This action might not be possible to undo. Are you sure you want to continue?

Welcome to Scribd! Start your free trial and access books, documents and more.Find out more

1

INTRODUCTION

, knowledge, and resources. How well one succeeds will be a function of adherence

rve to confirm and modify these initial ideas. An iterative loop is established be

erimental “variability.”

s “plan, do, check, act.” Both statements illustrate the role of statistics as an i

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

2

INTRODUCTION

statistics comes into formal play in two places: in helping design the experiments

nsidered course of action aimed at answering one or more carefully framed questio

tivity are too vast and varied to be left to a single individual.

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

3

INTRODUCTION

ructed graphical displays of data quickly assist in data analysis.

illustrative examples. Many statistical software programs go far beyond the subje

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

4

INTRODUCTION

o be qualitative, e.g., different machines, different operators, switch on or off.

different temperatures, then the factor temperature has four levels. In the case o

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

5

INTRODUCTION

eatment combination” is the set of levels for all factors in a given experimental

gical entities, natural materials, fabricated products, etc. in known or unknown ways.

han between different portions. Observations

taken within a day are likely to be

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

6

INTRODUCTION

use of certain tools called planned grouping, randomization, and replication.

influence . We say, “ (eta) is a function of x,” that is, η =f(x). Of course, observ

se. The errors (noise) attending a series of experiments have two primary componen

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

7

INTRODUCTION

nd then repeating an experiment. Experimental error, the failure of agreement betw

odel, thus allowing the data to identify appropriate subsets of models for the exp

n of the roles of ε and η is formally recognized in an analysis of variance (ANO

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

8

INTRODUCTION

batches, operators, machines, or days. These variables are commonly called “blocks,

ocks to accentuate the influences of the studied factors. Designs that make use of

of uncontrolled variables will balance out. It also improves the validity of esti

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

9

INTRODUCTION

ovides an opportunity for the effects of uncontrolled factors or factors unknown

to item measurements or to the variability occurring between adjacent items manu

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

10

CLASSIFICATION OF EXPERIMENTAL DESIGNS

These designs have certain rational relationships to the purposes, needs, and phys

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

11

COMPLETELY RANDOMIZED DESIGN : ONE FACTOR , k LEVELS

periment and there are k treatments (or levels of the factor) to be investigated.

e experiment. The advantages of the design are :

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

12

COMPLETELY RANDOMIZED DESIGN : ONE FACTOR , k LEVELS

the case of a one-way classification analysis of variance, the most appropriate m

d the column headings were “Batch 1,” “Batch 2,” “Batch 3,” where the “batches” repr

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

13

COMPLETELY RANDOMIZED DESIGN : ONE FACTOR , k LEVELS

imenter, the data may be represented by the Fixed Effects Model (Model I), whereas

lso be interested in knowing about the “components of variance”; that is, the vari

ments.

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

14

BLOCKED DESIGNS

of the studied factors are repeated each day or with a different operator, machin

ore using a block design data analysis.

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

15

RANDOMIZED BLOCK DESIGN

divide the experiment into blocks, or planned homogeneous groups. When each such

s may also have an influence upon the response, then we might plan to observe all units at random within a given block.

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

16

BALANCED INCOMPLETE BLOCK DESIGNS

t can be studied to three a day. The production manager is concerned that day-to-d

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

17

BALANCED INCOMPLETE BLOCK DESIGNS

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

18

LATIN SQUARE DESIGNS

r nonhomogeneity, i.e., two different blocking variables. Such designs were original

positions or operators and days. The studied variable, the experimental treatment

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

19

LATIN SQUARE DESIGNS

ur positions on the wear machine. A 4 X 4 Latin square will allow for both sources

3 X 3 to 7 X 7 are given in the table on next slide . Strictly speaking, every ti

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

20

LATIN SQUARE DESIGNS

at random, permute the rows at random, and assign the letters randomly to the tre

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

21

YOUDEN SQUARE DESIGNS

by the fact that the number of rows, columns, and treatments must all be the same

of of of of of

treatments to be compared levels of one blocking variable (columns) levels of another blocking variable (rows) replications of each treatment times that two treatments occur in the same block

square, t = b and k = r.

