1

EXPERT SYSTEM FOR FLEXIBLE MANUFACTURING SYSTEM

Ashraf S. Youssef; Ph. D.

AgendaThesis Objective & Scope.Motivation.Why Flexible Manufacturing System?QCES Parameters & AssumptionsQCES Parameters & Assumptions.QCES: A Framework for KB- DSS.Why Control Chart Patterns?Control Charts Patterns Identification Techniques.Pattern Identification Attributes (Features).The Computer Program (Analytical Part).QCES Rule-Base Building.CASE STUDIES.

Thesis Objective & Scope

The main purpose of the current research work is topropose a conceptual framework for an online realtime quality control system matching with the featurestime quality control system matching with the featuresand requirements of the recent advances inmanufacturing systems, especially FlexibleManufacturing Systems (FMS). According to theproposed framework, an online expert quality controlsystem (EQCS) is developed and implemented.

2

MotivationTo date, quality control methods remain underutilized.

Expertise is needed in quality control.

Application of Expert System (ES) in quality control pp p y ( ) q yactivities is important in FMS.

Most of the research efforts focus on obtaining good rules to help construct good quality control diagnostic system.

Most of the recent research suggests a framework of ES to develop statistical process control applications.

Motivation (Cont.)Most applications of ES in manufacturing concluded that the ES technology allows for implementation of sophisticated and efficient process control systems.

The new philosophy of manufacturing control became in demand when FMSs were installed.

All researchers that discussed pattern recognition concentrated on only six patterns: systematic, trends (upward & downward), stratification, cyclic, and natural pattern.

Why Flexible Manufacturing System?

Produce Different Products in Small Batches

Costly

Utilize Advanced Technology

FMSFully

Automated System

3

QCES Parameters & AssumptionsQCES is concerned with the analysis of control charts for variables (X-bar and R charts).QCES is based on the 3 sigma limits.I t t ti i QCES d d th AT&T lInterpretation in QCES depends on the AT&T rules and on the identification of the chart pattern. QCES identified the last six patterns as unknown.Attribute-value triple mode was used to represent the knowledge base of the ES part.Production rule was used to build the rule base.

QCES: A Framework for KB- DSS

Process Information

Patterns Information

Knowledge Base I

Model Base Statistical Analysis

Pattern Identification (computer program)ES Part

Inference Engine

QCSCharacteristicsComputations

UserInterface

Input Data

Why Control Chart Patterns?

Non-random patterns in control charts signal the presenceof special or assignable causes even if all points fallwithin the control limits.

Correct interpretation of patterns helps the qualityprofessional in identifying assignable causes; Westernand Electric Company stated fourteen unnatural patternsof control charts and the common causes of each pattern.

4

Control Charts Patterns Identification Techniques

Work on patterns identification in quality control chartshave taken four main directions:

Expert System. (QE Interpretation)Computer Vision. (QE Interpretation)Neural Network. (Time & Effort)Feature-Based Method.(More Accurate)

Feature-based method is the most accurate. Pham andWani (1997) stated that this has 99% accuracy for patternrecognition.

Control Chart Pattern Identification procedure

Investigate the Patterns and Extract the Features

B ild D t bBuild a Database

Convert these Data into a Decision Tree

Develop Computer Program

Verify the Results

Pattern Identification Attributes (Features)

Test of normalitySlopeMoving AverageXi>Xi+1 and Xi+1<Xi+2 for all samples9 points Consecutive above CL or lower CL9 points Consecutive below CL + σ8 of points Lie on warring Limits Zone Points > 5 σ + CL 15 Points consecutive in Zone C

5

Decision Tree of the Pattern Identification

SQCChart

Test of Normality

Fail

NormalPattern

PassMoving Average

Increased Decreased

points 51p1or 1

=σ+>σ−< XXXX ii

points 8p 2or 2

=−<

+>

σ

σ

XXXX

i

i

Xi>Xi+1 and Xi+1 < Xi+2For all Xi’s

Increasing Trend

Systematic Pattern

Decreasing Trend

Mixture

PassFail

Others

Fail Pass

Decision Tree of the Pattern Identification (Cont.)

