Process improvement to the anodize line through design of experiments

Rochester Institute of Technology

RIT Scholar Works

Theses

Thesis/Dissertation Collections

3-1-1996

Process improvement to the anodize line through

design of experiments

Shantel Gammie

Follow this and additional works at:

http://scholarworks.rit.edu/theses

This Thesis is brought to you for free and open access by the Thesis/Dissertation Collections at RIT Scholar Works. It has been accepted for inclusion

in Theses by an authorized administrator of RIT Scholar Works. For more information, please contact

[email protected].

Recommended Citation

Process Improvement to the Anodize Line

through Design of Experiments

by

Shantel Gammie

A Design Project

ill

Partial Fulfillment

of the

Requirements for the

MASTER OF SCIENCE

ill

Mechanical Engineering

Approved by:

Professor

Names Illegible

Thesis Advisor

Professor

-Professor

-Professor

-Department Head

DEPARTMENT OF MECHANICAL ENGINEERING

COLLEGE OF ENGINEERING

PERMISSION GRANTED:

I, Shantel Gamrnie, hereby grant permission to the Wallace Memorial Library of the

Rochester Institute of Technology to reproduce my design project entitled Process

Improvement to the Anodize Line through Design ofExperiments in whole or in part.

Any reproduction

will

not be for commercial use or profit.

March 20, 1996

Thanks

to the

KEMDMPO

Finishing

Department

and a special

thanks to

ABSTRACT

The

goal ofthisproject

is

to analyzetheanodize

line

and make process

improvements

which

directly

affectthe

product,

theresults

being

reduced

defects,

lower

variability

in

the

process and

product,

faster

cycle

time,

reduced_costs, and

higher

profits.

Possible

defect

conditions were

identified,

tracked,

and analyzed

in

orderto

determine

thegreatestproblem.

After

recognizing

a prime

improvement

opportunity,

a

design

of experimentswasconductedwith

thepurposeof

showing

relationships

between

key

process parameters and productcharacteristics.

Finally,

recommendationsweremadetoraisethe

quality

level

oftheanodize

line,

withthe

TABLE

OF CONTENTS

LIST OF TABLES

viii

LIST OF

FIGURES

ix

I.

INTRODUCTION

1

A.

Anodizing

1

1. Definition

1

2. Applications

2

3. The Process

4

4.

Types

of

Anodic

Coatings

5

a.

Barrier

5

b.

Porous

5

5.

Coating

Structure

of

Porous-type

6

6. Mechanism

8

B.

Theory

of

Design

of

Experiments

9

1. Definition

9

2.

Qualifications

9

3. General Procedure

10

4. JMP

Analysis

14

a.

Summary

of

Fit

14

b.

Analysis

of

Variance

(ANOVA)

17

c.

Parameter Estimates

20

d.

Effect Test

22

e.

Lack

of

Fit

23

f

Leverage Plots

26

H.

KEYPROCESS PARAMETERS

27

m.

DEFECTS

31

IV.

DESIGN OF EXPERIMENTS

42

A.

Pre-experiment

42

1. Brainstorm

42

a.

Project

Objective

42

b.

Response Variable

(Output)

43

c.

Factors

(Inputs)

44

2.

Revelations

Since

the

Brainstorming

46

a.

Material

46

C.

Experiment-

Final Design

52

D. The Experiment

53

E. Response Variables

54

F. Post-experiment

56

V

ANALYSIS

57

A. Experimental

Results

57

B.

Scatter Plots

58

C. Residuals

66

D.

t-Test

84

E. JMP DOE

Analysis

86

1.

Smut

87

a.

Ordinal

87

b.

Continuous

89

2.

Blue

90

3. Degree

of

Seal

91

4. Darkness

92

F.

Correlations

93

D.

Summary

of

Results

94

1.

Significant

Factors

94

2. Predition Profile

94

VI.

RECOMMENDATIONS

98

VH.

APPENDLX

105

A. Aluminum

Alloy

Compositions

and

Designations

106

1. Wrought Aluminum

106

2.

Cast

Aluminum

107

3.

Special Treatment

107

B.

Types

of

Designs

110

1. Full Factorial

110

2. Fractional Factorial

113

3.

Screening

118

4. Response

Surface

119

C.

Values

of

to,v

used

for

the t-Test

121

D.

Smut JPM DOE Analysis

122

1. Ordinal Response

122

2.

Analysis

1

126

3.

Analysis

II

130

E. Blue JPM DOE

Analysis

133

1. Analysis 1

133

F.

Degree

of

Seal

JPM DOE

Analysis

140

1.

Analysis

1

140

2.

Analysis

U

144

G.

Darkness

JPM

DOE

Analysis

147

1.

Analysis

1

147

2. Analysis

U

151

H.

Correlations

155

LIST OF TABLES

Table Page

#_

1.

Industrial Applications

of

Anodized

Aluminum

3

2.

Pore Diameter

and

Cell Wall

Thickness

of

Several

Oxide Coatings

7

3.

ANOVA Table

18

4.

Summarization

of

Key

Process

Parameters

30

5.

Customer Complaints

38

6.

Scrap/Rework

40

7.

Possible

Variables That

Cause Smut

(Brainstorm)

44

8.

Experiment Variables

and

Constants

(First

Draft)

45

9.

First Design Proposal

47

10.

Revised Table

8

.

Final Experiment Variables

and

Constants

51

11.

Final

Design

52

12.

Experimental Results

57

13.

Residual Values

67

14.

Residual Values

Excluding

Run

#9

76

15.

t-Test

Inputs

85

16.

t-Test

Results

85

17.

Smut

Results

(As

an

Ordinal Response

Variable);

%

Opportunity

of

Producing

Is

and

3s

for

Each Factor

at

3 Levels

88

18.

Smut

Analysis Results

(As

a

Continuous

Variable)

89

19.

Blue

Analysis

Results

90

20.

Degree

of

Seal Analysis Results

91

21.

Darkness

Analysis

Results

92

22.

Significant

Pairwise

Correlations

93

23.

Significant

Main

Effects

94

Al.

Wrought Al

Alloy

Groups

First

Digit

Designation

106

A2.

Cast

Al

Alloy

Groups

First Digit Designation

107

A3.