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

22

PLANNING INTERLABORATORY TESTS

of the true picture. It is always a difficult problem to decide whether or not o

of the materials to rank the laboratories. The data from the interlaboratory test

t is consistently high in its ability to measure the response will show a lower r

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

23

PLANNING INTERLABORATORY TESTS

est is performed by several laboratories, the results are disappointing. The reaso

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

24

PLANNING INTERLABORATORY TESTS

**he program. The two materials should be similar in kind and in the value of the pr
**

drawn through the median of all points in the x direction and a horizontal line

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

25

PLANNING INTERLABORATORY TESTS

e into four quadrants, and the first (and often revealing) step in the analysis is

upper right; if a laboratory is low on both samples, its point will lie in the lo

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

26

NESTED ( COMPONENTS OF VARIANCE ) DESIGNS

nvestigations associated with interlaboratory comparisons, or the repeatability a

d” or “hierarchical” designs.

l elements are random variables.

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

27

PLANNING THE SIZE OF THE EXPERIMENT

η

k

= η The corresponding averages y1, y2, …, yk computed from the recorded da

fied, existing tables or charts can be used to determine the necessary sample siz

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

28

FACTORIAL EXPERIMENTS — GENERAL

ach of all possible combinations of levels that can be formed from the different

force, and the factorial experiment would consist of 20 trials. In this example, t

timated main effect is the difference between the average responses at the two le

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

29

FACTORIAL EXPERIMENT WITH TWO FACTORS

at happens when other values of force and amperage are employed. Four values of f

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

30

FACTORIAL EXPERIMENT WITH TWO FACTORS

ticeably higher resistivity values at amperage level 5 and the apparent changes i

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

31

SPLIT - PLOT FACTORIAL EXPERIMENTS

r k to give a total of N = m1 X m2 X… X mk experimental trials.

alyst, that is, a 3 X 3 X 4 factorial design in N = 36 trials. To be a standard fa

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

32

ACTORIAL EXPERIMENTS WITH k FACTORS ( EACH FACTOR AT

TWO LEVELS )

by a letter (or numeral) and then to denote the two levels (versions) of each fac

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

33

EVOP : EVOLUTIONARY OPERATION

uce information about itself while simultaneously producing product to standards.

ate and temperature. The value of a dependent variable is the result of the settin

better operating conditions selected for any desired production rate. The response

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

34

EVOP : EVOLUTIONARY OPERATION

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

35

EVOP : EVOLUTIONARY OPERATION

rease profit in an operating plant with minimum work and risk and without upsetti

ure on the process. Study cost, yield, and production records. ly the definitive text. ction management. Organize a team and hold training sessions. s that are likely to influence the most important response. steps according to a plan. (Cycle 2) and each succeeding cycle, estimate the effects. ificant, move to the indicated better operating conditions and start a new EVOP p en shown to be effective, change the ranges or select new variables P plan and adjust the ranges as necessary. he rate of gain is too slow, drop the current factors from the plan and run a new

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

36

EVOP : EVOLUTIONARY OPERATION

urers’ literature, patents, textbooks, and encyclopedias of technology. Do not negl

effect of past changes in equipment and conditions.

e use is the two-level complete factorial. There are important reasons to maintain

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

37

EVOP : EVOLUTIONARY OPERATION

the average of the other four peripheral points. It is therefore a signal of curv actors. In the Taguchi literature the response would be termed “robust” to changes

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

38

EVOP : EVOLUTIONARY OPERATION

ts, as can be seen by adding a constant to the five runs of Cycle 3, and recalcula

rate, percent impurity, or pounds of byproduct. A calculation sheet is made for ea

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

39

BLOCKING THE 2 k FACTORIALS

e question is how to choose the trials to be run each day so as not to disturb th

action is constructed and labeled the block “generator.” Those runs carrying a plu

fect.

