DEFECT-RELATED KNOWLEDGE ACQUISITION FOR

DECISION SUPPORT SYSTEMS IN ELECTRONICS ASSEMBLY

Sébastien Gebus and Kauko Leiviskä

Control Engineering Laboratory, Department of Process and Environmental Engineering

University of Oulu, P.O. Box 4300, FIN-90014 Oulu, Finland

Keywords: Decision support system, knowledge acquisition, quality, optimization, traceability, feedback.

Abstract: Real-time process control and production optimization are extremely challenging areas. Traditional

approaches often lack in robustness or reliability when dealing with incomplete, inaccurate, or simply

irrelevant data. This is a major problem when building decision support systems especially in electronics

manufacturing, where blind feature extraction and data mining methods on large databases are common.

Performance of these methods can be drastically increased when combined with knowledge or expertise of

the process. This paper describes how defect-related knowledge on an electronic assembly line can be

integrated in the decision making process at an operational and organizational level. It focuses in particular

on the acquisition of shallow knowledge concerning everyday human interventions on the production lines,

as well as on the conceptualization and factory wide sharing of the resulting defect information. Software

with dedicated interfaces has been developed for that purpose. Semi-automatic knowledge acquisition from

the production floor and generation of comprehensive reports for the quality department resulted in an

improvement of the usability, usage, and usefulness of the decision support system.

1 INTRODUCTION

A decision support system (DSS) can be defined as

“an interactive, flexible, and adaptable computer-

based information system, especially developed for

supporting the solution of a non-structured

management problem for improved decision

making. It utilizes data, provides an easy-to-use

interface, and allows for the decision maker’s own

insights” (Turban, 1995). Data, however, does not

exist naturally in a factory; it has to be collected,

stored, prepared, and eventually mined. Morover, it

might be incomplete, inaccurate, or simply irrelevant

to the problem that is being investigated thus leading

to the inability of decision makers to efficiently

diagnose many malfunctions, which arise at

machine, cell, and entire system levels during

manufacturing operations (Özbayrack & Bell, 2002).

These difficulties might be overcome by taking into

consideration knowledge about the environment, the

task, and the user (Gebus, 2006).

Knowledge-based approach takes advantage of

the fact that it is the people operating the process

who are most likely to have the best ideas for its

improvement. It is through the integration of these

ideas into the problem solving approach that a

solution for long term process improvement can be

found (Seabra Lopes & Caraminha-matos, 1995).

Additionally, as the use of knowledge and more

generally qualitative information better explains the

relationships between input process settings and

output response, it is well indicated for improving

the understanding and usability of DSS (Spanos &

Chen, 1997). In this paper, we shall examine the

possibility to integrate knowledge in general, and

especially shallow knowledge, into the decision

making process. Section 2 presents the problematic

related to knowledge acquisition and knowledge-

related improvements in man/machine interractions.

Section 3 presents our contribution to that field

through a case study that is followed by a discussion

on the results and the conclusion.

2 KNOWLEDGE ACQUISITION

Unlike data, knowledge does exist naturally in the

factory, but collecting and interpreting it constitutes

a major issue when building knowledge-based DSS.

270

Gebus S. and Leivisk

a K. (2007).

DEFECT-RELATED KNOWLEDGE ACQUISITION FOR DECISION SUPPORT SYSTEMS IN ELECTRONICS ASSEMBLY.

In Proceedings of the Fourth International Conference on Informatics in Control, Automation and Robotics, pages 270-275

 SciTePress

These tasks commonly carried out by a knowledge

engineer are often referred to as the bottleneck in the

expert system development (Feigenbaum &

McCorduck, 1983).

First and main obstacle is the knowledge

engineering paradox (Liebowitz, 1993). Knowledge

and skills that constitute expertise in a particular

domain is tacit. Furthermore, the more competent

experts become, the less able they are to describe

how they solve problems. Another contribution to

the bottleneck is the lack of willingness to share

knowledge. It is often said that knowledge is power

and people can be reluctant to give up what makes

them inexpendable (Verkasalo, 1995). Finally,

knowledge availability constitutes another obstacle

as experts are not always known and have little time

to spare. Additionally, today’s global working

conditions make it hard reaching experts located at

the other end of the world or across the street at the

subcontractor’s plant. Distributed decision making

becomes therefore a major issue (Verkasalo, 1995).