Alloy

Suffix

Designations

107

Bl.

Full Factorial Design

Ordering

Pattern

112

B2.

Fractionation

Table

117

B3.

Design Resolution

118

LIST OF FIGURES

Figure Page

#

1.

Anodize Process

Flow

4

2.

A

Cross-section

of a

Barrier-type

Coating

5

3.

A

Cross-section

ofa

Porous-type

Coating

6

4.

Causes for Lost Contact

33

5.

Causes for

Overanodizing

33

6.

Causes for Smut

34

7.

Causes for White Spots

34

8.

Causes for

Bleed

Out

35

9.

Causes for

Bent Parts

35

10.

Causes for Crashes

36

11.

Causes for

Burnt Parts

36

12.

Causes

for

Staining

37

13.

Customer

Complaints

(1994)

39

14.

Scrap/Rework

(10/24/94-12/3

1/94)

41

15.

Smut

vs.

Free

Sulfuric Acid

59

16.

Smut

vs.

SealpH

59

17.

Smut

vs.

Seal

Temperature

59

18.

Smut

vs.

DI Rinse Temperature

59

19.

Blue

vs.

Free

Sulfuric

Acid

60

20.

Blue

vs.

SealpH

60

21.

Blue

vs.

Seal Temperature

60

22.

Blue

vs.

DI Rinse Temperature

60

23.

Weight Loss

vs.

Free Sulfuric Acid

61

24.

Weight Loss

vs.

SealpH

61

25.

Weight Loss

vs.

Seal Temperature

61

26.

Weight

Loss

vs.

DI Rinse Temperature

61

27.

Darkness

vs.

Free

Sulfuric Acid

62

28.

Darkness

vs.

Seal

pH

62

29.

Darkness

vs.

Seal

Temperature

62

30.

Darkness

vs.

DI Rinse Temperature

62

31.

Response

Variables

vs.

Seal Temperature

64

32.

Smut

and

Degree

of

Seal

vs.

Seal Temperature

65

33.

Residual

Smut

vs.

Smut

69

34.

Residual Blue

vs.

Blue

69

35.

Residual

Degree

of

Seal

vs.

Degree

of

Seal

69

36.

Residual

Darkness

vs.

Darkness

69

37.

Residual

Smut

vs.

Run

70

41.

Residual Smut

Normality

Test

71

42.

Residual Blue:

Normality

Test

72

43.

Residual Degree

of

Seal:

Normality

Test

73

44.

Residual Darkness:

Normality

Test

74

45.

Residual

Smut

vs.

Smut

(Excluding

Run

9)

77

46.

Residual

Blue

vs.

Blue

(Excluding

Run

9)

77

47.

Residual

Degree

of

Seal

vs.

Degree

of

Seal

(Excluding

Run

9)

77

48.

Residual Darkness

vs.

Darkness

(Excluding

Run

9)

77

49.

Residual

Smut

vs.

Run

(Excluding

Run

9)

78

50.

Residual Blue

vs.

Run

(Excluding

Run

9)

78

51.

Residual Degree

of

Seal

vs.

Run

(Excluding

Run

9)

78

52.

Residual Darkness

vs.

Run

(Excluding

Run

9)

78

53.

Residual

Smut

Normality

Test

(Excluding

Run

9)

79

54.

Residual Blue:

Normality

Test

(Excluding

Run

9)

80

55.

Residual

Degree

of

Seal:

Normality

Test

(Excluding

Run

9)

81

56.

Residual Darkness:

Normality

Test

(Excluding

Run

9)

82

57.

Prediction

Profile;

Minimized

Smut

95

58.

Prediction

Profile;

Most

Desirable

97

59.

Process Window

Example

100

60.

Smut

and

Darkness

vs.

Free

Sulfuric

102

61.

Smut

and

Darkness

vs. pH

Seal

102

62.

Smut

and

Darkness

vs.

Temp. Seal

103

63.

Smut

and

Darkness

vs.

Temp. DI

103

INTRODUCTION

The

introduction

section

is

to

introduce

thereaderto the

anodizing

process and

Design

of

Experiments.

Anodizing

Definition

Anodizing

oranodic oxidation

is

anelectrolytic process

for

oxidizing

aluminumto

producean

improved

surface quality.

"Aluminum

withoutsome surfacetreatment

is like

goodwoodwithout varnish.

The

wood

is

strong

and

may

make a good structural

member,

but it does

not

look

as good as

it

could,

and

it is

susceptibletowearand

weather."1

Anodizing

is like

varnishing in

theexample above.

It

addsto the

quality

of

thealuminum

by

making

it

more resistantto theenvironment.

Exposing

aluminumtoair produces athinoxide

film,

that

is

0. 1-0.4

x

10"6

inches

(0.25-1.0

x

10'2

u,m)

thick.

Anodizing

willproduce athickeroxide

coating

than the

film

formed

naturally

in

air.

With

a_{thicker coating,} aluminum

has

improved

physical and

chemicalpropertieswhichallows

for

expandedapplications.

Some

ofthe

improved

properties

include

excellent resistancetomarine and atmospheric

corrosion,

abrasion

Anodizing

occurswhenanelectrochemical conversionoccurs

from

metallic

aluminumtoaluminum

oxide,

A1203

.

This

conversionrequires asource of

direct

current

passing

througha suitable acid electrolytewhich will produce oxygen

ions.

The

most

commonly

used electrolyte

is

a

dilute

sulfuricacid_solution,

but

chromic_acid, oxalic_acid,

phosphoric acid plus

additives,

andother specialized electrolytes with

limited

applications

are also possibilities.

Applications3

Anodic

coatingsare

widely

appliedtoaluminum

because

of

its

unique responseto

anodizing.

There

are

many

advantagesgained

from

anodizing

aluminum.

The

following

is

a

list

of principal

functions for

anodizing.

Undercoat

for

organic

coatings,

electroplated metallic_coatings, and solid

lubricants

Corrosion

resistant

coating

Coloring

(a

wide range of_colors,

for

_example,

black, bronze,

purple, orange)

Antimark

applications

Heat

reflectionand radiant

heat

absorption

Wear

resistance and

lubrication

Electrical

resistance

Abrasion

resistance

Thermal

resistance

To

get an even

better idea

oftheapplications

for

anodized_aluminum,

Table 1

below lists

some ofthemore

important

applications

in

present-day

industrial

practice.