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

40

BLOCKING THE 2 k FACTORIALS

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

41

FRACTIONAL FACTORIAL EXPERIMENTS ( EACH FACTOR AT TWO LEVELS )

ed subset of all possible combinations. The analysis of fractional factorials is

therefore only 2kp independent estimates are possible. In designing the fractional

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

42

DESIGNING A FRACTIONAL FACTORIAL DESIGN

he column of signs associated with the highest order interaction. These signs are

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

43

OTHER FRACTIONAL FACTORIALS

ny factors assumes great simplicity in the mathematical model for the response fu

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

44

TAGUCHI OFF - LINE QUALITY CONTROL

se industrial environments experiments are run to identify the settings of both p

rtant to note that the word “design” takes differing connotations: product design,

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

45

TAGUCHI OFF - LINE QUALITY CONTROL

hogonal arrays. The inner array consists of a statistical experimental design emp

noise statistics. No closure to the debate seems imminent. One thing is clear. The

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

46

RESPONSE SURFACE DESIGNS

rch and development laboratories and sometimes on actual plant equipment itself. I

lled variables and a single response variable are studied. The data obtained are

rved (a second-order approximation).

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

47

RESPONSE SURFACE DESIGNS

analyst with vivid insights into the nature of the responses and factors under in

ursued.

e with this approach is that a false optimum can be reached. Consider the followi

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

48

RESPONSE SURFACE DESIGNS

ant yield. For example, there is an entire set of conditions of concentration and failed for a fairly simple reason.

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

49

RESPONSE SURFACE DESIGNS

it deserves. This often leads to difficulties later on. In the present example the

o be done intelligently. The specific scale over which each factor is to be studie

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

50

MIXTURE DESIGNS

ends on the proportions of the metallic elements present; gasoline is ordinarily a

trained to fall within narrow ranges, thus forming isolated mixture regions and r

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

51

GROUP SCREENING DESIGNS

med, each containing several factors; the groups are tested; and individual factor

Sec 47 DESIGN AND ANALYSIS OF EXPERIMENTS

52

- Jordan
- Jordan [Compatibility Mode]
- Chap 06
- Chap 05
- Chap 04
- Chap 03
- Chap 02
- Chap 01
- Sec 36 Quality and The National Culture
- Sec 48 Reliability Concepts and Data Analysis
- Sec 46 Acceptance Sampling
- Sec 45 Statistical Process Control
- Sec 44 Basic Statistical Methods
- Sec 43 Quality In Latin America
- Sec 42 Quality In China
- Sec 41 Quality In Japan
- Sec 40 Quality In The United States
- Sec 39 Quality In Central & Eastern Europe
- Sec 38 Quality In Western Europe
- Sec 37 Quality In Developing Countries
- Sec 34 Second Generation Data Quality Systems
- Sec 33 Financial Service Industries
- Sec 32 Health Care Services
- Sec 31 Government Services

- Anova
- Improvement of the Decision Efficiency of the Accuracy Profile
- Chap 001
- One Wayanova
- QT-MBA Sem I
- Analytical Chemistry
- Metrics HealthCare
- Forecasting
- Regression Analysis
- Stat 104 Textbook
- Response Surface Methodology and MINITAB
- RM5
- CH1 Introductionrw
- Work Sampling Ppt
- SM+1010+Introduction
- Note
- June2010.pdf
- Autocorrelation - computer lab
- Signal Detection
- CHAP DOE.pdf
- Statistical Terminology
- probit 2
- St-Pierre Stats Paper for Meta-Analysis 2001
- CHAPTER 1
- B.S.(G)Final Year Math Ma Tics
- 31301578 Quantitative Techniques 1
- Class 3 - CI
- Reasearch
- Kla Taguchi

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue listening from where you left off, or restart the preview.

scribd