Currently, face to face discussions are still the

most widely used way of transferring knowledge as

they have the ability to make tacit knowledge more

explicit by allowing the expert to provide a context

to his actions. But expert interviews and other

manual techniques are not always possible and

depend very much on the knowledge engineer’s own

understanding of the domain. The challenge in a

global company is therefore to develop tools and

methods that enable experts to be their own

knowledge engineers. Three topics are commented

upon here: knowledge representation, automatic

knowledge extraction and the user interface.

2.1 Knowledge Representation

Experts reasoning is often incomplete and not

suitable for machine processing. Creating the proper

ontology is therefore an essential aspect of sharing

and manipulating knowledge. Based on the notion

that different problems can require similar tasks, a

number of generic knowledge representations have

been constructed, each having application across a

number of domains (Holsapple et al., 1989).

Common classes of knowledge representations are

logic, semantic network, or production rules.

Computer programs can use forms of concept

learning to extract from exemples structural

descriptions that can support different kinds of

reasoning (MacDonald & Witten, 1989). More

generally, automatic elicitation of knowledge, if

possible, offers great advantages in terms of

knowledge database generation.

2.2 Automatic Knowledge Extraction

Automatic knowledge extraction methods make it

possible to build a knowledge base with no need for

a knowledge engineer and only very little need for

an expert, for example by using case-based

reasoning. This poses however a knowledge

acquisition dilemma: If the system is ignorant, it

cannot raise good questions; if it is knowledgeable

enough, it does not have to raise questions. Scalable

acquisition techniques such as interview

metasystems (Kawaguchi et al., 1991) or

interviewing techniques using graphical data entry

(Gaines, 1993) can help overcoming this difficulty.

Because the domain knowledge is often very

specific, knowledge acquisition is a labor-intensive

task. For that reason, generic acquisition shells have

been developed (Chien & Ho, 1992) and extended

with methods for updating incomplete or partially

incorrect knowledge bases (Tecuci, 1992) (Su et al.,

2002). The work has also been facilitated by studies

on the automatic acquisition of shallow knowledge,

which is the experience acquired heuristically while

solving problems (Okamura et al., 1991), or by

compensating for the knowledge engineer’s lack of

domain knowledge, so that the resulting knowledge

base is accurate and complete (Fujihara et al., 1997).

2.3 User Interface

In DSS users are often presented with an exhaustive

amount of data upon which they have to make

decision without necessarily having the proper

understanding or knowledge to do so. The user

interface (UI) is the dialogue component of a DSS

that facilitates information exchange between the

system and its users (Bálint, 1995).

The choice of an interface depends on many

factor, but there are only few reasons for its

inadequacy (Norcio & Stanley, 1989). Mainly, the

UI is often seen as the incidental part of the system.

Consequently it is not well suited to the system or to

the user, and more often to neither. Usability can be

seen as the degree to which the design of a particular

UI takes into account the psychology and physiology

of the users, and makes the process of using the

system effective, efficient and satisfying.

For its response to be understandable, a DSS

should be able to tailor its response to the needs of

the individual. UI adaptability can be achieved by

mapping user’s actions to what they intend to do

(Eberts, 1991) or need (Lind et al., 1994). This can

undermine however the user’s confidence in the

information given to him. Adaptability can therefore

DEFECT-RELATED KNOWLEDGE ACQUISITION FOR DECISION SUPPORT SYSTEMS IN ELECTRONICS

ASSEMBLY

271

be applied to the content instead of the interface

itself by increasing the ability of a DSS to explain

itself for example by using graphical hierarchies

instead of the equivalent flat interface to describe the

structure of a rule base (Nakatsu & Benbasat, 2003).

Different planning and design methodologies

have been developed to insure that user

specifications are taken into consideration (Wills,

1994) (Balasubramian et al., 1998) (McGraw, 1994).

The following steps, also used in the case study, aim

to build a separate methodology to develop user

interfaces for knowledge acquisition:

• Identify and characterize the real users;

• Define a work process model;

• Definition of a general fault model;

• Design of a prototype, and

• Test, debugging, and redesign.

3 CASE STUDY

The electronic subcontractor involved in this case

study is lacking the resources for long-term process

improvement. Process control is left to the operators

who make adjustments only based on experience and

personal knowledge of the production line. Thus

tuning of the system and consequent quality of the

product depend very much on human interpretation

of machine problems. A DSS integrating this

“know-how” could lower the variability inherent to

human choices and greatly improve the efficiency of

any response when a problem occurs.