Table

1. Industrial

Applications

of

Anodized

Aluminum4

Industry

Application

Building

Decoration,

protection of exterior

building

_components,

structural_members,

storefronts,

entrance_ways, window

frames,

ceiling panels,

handrails, hardware,

telephone

booths.

Transportation

Auto:

Headlight

bezels,

grills,

window

frames,

garish

moldings,

brake

pistons.

Air: Aircraft

instrument

_panels,

landing

gear, propellers,

fuel

pumps, wing

skins,structural

components,

_rivets,

instruction

plates, trim.

Consumer Durable

Goods

Refrigerator:

trim,

shelves,

evaporators, appliance

trim,

cooking

utensil

covers,

baking

_pans,name

plates,

furniture,

giftware,

costume

jewelry,

firearm/military

components.

Lighting

Reflectors

for

highway

and stadium

lights,

indoor

lighting

fixtures.

Electrical

Capacitors,

insulated

wireand

strip

conductors.

The Process

Understanding

the

anodizing

process

is

simplified

by

looking

atthe

flow

ofthe

process.

A flow

chart

is

shown

below in

figure

1.

(

Load

J

?

Clean

Neutral

Rinse

Etch

Rinse

No

Rinse

*

Anodize

Dye

Clear

Rinse

Deox

Rinse

Seal

DI

Rinse

(inspect

J

HotDI

Rinse

Unload

4

Hot

Dry

<

Cold

Dry

* '

Types

of

Anodic

Coatings

Anodic

coatings are classified as

barrier

or porous

depending

onthesolvent action

oftheelectrolyteonthe

naturally

occurring

oxide

layer.

Deciding

onthe typeof

coating

is

based

ontheapplication ofthepart

being

anodized.

Barrier-type

Electrolytes

with

little

orno

capacity

to

dissolve

theoxide

form

barrier-type

coatings.

These

typeofcoatings arethin

(less

thanonetenthousandthof an

inch),

compact,

nonporous,

and

electrically

resistant.

In addition,

withsuitable

etching

conditions,

high

capacitance

is

obtainable.

Sodium

borate/boric

acid electrolytes are

examplesofthis typeof

film

producer.

Electrical

capacitors

have barrier-type layer.

Aluminum

Figure 2. A

cross-section ofa

barrier-type

coating

Porous-type

Porous

typeof coatings are

formed in

anelectrolytewith

high

solvent actionon

thenatural oxide.

The

formed film

consists ofa porousouterportion and athin

barrier

portionadjacentto themetal.

Porous-type

coatings

have

widerangesofapplications

Porous Layer

Barrier Layer

Aluminum

Figure 3. A

cross-sectionofa poroustype

coating

Coating

Structure

ofthe

Porous-type

The

structure oftheanodic

coating

is

a

group

of

hexagonal-shaped

oxide cells

each

having

acentralporethatextendstoathincompact

barrier layer

ofoxide.

The

barrier layer

is

continuously

transformed

into

theporous

form

during

the

anodizing

process.

The

cell size equalstwice thecellwallthicknessplusthe central pore

diameter.

There

are

approximately

a million cells per squareinch.6

The

cell structure oftheoxide

layer formed from

sulfuric, oxalic,

chromic, and phosphoric acids are

similar,

but

vary

in

Table 2. Pore

Diameter

and

Cell

Wall Thickness

of

Several Oxide

Coatings7

Electrolyte

Pore Diameter

(Angstrom)

Wall Thickness

(Angstrom/volt)

15%

sulfuric

acid, 50

F

120

8.0

2%

oxalic

acid, 75 F

170

9.7

3%

chromic

acid, 100

F

240

10.9

4%

phosphoric

acid, 75

F

330

10.0

The

pore

diameter is completely dependent

onthe typeofelectrolytewhereasthe

wallthickness

is

highly

dependent

ontheappliedvoltageand

slightly

dependent

onthe

electrolyte.

Additionally,

the

coating

thickness

depends

onthesametwomain

factors:

applied voltage andtheelectrolyte.

For

examplesodium

borate/boric

acid electrolytes and

300-500

voltsare conditionsthatproduce athin

film

withathickness that

is less

than

0.0001

inch. A

sulfuric acid solution and

12-24

volts are conditionsthatproducethicker

films

of

up

to

0.001

inch.

Other

factors

affecting

the thicknessarethecurrent

density

and

Mechanism

The anodizing

process

is dissimilar

to

electroplating in

the

way

that the

coating

forms.

Porous-type

anodic

films

startontheoutside surfaceofthemetal anodeand

progress

inward,

sothat the

last-formed

coating

is

nearthe

metal-coating

interface

andthe

first-formed

layer

is

onthesurface.

By

contrast,

themetal

being

plated

in

the

electroplating

process

is

a

conducting

substrate

acting

asthecathode

in

theelectrolytic

cell.

A

metallic

coating

is deposited

onthesurfaceofthe substrateandgrowsoutwards.

Furthermore,

when_anodizing, additionalmetallicmaterials are not

being

addedto the

aluminum,

instead

aconversion ofthesurface

is

occurring.

The

mechanism of

forming

the

barrier

coatings

is ionic.

Aluminum

ions

combine

withtheoxygen

ions

oftheelectrolyte.

The barrier

thickness representsthe

distance

thoughwhichthe

ions

canpenetratethe

layer

ofoxideunderthe

influence

oftheapplied

potential.

Therefore

thevoltage

is

the

driving

force behind

the

ions,

and

determines

the

thicknessofthe

barrier layer.

For

barrier-type

coatings,

a

limiting

thickness

is

reached and

current

flow

ceases.

Porous-type

coatings

do

not reacha

limiting

thickness

due

to the

solvent action oftheelectrolyte.

However,

a

barrier layer

withathickness that

is

equalto

fourteen

times theapplied voltagetimesa

factor less

than

unity

determined

by

the

electrolyte will stillexist

between

themetal and

base

ofthepores

for

a porous

film.

