESULT IS (RULE COMBINE6 SAYS THE DIE NO. IS 12400)

The drawing displayed by the graphic mechanism is shown in Fig. 4.

175

5. Conclusions

This study has established a prototype expert system of sheet-metal bending design with preliminary learning capabilities, having the following characteristics:

(1) The sheet-metal bending knowledge group, consisting of knowledge base, data base, and design experience knowledge base, possesses flexible developing capabilities concerning sheet-metal bending design. (2) Since there are numerous types of sheet-metal bending, to design sheet-metal bending with a modular concept not only defines clearly the designed type, but also reduces the used memory at the same time, which increases the speed of execution. (3) The introduction of the calculation-type knowledge group enables this system to perform calculations. (4) The introduction of the graphic mechanism enables the user to actually under- stand the designed punches and dies. (5) This system is executed on a PC/AT, thus possessing learning capabilities and high applicability. This system uses concept learning and rote learning and possesses the function of automatic rule addition. The most important part of the learning process developed in this study is that when the inference by the inference engine is unsuccessful, the system automatically gives this rule the drawing numbers of punches and dies. The size of the graphic element within this drawing number is in the form of a variable. Therefore, different sizes represent different design experience knowledge. In this way new design experiences can be learned.

Acknowledgement

The authors gratefully acknowledge that Metal Industries Development Centre has supported this research.

References

[1] Z.C. Lin, C.L. Hsu and K.H. Yao, "Planning and Building an Integrated Design Software of Blanking and Piercing Dies (ESSCP) With A Micro Expert System as A Main Structure," J. Chinese Soc. Mech. Engrs., 10(2)(1989) 101-120.

Page 12: An investigation of an expert system for sheet-metal bending design

176 Z.-C. Lin, G.-J. Peing/Journal of Materials Processing Technology 43 (1994) 165-176

[2] Z.C. Lin and C.H. Huang, "An Investigation of an Expert System Employing Chinese for Deep- drawing of a Press Die Design", J. Chinese Soc. Mech. Engrs., 14(1) (1993). 63-77.

[3] Bending Technique, Amada Sheet Metal Working Research Association, (1981). [4] The ABC of Bending Tools, Amada Sheet Metal Working Research Association, (1981). [5] K., Lange, Handbook of Metal Forming, McGraw-Hill Book Company, (1985). [6] Y.-J., Da, "Press Working and Die Design", Shin luh Book Company, (1989). [7] Golden Common Lisp Large Memory, Gold Hill Computer Inc., (1984).