Startis-fication More than one point

> 5 σ

FailPass

Freaks9 Consecutive Points

<

Sudden Shift in level

Unknown Pattern

Pass

PassFail

Fail

9 Consecutive Points >

Gradual Change

Fail Pass

σ−>σ+ XX or

2σ−X

The Computer Program (Analytical Part)

A computer program was developed based on the VB language; the program includes the following activities:

Computation of the quality control chart characteristics (X-bar and R charts).Plotting the quality control charts.Computation of the process capability.Interpretation of the quality control charts by applying AT&T rules first and then applying pattern identification procedure.Identification of the process status i.e. in or out of control & why?

6

QCES Rule-Base BuildingCollect data (assignable causes) related

to the unusual patterns

Construct an attribute value table

Develop a rule-base

Develop computer program

Verify rule-base

CASE STUDIES

Three case studies were considered to verify the proposed EQCS framework:

Case Study I (From Statistical Quality Control Hand Book)Case Study II (Conventional M/C)Case Study III (FMS)

CASE STUDY I(SQC Handbook)

Sample No.

Sample Size1 2 3 4 5

1 11.1 9.4 11.2 10.4 10.12 9.6 10.8 10.1 10.8 11.03 9.7 10.0 10.0 9.8 10.44 10.1 8.4 10.2 9.4 11.05 12.4 10.0 10.7 10.1 11.36 10 1 10 2 10 2 11 2 10 16 10.1 10.2 10.2 11.2 10.17 11.0 11.5 11.3 11.0 11.38 11.2 10.0 10.9 11.2 11.09 10.6 10.4 10.5 10.5 10.910 8.3 10.2 9.8 9.5 9.811 10.6 9.9 10.7 10.2 11.412 10.8 10.2 10.5 8.4 9.913 10.7 10.7 10.8 8.6 11.414 11.3 11.4 10.4 10.6 11.115 11.4 11.2 11.4 10.1 11.616 10.1 10.1 9.7 9.8 10.517 10.7 12.8 11.2 11.2 11.318 11.9 11.9 11.6 12.4 11.419 10.8 12.1 11.8 9.4 11.620 12.4 11.1 10.8 11.0 11.9

7

CASE STUDY I(Computer Implementation)

CASE STUDY I (Consultation Session)

CASE STUDY II(Conventional M/C)

8

CASE STUDY II(Computer Implementation)

CASE STUDY II(Consultation Session)

1 50-1.00-0.500.000.501.001.502.002.50

0 0.5 1 1.5Z (

-2.50-2.00-1.50

R ( i )

CASE STUDY III(FMS)

9

CASE STUDY III(Computer Implementation)

CASE STUDY III(Consultation Session)

DISCUSSION

Item R Chart X-bar Chart

Center line No Difference No Difference

Upper Control No Difference No Difference

Case Study III

Item R Chart X-bar Chart

Center line No Difference No Difference

Case Study I

LimitsLower Control

LimitsNo Difference No Difference

Case Study II

Item R Chart X-bar Chart

Center line No Difference No Difference

Upper Control Limits

0.003mm -0.002mm

Lower Control Limits

No Difference 0.003mm

Center line No Difference No Difference

Upper Control Limits

No Difference No Difference

Lower Control Limits

No Difference No Difference

10

DISCUSSION (Cont.)Three case studies were used for verification.In Case Study II, the causes of out-of-control were fired as follows: rotation of fixture, tool in need of sharpening,

f l f d f i i l d hiwear of tool, use of defective raw materials, and machine in need of repair.There is no significant difference in the results of the characteristics of the control charts.The normality test was also performed to verify the EQCS decisions.

ConclusionIn this research, a quality control expert system (QCES) framework was developed. This framework integrates statistical quality control system with expert system (Hybrid DSS). The proposed framework:( y ) p p

Allows for the identification of instability patterns of control charts.Helps to develop an on-line statistical quality control and process diagnostic system for FMS.Makes on-line quality control and adaptive sampling plans more effective in FMS.Was applied and verified in three different case studies.

Thank YouThank YouThank YouThank You