9

Barrier Layer Thickness

=

Theory

of

Design

of

Experiments

Definition

Design

ofexperiments

(DOE)

is

a systematicapproachtoexperimentationthat

allows an efficient and effective efforttowards

improving

the

quality

and

productivity

of a

process.

The

goal of

DOE

is

tounderstandthe

relationship between

processparameters

and product

characteristics,

saveexperimentation

time,

decrease

scrap

_rates,

decrease

production

times,

decrease

inventory,

and save costs associatedwith each ofthese. 10

Qualifications

DOE

is

auseful problem

solving

process

in

many

different

situations,

for

example:

1)

thereexists apart with

high

nonconformity

or

many

defects, 2)

thereexistsaprocess

with

high

nonconformity,

3)

anew_machine,processorpart

is

being implemented,

or

4)

a

newmachine

is

being

purchased.11

A nonconformity

is

a

departure from

specification requirements.

A

defect is

any

variation

of a required characteristicoftheproductor

its

parts,

which

is far

enough removed

from

its

targetvaluetopreventtheproduct

from

fulfilling

thephysical and

functional

requirementsofthecustomer. 12

The quality

of a productorprocess

increases

asthe

number of

defects decrease.

A

measure ofthe

quality

oftheprocessorproduct

is

a

defect

perunit

(DPU).13

The

most commonapplicationof a

DOE

in

the

manufacturing

area

is

thatwhich

deals

with a part

being

produced ata

high DPU

(case

1).

Other

timesa process

is

producing

too

many

defects,

regardless ofthepart

being

produced

(case 2).

In

this

situation a representative part oftheprocess

is

chosen

for

_{the experiment,} andtheresults

oftheexperiment are relatedtoallparts produced

by

theprocess.

Case

threeaddresses

problems

before

they

happen.

By

way

of

DOE,

insight into

anew process can

be

gained.

Experimentation

willteach

how

thevariablesoftheprocesswillaffectthecritical

parameters oftheparts

being

produced.

Lastly,

before

purchasing

a new machine a

DOE

is

a good

idea in

orderto test to see

if

themachine

does

what

it is desired.

By

running

a

DOE,

themachinecan

be

tested

for

_{output, variability,} easeof_use,

set-up

time,

and

overallmachine performance.

Basic

General

Procedure14

There

aretensteps

in

a

designed

experiment.

They

are as

follows:

1. Brainstorm

2. Design

theexperiment

3. Obtain

materials and clean machine

4. Conduct

experiment/collect

data

5. Clean

the

data

6. Analyze

the

data

7. Interpret

theresults

8.

Confirmation

run

9.

Write

report

The

first

step

of

any design

of experimentproject

is

brainstorming. A

teamof

expertsshould

be

gathered

for

a

brainstorming

session

in

orderto

discuss

theproblems

associatedwiththeprocess at

hand.

The

teamshould consist of

different

skill

levels

including

operators, maintenance, engineers, managers,

andother experts.

Several

key

questionsneedto

be

answered

during

the

brainstorming

step.

These

include:

What

is

theproject goal?

What

is

theproject objective?

What

aretheoutputs/responsesoftheprocess?

What

arethe

inputs/factors

oftheprocess?

What

arethe

levels

ofthe

inputs?

Which

inputs

are

inter-related

toeachother?

What

parts/materialare

going

to

be

used

for

theexperiment?

How many

partscan

be

produced

during

the experiment? 15

The

brainstorming

has

resulted

in

a

list

ofresponses

(outputs),

factors

(inputs),

and

levels

for

the

factors

oftheprocess.

Having

completedthis

first

step,

the

list

of

factors

and

responses

is

usedto

design

an efficient

experiment,

and create a

Design

of

Experiment

sheet,

step

2.

There

areseveral

different

experimentsthat

may

be

chosen,

but

thegoal

is

to selectthe

mosteconomical

design

thatwill renderthemost

information

abouttheprocess.

See

The

third

step

is

toobtainmaterialsand clean_{the machine,} which

is

pretty

self-explanatory.

Obtaining

materials

simply

means

ordering

and

receiving

the

desired

number

of parts

for

theexperiment.

Cleaning

themachine means

doing

any necessary

maintenance

oradjustmentsto themachine

before

experimentation.

After

brainstorming, designing

the

experiment, obtaining

_{the parts,} and

preparing

the

machine,

theexperiment

is

runatthe

levels indicated

onthe

Design

ofExperiments

sheet

(step

4). Parts

needto

be

tagged,

_recorded,and measured

for

theresponse.

If

the

response

is

quantitative

data,

then this

is

an

easy

task.

On

the contrary,

if

theresponse

is

attribute or qualitative

data

and requires

judgment,

then

measuring

theresponse

is

difficult

and not arecommended practice.

For

_example,

it is

simpletoobtainathicknessvalue

for

an anodic

coating using

apermascope.

If

therewereno

measuring

tools,

assigning

a

thickness toeachtest sample

by

visualmeanswould

be impossible.

To summarize,

attribute

data is

not recommended

for

analysis,

and should

be

replaced

by

quantitative

data

if

possible.

Bad

partswill

be

made

during

the

DOE. The

idea

of a

designed

experiment

is

to

change process parameters

in

orderto

induce

changes

in

the

final

part.

Both

good and

bad

partsareexpectedto

be

made

allowing

onetoseewheretheoptimal settings are

located.

Step

five

ofthewhole

design

of experiments process

is

cleaning

the

data.

Checking

the

accuracy

ofthe

data is

important

toensurethatmistakes

did

not occur

in

the

transmissionofthe

data. The

result ofthis

step is

a

list

ofthe

factor

settingsof each

Now

the

data

are

ready

to

be

analyzed and

interpreted (steps

6

and

7)

for

two

items:

1)

relationships

between factors

and responses and

2)

significant verses

insignificant

factors.

An

empirical

equation,

describing

the

relationship between

the

factors

andthe

responses,

is

also obtained

from

the

data

analysis.

This

equation will

be

usedtopredict

whattheprocesswillproduce atvarious

factor levels. The

"true"

functional

relationship

between

theresponse andthe

factors,

themechanistic

model,

is

oftentoocomplicatedto

allowparameter_estimation,

but it

can

be

approximated

by

anempirical

(polynomial)

model.