The proposed DSS tries to provide understanding

and formalization of the parameters influencing the

quality of the products, which are needed to improve

the operative quality. Traceability in terms of know-

how from the production floor is achieved through

an Expert Knowledge Acquisition System (EKCS)

recording the main breakdown information. Cross-

analysis of the subsequent enriched information with

measurement data can then improve the ability to

control the production as described in Figure 1.

3.1 Prototype Version

The approach based on fast prototyping has already

been described (Gebus, 2006). It is well suited for

creating tools that are both user and context specific.

This section serves as a reminder of the main points

and conclusions concerning the prototype version.

Line operators are the main users and knowledge

providers of the DSS, whereas quality engineers are

interested in getting knowledge in a way, which is

most suitable for a comparative analysis with

measurement data. The process model describes the

interactions between the process and the knowledge

sources. The fault model analyzes the cause and

effect relationships in the process model. Finally, a

prototype interface is designed to guide the operator

through the knowledge acquisition process using

pictures of the product and check boxes representing

the different cells of the production line. Fault

information is entered manually.

Prototype versions are time consuming but have

the advantage to allow the identification of problems

that the knowledge engineer would otherwise be

unaware of. All these problems can then be

addressed when designing a more complete version.

3.2 Final Version

3.2.1 Axes of Improvement

Based on the feedback from the test period, three

different user-oriented axes of improvement were

identified, Usability, Usefulness, and Usage (3U).

Usability concerns mainly knowledge input that

needs to be simpler, faster, and more intuitive. The

input method also has to favor system portability as

equipment is in constant evolution. In this context,

checkboxes that affect the appearance of the UI are

confusing and are replaced by clickable areas on

digital photographs of the production line and cells.

A defect is located by zooming in on an area, which

is linked to a list of problems and corrective actions.

This approach enables the needed flexibility and

portability while keeping the environment and the

interface very familiar, resembling a factory floor.

Usefulness probably is the most important target

of any system. However, experience has shown that

a system aimed at simplifying operators’ work, but

updated by design engineers, does not have any

long-term continuity. Motivating every single user

of the system into actually using is achieved by

transforming the DSS from a simple fault collecting

system to a factory wide information sharing system

providing user-specific levels of added value.

Figure 1: Flowchart representing the Expert Knowledge

uisition S

ste

ICINCO 2007 - International Conference on Informatics in Control, Automation and Robotics

272

Improved usage is consequent to improved

usefulness. Defect information is sent back to the

operators in forms that can be used during meetings

to discuss encountered production problems. In the

same way, quality engineers are more inclined to use

a system that automates some of their tasks and

provides information tailored to their needs it.

3.2.2 Overall Structure

In addition to specific interfaces for operators and

quality engineers, a dedicated UI has been created

for updating the system. As an information sharing

system benefiting all the users, Figure 2 shows only

closed loops of information flows. Operators provide

defect information stored in a database. The quality

department can access any relevant historical data to

produce statistical information. After analysis,

quality feedback is generated and sent back to the

production line. An administrator uses defect

information only casually for updating the system.

From the practical point of view, the three

interfaces use a unique database allowing an

automatic and immediate update of the system. This

database is stored on a SQL server providing the

needed flexibility that was missing in the prototype

version. Such a structure enables storing not only

information about date, time, defect and corrective

actions, but also all the settings relative to a certain

production line. This was added in order to make the

administrator interface a fully integrated subsystem.

3.2.3 Some Features in Details

The administrator interface was designed so that

new setups can be created easily. Updating is done

by selecting digital photographs of the production

lines and cells, and creating clickable areas that are

then linked to defect types and possible solutions.

Concerning the operator interface, emphasis has

been put on simplicity and intuitiveness. Setup

options are limited to choosing the database location

and selecting the current line. This is done only once

when the DSS is implemented on a new production

line or when the digital photographs are updated. In

normal use, data input is done by choosing the

product part from a list, and choosing the defect area

by zooming in on the pictures as in Figure 3. The list

of known causes and corrective actions is made

available as well as a comment window in case of

unreferenced problems. Selection of causes and

actions generates automatically a decision tree and

updates charts representing short-term quality and

usage information.

The supervisor interface is more complex and

offers three main options as shown in Figure 4.