Step

eight

is

toconduct a confirmation run.

By

doing

this,

the

results,

theories,

and

suggestedoptimalsettings attained

from

the

design

of experimentand areverified.

Finally,

thepurposeofthe two

remaining

_steps, nine and

ten,

is

to

inform

the teamofthe

resultssothat theresults

in

awrittenreport and anoralpresentation.

A

planto

keep

the

JMP

Analysis

JMP

(Statistical Software for

the

Macintosh

from SAS

Institute

Inc.)

is

asoftware

packagecapableof

performing

the

DOE

analysis.

The

resultsto

look

at

from

the

JMP

output arethe

Summary

of

Fit,

the

Analysis

of

Variance

(ANOVA),

the

Parameter

Estimates,

the

Effect

Test,

the

Lack

of

Fit,

andthe

Leverage Plots.

Summary

ofFit

17

RSquare

RSquare

Adj

Root Mean

Square

Error

Mean

of

Response

Observations (or

Sum

of

Weights)

RSquare

(R

)

is

thecoefficientof

determination

and measuresthepercentofthe correctedtotalsum ofthe squaresthat

is

explained

by

allofthe terms

in

themodel

(except

for

the

intercept

term).

The

equationthat calculates

R2

is

given

by:

B2

=

Model

sum

of

squares/Correctedtotalsum

of

squares

B?=

SSModel/SS

C Total

The

R2value

is

constrained

between 0

and

1

.

Multiplying

R2

by

100

yieldsthe

DOE

Equation Prediction Rating. An RSquare

value of

0.

95-

1

.

0

is

desired.

The

higher

the

RSquare

valuethemore

"adequate"

RSquare

Adj

(R2

Adjusted)

adjusts

RSquare

tomake

it

morecomparable over

modelswith

different

numbers ofparameters

by

using

degree

of

freedom in

its

computation.

It

is

acalculation of meansquares

instead

ofsums of squares and

is

calculated

by

B2

Adj

=

1

-

Error

Mean Square

C

Total Mean

Square

where,

Error

meansquare

is found in

the

ANOVA

table

found

onpage

17-19.

C

Total

meansquare=

C Total SS/C Total DF

(C

Total

SS

and

C

Total DF

found in

the

ANOVA

table)

Root Mean

Square

Error

(Root

MSE)

is

an estimate ofthestandard

deviation

of

errors aboutthe

fitted

regression model

(random

error).

The

calculation

for

thisvalue

is:

Boot

Mean

Square

Error

=

^

ErrorMeanSquare

A

prediction withthe

least

amountof

variability

is desired.

Therefore

asmall

Root

MSE

The Mean

of

Response

is simply

themean of_{the responses,} calculated

by:

n

Yvi

Mean ofResponse

=

r^

n

where,

n=

The

numberofexperimental runs

y,=

The

i*response value

The

number of experimental

runs, n,

equalsthenumberof

Observations

(or

sum of

Analysis of Variance

(ANOVA)

18

Degrees

ofFreedom

(Model,

Error,

Total)

Sum

of

Squares

(Model,

Error,

Total)

Model Mean

Square

Error Mean

Square

F Ratio

Prob>F

"The

analysis ofvariance

is

a means

for

partitioning

thetotal

variability

ofthe

observed responsevariables

into

variouscomponentswhichcan

be

attributedto

known

sources."101

The

total

variability

is broken down into

theexperimental error

variability

andthemodelvariability.

The

experimental error

variability

representsthe

variability

withinthegroupsofresponse values.

The

model

variability

is

the

variability

due

to

changing

the

factor levels

and

it

representsthe

variability

acrossthe three

factor

levels

(high,

medium, and

low).

Notation

for

the

ANOVA

table:

n=

The

number of experimentalruns

p

=

The

_{numberof modelparameters}

yi=

The

i*responsevalue

y

=

The

averageofthenresponse values

*yj

=

=

The

summation

from i=l

ton

(over

allthe

responses)

Table 3. The ANOVA Table

Source

of

Variability

Degrees

of

Freedom

Sum

of

Squares

Mean

Square

Model

Error

Total

p-1

n-p

2XV

y)2

Z(yi-yf

(p-l)+(n-p)

=

n-1

Kyi- y)2

OR

KV

y)2/(P-i)

Itory?/(M>)

Source

ofVariability-

indicates

the specific componentof variability.

Degrees

of Freedom- representsthenumber of

independent

piecesof

information

used

toestimatetheparticularcomponentofvariability.

Sum

of

Squares

(SS)-

is

_{thenumericalestimateofthecomponent of}

variability

(unadjusted for

the

degrees

of

freedom). It

is

thesum ofsquares ofthe

difference

between

the

fitted

response andtheactualresponse.

Mean

Square-

is

anestimate ofthe

variability

contribution

from

the

corresponding

sourceof

variability

after

adjusting for

the

degrees

of

freedom.

In

additionthe

ANOVA

table

has

the

F

Ratio

andthe

Prob>F.

FBatio

=

Model

Mean Square

Error Mean

Square

It

estimatesthe

following

quantity:

Experimental variability

+

Factor variability

Experimental variability

The

larger

the

F Ratio

(the further

it deviates

from

one

in

thepositive

direction),

the

more evidencethere

is

ofsignificant

factor

effects.

The F Ratio

is

the

"F

Value"

for

the test statistic andthe

Prob>F

value

(the

p-value)

is

thesignificance

level

which are usedto test the

following

hypotheses:

Ho: No

factors have

aneffect ontheresponse or

Pi=P2

=

33

=

Pi

=

₀

H,: At

least

one

factor has

aneffect ontheresponseor

0iort32orp3or...

p\*0

wherethe

3s

arethecoefficients ofthemaineffects

in

theequationthatwillresult

from

the

DOE

analysis.

See

Appendix B

for

more

information.

A

"Prob>F'value <

0.05

indicates

sufficient experimentalevidencetorejectthenull

Parameter

Estimates19

Term

Estimate

Std

Error

t

Ratio

Prob>

1

1

1

Term

is

theparameter

in

themodel

being

estimated.

The Estimate

values areestimations ofthecoefficientsof themodel

found

by

least

squares.