Similar reporting capability as in the operator

interface is available, but with no limitations in time.

Long-term trends can be visualized from historical

data. Furthermore, advanced database exploration

properties have been added. Customizable SQL

requests can be created and run on the entire

database before exporting the result to other

software.

Production Line

Data

base

Defect

information

erator

Statistical

information

Qualit

artment

Defect

information

New

setups

Administrator

Feedback

control

Control

Algorithms

Quality

feedback

Actions can be

taken to improve

production

Figure 2: General structure of the DSS with information flows between the different entities.

DEFECT-RELATED KNOWLEDGE ACQUISITION FOR DECISION SUPPORT SYSTEMS IN ELECTRONICS

ASSEMBLY

273

4 RESULTS AND DISCUSSION

During the testing of the prototype version 183

machine stops occurred, out of which only 70 were

commented on representing therefore a usage rate of

38%. After having made the modifications, the final

version started to be used in a systematic way at

different levels of the company, thus promoting

employee involvement towards quality issues. Line

operators, in particular, not only increased the usage

rate to nearly 100%, but extend that use to their

weekly quality meetings. For them the increased

usage has been triggered by an improved usefulness.

The acceptation level for this new tool is based

on fully graphical interactions and digital pictures.

These enable quick updates of the system while

keeping a familiar framework. It also shows that in

order to satisfy the user needs, the real challenge for

an adaptable interface is not to evolve with the

problem, but rather to remain static while presenting

an evolving situation.

The new systems enables the cross analysis of

data and expert information, which is a prerequisite

for developing feedback control policies that will

lead to a more efficient factory-wide knowledge and

defect management.

The supervisor interface in

particular can be the backbone for implementing

monitoring and control algorithms. It has shown that

any kind of previously stored data can not only be

easily accessible, but also be processed by any

chosen algorithm and the results can be sent back in

various forms (charts, decision trees etc.).

One can also imagine replacing manual feedback

with control algorithms generating automatic

feedback control. This is not possible in the current

state of the system as proper actuators necessary to

transform information into action on the production

line are missing. Even if they did exist, automatic

feedback control would still be highly dependent on

line technology and therefore not portable.

5 CONCLUSION

When building a knowledge-based system, the

approach that is used has to be very human oriented.

Defining the right interfaces for real-time knowledge

acquisition can be a major problem. They have to be

adapted to users with various degrees of knowledge.

In addition to this, the complexity of any interface

must be sufficient enough to catch the full scope of

information, but simultaneously keep the data

extraction process as simple as possible.

The general process for designing a knowledge

acquisition interface applied to this case study

presents the different tasks that have been

undertaken and the problems encountered. Unlike

traditional design techniques that emphasize doing it

right the first time, the 3U approach proposed in this

section leads to a better match with user concerns.

Knowledge acquisition software has been

implemented on the production floor in a factory

producing components for the electronics industry.

Based on a test period, the knowledge gained from

the use of this tool enabled defect classification and

standardization. This is the first step towards cross

analysis with monitored parameters from the

production floor, leading eventually to on-line fault

diagnosis.

REFERENCES

Balasubramanian, V., Turoff M., Ullman, D., 1998, A

Systematic Approach to Support the Idea Generation

Phase of the User Interface Design Process. 31st

Hawaii Int. Conf. on System Sciences. vol. 6, pp. 425 –

434.

Bálint, L., 1995, Adaptive Interfaces for Human-Computer

Interaction: A Colorful Spectrum of Present and

Future Options. IEEE Int. Conf. on Systems, Man and

Cybernetics. vol. 1, pp. 292 – 297.

Chien, C.C., Ho, C.S., 1992, Design of a Generic

Knowledge Acquisition Shell. IEEE Int. Conf. on

Tools with AI. pp. 346 – 349.

Eberts, R., 1991, Knowledge Acquisition Using Neural

Networks for Intelligent Interface Design. IEEE Int.

Conf. on Systems, Man, and Cybernetics. vol.2, pp.

1331 – 1335.

Feigenbaum, E.A., McCorduck, P., 1983, The fifth

generation: artificial intelligence and Japan's

computer challenge to the world. Addison-Wesley

Longman Publishing Co.

Fujihara, H., Simmons, D.B., Ellis, N.C., Shannon, R.E.,

1997, Knowledge Conceptualization Tool. IEEE

Transactions on Knowledge and Data Engineering

9(2): 209 – 220.