For

_example,

Smut

=

p0

(intercept)

+

Pi

(free sulfuric)

+

p2

(pH)

+

p3

(t seal)

+

p4

(t

DI)

+e

where,

Po, Pi, P2, P3,

and

P4

aretheparameter_estimates, where

Pi

estimatesthe

free

sulfuric _effect,

P2

estimatesthepH

effect,

etc.

The Std

Error

(the

standard error ofthe

estimate)

is

thesquare root oftheestimated

variance oftheparameterestimate and

is

usedto

quantify

the

uncertainty

or

variability

in

theparameterestimates.

In

other_words,

it is

anestimate ofthestandard

deviation

The

t

Ratio (t

value)

is

computed as

follows:

t=

Estimate/Std

Error

The

hypothesis

that

is

being

tested

by

the test

statistic,

t,

is:

Ho: The

parameter=

₀

(model

_term

insignificant)

Ha:

The

parameter*

₀

(model

_term

is

significant)

If

"Prob>

1

11"<

0.05,

thenrejectthenull

hypothesis

andassumethemodelterm

is

Effect

Test

Source

Sum

of

Squares

F Ratio

Prob>F

The

"Effect

Test"

providesthesame

information

asthe

'Tarameter Estimates". It

is

atype

DI

statistic

meaning

it

presents apartial

partitioning

ofthemodel sum of squares.

The

individual

sumsofsquaresare saidto

be

partial

in

thateach sum ofsquares represents

theamount of

variability

the

corresponding

modeltermswouldexplain

if it

wasthe

last

termentered

into

themodel. 21

The F

Ratio,

"F

Value",

test statistic

is:

F

=

Sum

of

Squares (type

IID/DF

Error Mean Square

The

following

hypothesis

test

is

constructedto

determine

modeltermsignificance.

Ho: The

variability

explained

by

themodelterm

is

insignificant

Ha:

The variability

explained

by

themodelterm

is

significant.

If

"Prob>F"

<

0.05,

_thenrejectthenull

hypothesis

and assumethemodelterm

is

Lack ofFit

22

Source

Sums

of

Squares (Lack

of

Fit,

Pure

Error,

Total)

Lack

of

Fit

Mean

Square

Pure

Error Mean

Square

F Ratio

Prob>F

MaxRSq

The

lack

of

fit

analysis provides a

breakdown

ofthe error sumof squares.

The

errorsumsorsquares

is

made

up

oftwocomponentsof_variability,

lack

of

fit

errorand

pure error.

To

separatethetotalsum ofsquares

into

the

lack

of

fit

andpure error

componentsthereare

four

steps.

1.

For

each

distinct

factor

combinationwhich

is

replicated,

compute a standard

deviation,

s,

or_variance,

s2,

from

theresponsevalues.

If

thereare

k

distinct

factor

settingswith

replication, then therewill

be k

variancescomputed.

These k

variancesrepresent

k

estimatesoftheexperimental

variability

orpure error.

2.

A

"total"pureerrorsums of squares

is

computed as:

Pure

Error

Sum

of

Squares

=

(dfO(s2)

+

(df^)(s22)

+

(df3)(s32)

+ ... +

(df^(sk2)

where, dfk=

degree

of

freedom

for

k*

factor setting

Sk2 =

estimated variance

for

k*

3

.

The

lack

of

fit

sum ofsquares

is

obtained

by

subtracting

thepureerror sumof squares

from

the totalerror sum ofsquares.

lack

of

fit

sumof squares=

(total

errorsum of

squares)

-

(pure

_{error sum of}

squares)

4.

The

degree

of

freedom

associatedwithpureerrorand

lack

of

fit

are obtained

from

the

degrees

of

freedom

chart.

(The

pureerror

degrees

offreedomcanalso

be

computed

by

summing

the

degrees

of

freedom for

each

individual

variance estimate).

The

pure error and

lack

of

fit

sums of squares are

divided

by

theirrespective

degrees

of

freedom

toobtainthepureerror meansquare and a

lack

of

fit

mean square.

The

pure error mean square

is

an estimate ofthepure error variance.

However,

the

lack

of

fit

mean square

is

anestimateofthe sumofthe

"pure

error"

variance and a "bias"

component.

The

bias

component representsthe

bias

or errorassociatedwith

using

an

inappropriate

modelto

describe

the truerelationship.

Testing

the

bias

significance,

whichreflectsthe

lack

of

fit,

is done

withthe

following

hypothesis

test.

The

F

ratiotests that the

lack

of

fit

error

is

_zero, and

is

calculated

by:

F

Ratio

=

(Lack

of

fit

mean square)/(Pure error mean

square)

The

hypotheses for

thetestareas

follows:

Ho: The

model

is

_adequate, no

lack

of

fit

A

"Prob>F'

value <

0.05

implies

rejectionofthenull

hypothesis

or

lack

of

fit.

A

lack

of

fit indicates

thatadditional parameters should

be

addedto themodel.

The F

ratio estimates

Experimental variability

+

Model

bias

Experimental variability

The larger

the

F statistic,

themore evidencethere

is

ofa

bias

due

to an under-specified

model.

Max

RSq

is

themaximum

R2

that can

be

achieved

by

amodel

using

only

the variables

in

themodel.

It's

calculation

is:

MaxB2

=

1

-

SS

(Pure

Error)

Leverage Plots

n

Leverage

plots

graphically

illustrate

thesignificant parameters and atwhich

level

willproducethemost

favorable

response.

Essentially

the

leverage

plot

is

agraphical

display

ofthe

Effect

Test.

Dotted

confidence curvesontheplots

indicate

whetherthe test

is

significantatthe

5%

level

by

showing

a confidence region

for

the

line

of

fit.

If

theconfidence region

between

thecurves containsthe

horizontal

line

then theeffect

is

not significant.

If

the

KEY

PROCESS

PARAMETERS

The anodizing

process

has many

parametersthatare

important

to theproduction

of good parts.

One

ofthe

hardest anodizing

parameterstocontrol

is

thealuminum

quality

from

suppliers.

Because

anodic oxidation

involves

aconversionof thealuminumsurface

into

anoxide

coating,

the

alloy

and

its

metallurgical structure

have important

effectson

thecharacteristicsofthe

finished

surface.

Differences

in

coatingsarisewiththe

purity

of

the aluminum, the typeand

quantity

of

alloying

_{elements, type}of mill

product,

different

production

lots,

interchanged

manufacturer

lots,

typeof

fabrication,

or

different

temper/aging

treatments.

All

ofthese

factors have

significant effects ontheappearance

and

functional

propertiesofthe

finished

parts. 24

Properties

ofanodic coatingsthat are affected

by

alloy

composition

include

appearance

(color,

_reflectance, and

transparency),

continuity

(protectiveness),

abrasion

resistance,weight

density,

_porosity,

dielectric

strength,

andcomposition.

As

far

as

appearance, purealuminum willproducethemosttransparentanodic

coating

ofall

its

alloys.

Clear

anodized coatings

(not

dyed)

could

look

opaque, gray, gold,

tan,

or

brown

depending

onthemajor

alloying

element. M

The

aluminum

alloy

system assigns a

four-digit

numerical

designation

toeach

grade.

The

numerical

designations for

the

alloy

and cast

alloy

andthesuffix

designations

are

in

Appendix

A.

Racking

is

another

important

factor in

theproduction of goodanodic coatings.26

current

flows

to thepart

during

anodizing.

Rack

design

and partplacement ontherack

are

important.

A

goodrack

design

will

hold

parts_securely, conduct current adequately,

and

carry

a

full

load

without shielding.

For

themost

part,

racks aremade oftitanium

but

may

also

be

made of aluminum.

Aluminum

racks require

stripping

after each use.

Titanium

racks

last longer but

are more expensive and require

larger

contact area

because

oftheir

lower

electricalconductivity.

Part

positionmust allow

for

good

drainage

and

avoidance of air pockets.

Having

rackedthe

parts, processing

commences and parts aremoved

from

tank to

tank.

There

are

many

tanks

involved

in

the

anodizing

process

(Figure

1).

Furthermore,

thereare several

factors

of eachtank

like

_{concentrations,}

times, temperatures,

etc.that

have

considerable contributionsto the

final

products.

Adequate cleaning

is

the

first

requiredtankprocessoperation.

Because many

organic compoundswillacttoresist

etching

and

anodizing

_steps,

they

needto

be

removed.

Control

of cleaner

concentration,

temperature,

andoilaccumulation areall

necessary.

Following

the

cleaning

is

rinsing.

Actually,

thorough

rinsing

must

follow

each

chemical

step

in

thesequence oftanks.

The

requirements

for

therinsetanksare

clean,

flowing

waterand an overflow

lip.

Rinsing

may

be

singleormultipletankrinses and

they

may be

spray

or

immersion

tanks.

Deoxidizing

is

the

step

to

follow cleaning

and rinsing.

During

this

step

anacid

solutionatan elevatedtemperatureremoves nonuniform oxide

films

andcontaminants

Etching

is

a

step

that

may

or

may

not

be

used.

Its

purpose

is

toremovethenatural

shine and provide a

soft, matte, textured

appearance.

On

average

etching

is

a

3-5

minute

process

in

a

nominally

five

percent sodium

hydroxide

solutionat

90-120

F.

The anodizing

tanks

have

many

key

process_{parameters, the}

first

being

the

chemical concentration.

In

the

anodizing

tank,

thesulfuric acid solution

is

controlled

in

industry

at a nominal

fifteen

percent solution.

It

is important

that thetemperatures are

held

withina couple of

degrees in

ordertoproduce consistent

coating

properties.

Current

flow is

also recommendedto

be

controlled

in

therange

from

12-16

amperes/ft2.

However

many

plants operateat

fixed

voltages

instead.

Agitation

is

another

factor

essentialto the

process

in

ordertoprovide auniformsolutiontemperature throughout the tank.

Cathode

location

can

have different

effectsonthethickness'

oftheoxidecoatings.

The

closerthe

surfacesareto the

cathode,

thethickertheanodic

coating

will

be.

When

desirable,

dying

is

carried out next.

The

most

important

parameters

in

the

dye

tankarethe

dye

concentration,

pH,

andtemperature.

Agitation

is

neededto

keep

concentrations andtemperaturesuniform.

Also,

time

is

a mentionable variable.

Longer

immersion

timeswillpromote

deeper dye

penetration

into

theporesoftheoxide coating.

Contamination

is

one

last

parameter.

Impurities

suchas

aluminum, sulfates,

and

iron

affectabsorptioncharacteristics and

dye life.

The

last important

chemicaltank

is

thesealtank.

Without sealing

parts,

they

become

subjectto

lower

corrosion

resistance, staining,

and

bleeding.

The

important

Table 4. Summarization

of

Key

Process Parameters

Metal

Quality

Impurities

Alloy

Composition

Processing

Temper/aging

treatment

Cleaning

Tank

Cleaner Concentration

Temperature

Oil Accumulation

Time

Anodizing

Sulfuric Acid

Concentration

Aluminum Concentration

Temperature

Current

Agitation

Cathode Location

Time

Sealing

pH

Nickel Concentration

Fluoride Concentration

Temperature

Agitation

Contaminants

Time

Backing

Electrical

Contact

Rack Design

Part Placement

Rack Material

Binse

Tanks

pH

Temperature (for

some

tanks)

DEFECTS

Because

ofthe

complexity

ofthe

anodizing

process, thereare

many

possibilities

where

defects

can occuronthe

line. Different defects

thatoccur

include,

lost

_contact,

smut,

overanodizing,

white_spots,

bleed

out,

bent

parts,

crashed_parts,

burnt

_parts, and

staining.

Causes for

these

defects may be due

to

materials, manpower, methods,

machines,

ormeasurements.

The

key

process parameters ofthe

anodizing

processwere

discussed in

the

previous section.

The

possible

defects

thatcan occur whentheseparameters are not at

theiroptimumare

discussed

next.

Poor racking

or poorcontact

due

to

insufficient

rack-contact areaor

loose

contactscan cause

iridescent

appearanceonclear_parts,

blue

appearanceon

black dyed

parts,

powdery coatings,

burning,

andotherproblems.

Operator's

technique and proper

rack

design play

an

important

role

in

producing

good contact.

Cleaning

wasthenextvariableto theprocessthatwas considered.