Figure 3: Interface for fault selection.

ICINCO 2007 - International Conference on Informatics in Control, Automation and Robotics

274

Gaines, B.R., 1993, Eliciting Knowledge and Transferring

it Effectively to a Knowledge-Based System. IEEE

Transactions on Knowledge and Data Engineering

5(1): 4 – 14.

Gebus, S., 2006, Knowledge-Based Decision Support

Systems for Production Optimization and Quality

Improvement in the Electronics Industry. Oulu

university press.

Holsapple, C.W., Tam, K.Y., Whinston, A.B., 1989,

Building Knowledge Acquisition Systems – A

Conceptual Framework. 22nd Hawaii Int. Conf. on

System Sciences. vol. 3, pp. 200 – 210.

Kawaguchi, A., Motoda, H., Mizoguchi, H.R., 1991,

Interview-Based Knowledge Acquisition Using

Dynamic Analysis. IEEE Expert: 47 – 60.

Liebowitz, J., 1993, Educating Knowledge Engineers on

Knowledge Acquisition. IEEE Int. Conf. on

Developing and Managing Intelligent System Projects.

pp. 110 – 117.

Lind, S., Marshak, W., 1994, Cognitive Engineering

Computer Interfaces: Part 1 – Knowledge Acquisition

in the Design Process. NAECon 1994 Conf. vol.2, pp.

753 – 755.

MacDonald, B.A., Witten, I.H., 1989, A Framework for

Knowledge Acquisition through Techniques of

Concept Learning. IEEE Transactions on Systems,

Man, and Cybernetics 19(3): 499 – 512.

McGraw, K.L., 1994, Knowledge Acquisition and

Interface Design. Software, IEEE 11(6): 90 – 92.

Nakatsu, R.T., Benbasat, I., 2003, Improving the

Explanatory Power of Knowledge-Based Systems: An

Investigation of Content and Interface-Based

Enhancements. IEEE Transactions on Systems, Man,

and Cybernetics 33(3): 344 – 357.

Norcio, A.F., Stanley, J., 1989, Adaptive Human-

Computer Interfaces: A Literature Survey and

Perspective. IEEE Transactions on Systems, Man, and

cybernetics 19(2): 399 – 408.

Okamura, I., Baba, K., Takahashi, T., Shiomi, H., 1991,

Development of Automatic Knowledge-Acquisition

Expert System. Int. Conf. on Industrial Electronics,

Control and Instrumentation. vol.1, pp. 37 – 41.

Seabra Lopes, L., Camarinha-Matos, L.M., 1995, A

Machine Learning Approach to Error Detection and

Recovery in Assembly. IEEE/RSJ Int. Conf. on

Intelligent Robots and Systems. vol. 3, pp. 197 – 203.

Spanos, C.J., Chen, R.L., 1997, Using Qualitative

Observations for Process Tuning and Control. IEEE

Transactions on Semiconductor Manufacturing 10(2):

307 – 316.

Su, L.M., Zhang, H., Hou, C.Z., Pan, X.Q., 2002,

Research on an Improved Genetic Algorithm Based

Knowledge Acquisition. 1st Int. Conf. on Machine

Learning and Cybernetics. pp. 455 – 458.

Tecuci, G.D., 1992, Automating Knowledge Acquisition

as Extending, Updating, and Improving a Knowledge

Base. IEEE Transactions on Systems, Man, and

cybernetics 22(6): 1444 – 1460.

Turban, E., 1995, Decision support and expert systems:

management support systems. Englewood Cliffs, N.J.,

Prentice Hall.

Verkasalo, M., 1997, On the efficient distribution of expert

knowledge in a business environment. Oulu university

press.

Wills, C.E., 1994, User Interface Design for the Engineer.

Electro/94 International Conference.

Özbayrak, M., Bell, R., 2003, A Knowledge-based

decision support system for the management of parts

and tools in FMS. Decision Support Systems 35(4):

487 – 515.

Reports with no limitations in time

and the possibility to edit the

information given by operators

Customizable SQL requests and

ossibility

o export data to othe

software

Same charts as for operators bu

without limitations in time

Figure 4: Structure of the supervisor interface.

DEFECT-RELATED KNOWLEDGE ACQUISITION FOR DECISION SUPPORT SYSTEMS IN ELECTRONICS

ASSEMBLY

275