When

the

cleaner concentration

is

too

low

white spotsor

staining may

result.

A

dried-on foam

pattern

may

result

if

the temperatureofthesolution

in

the tank

is

too

high.

Overly

vigorous agitationproduces excessive

foam

whichstays ontherack and parts.

In

the

deoxidizing

tank,

white

spots,

film,

or"smut"result when allthe

contaminatesare not removed

from

the surfaceoftheparts.

The

same

thing

can

happen if

solutiontemperature

is

too

high,

caustic

burning

results.

Caustic

burning

is

anon-uniform

etch patternthat

is

a rejectable product condition.

Many

factors

are

involved in

the

anodizing

tank.

Too

high

ofasulfuricacid

concentration

may

cause

smutty,

overanodized partsor

burnt

parts,

andtoo

low

of a

sulfuric acid concentration

may

cause white spotted parts.

The

concentration ofthe

aluminum

in

thetank

is important for

the

conductivity

that

is

necessary in

theoxide

film

formation.

Too

high

ortoo

low

aconcentration can cause

overanodizing

or

underanodizing

respectfully.

Extended

time

in

the tank

may

result

in

smutontheparts.

High anodizing

temperatureswillproduce asofter

coating,

leading

to

dye bleeding.

Low

current could cause white spots or

dye

bleeding

as well.

Air

agitation

is necessary

to

prevent part

burning.

Generally,

the

dye

tank

is

a

low

maintenance

tank,

andtherefore tank

life

can

be

many

years.

As

contaminants

increase

over

many

_{years, the}

dye

"spoils"

andthiswill

cause

defective

colored parts.

Black parts,

for

example,

would

have

a

blue

tint.

Concentrations

and pH

play

a role

in

getting

the

right

colortoo.

Sealing

theparts

is

a place

in

theprocesswherenumerousthingscanand

do

go

wrong.

Low

nickelconcentrations and

low

pH

leads

to

dye

bleeding.

High pH,

nickel

precipitation, andtoomuchtimeyield

smutty

product.

On

the

contrary,

low

temperatures,

pH,

and

sealing

timeproducean

inadequate

seal.

Figures 4-12 graphically

illustrate

possible

anodizing

defects

andthe explanation

Racking

Rack

Design

Operator's

Technique

Rack

Welds

Flight

bar

Contact

Figure 4.

Causes for

Lost

Contact

Temperature

oftheanodizetank too

high

Sulfuric

acid concentrationtoo

high

Current

density

too

high

Anodize

Tank

New

seal- too

active

Machine

quits

-parts

in

too

long

DIH2Otoohot

Not

enough

Ni

in

thesealant

Not

enough

Ni

in

thesealant

pHtoo

high

In

the tank too

long

Temp,

of

black dye

too

high

In

the tank too

long

Sulfuric

acid

concentration too

high

Figure 6.

Causes for Smut

iMffiflQWffljw^^^,.v...v.,.x.. ... ...w4s#ss^vr*-**XKKga

^ '

' 1 ' r '

Cleaner

Concentration

too

low

Sulfuric Acid

Concentration

too

low

Current

too

low

Lost

Contact

Black

die;

rectifier not

turnedon

Dead

holes

not pluggedor

not

facing

down

'r

'r 1'

Sulfuric Acid

gets

in

the

dead

holes

Salt

dries in

the

dead holes

Figure

8. Causes for

Bleed

Out

'

'' ^r

Operators

Cleaner Low

Computer

Lack

of

Maintenance

Airline

pushed out

Hoist

misses

the

flight bar

Figure 10.

Causes for Crashes

I

1

liKpurnt

Parts

'7-.

~.

??

^%<<CW-:MIIIIIIMMMMtlllllllllMIIIIIIIIIIIIIMIIItlMIIIIM MMIIIIlT:

Sulfuric Acid

Concentration

too

high

^SS^w^^^^^^^^a^^gg^&^g^fi&iw^^^

Air Agitation

Air

not

turnedon

Current

too

high

Airline

loose

Parts

in

theair

too

long

Cleaner Low

A

paretochart analysiswasusedto

determine

the

top

defects

oftheanodize

process.

A

pareto analysis

is

atechnique

for

prioritizing

typesor sources of problems.

It

"separates

the 'vitalfew'

from

the

'trivial

many'

and provides

help

in

selecting

directions

for

improvement."28

These

data

werecollected over atimespan of oneyear

based

on

customer complaints.

Table 5. Customer Complaints

Defect

#of

occurrences

Percentage

of

occurrences

Percentage

of

defective

parts

Over

anodized

(film)

2

12.5

%

31.4%

Scratches

3

18.8

%

21.6%

Smut

6

37.5

%

20.7

%

White

Spots

2

12.5

%

15.2

%

Masking

1

6.25

%

4.5

%

Bleed

Out

1

6.25 %

4%

Rack

Marks

1

6.25

%

2.6

%

See figure

13

for

thegraph

showing

thepareto analysis.

Based

onthenumber of

occurrences, smut

is

the

biggest

problemand accounted

for

sixout ofthesixteen reported

defects

thatreachedthecustomer.

In

addition

it

was

in

the

top

three

for

thenumber of

c

Q.

E

o

o

E

o

Another

setof

data

wascollected onthescrap/rework

for

_{ten weeks,} andthis

data

is

represented

below in

table

6

and

figure

14.

Table

6. Scrap/Rework

Defect

#

of occurrences

Percentage

of

occurrences

Percentage

of

defective

parts

Lost Contact

25

73.5

%

43%

Color

1

2.9

%

20%

Crash

1

2.9

%

17%

White

Spots

1

2.9

%

11%

Bent

3

8.8

%

5%

Dirty

Parts

2

5.9 %

2%

Racking

1

2.9

%

1%

From

thisscrap/rework

data for

tenweeks

it is

apparentthat

lost

contact parts are

thenumberoneproblem with

both

thegreatestnumberof

defective

parts andoccurrences.

This

data

was

determined

to

be incomplete because

everything

wasnot

being

recorded,

especially

thereworkedparts.

For

exampletherewere

large

lots

of

smutty

parts

many

of

8

i t

o

t-xt) I

03

m

w 3 n

O

c o

O 55

o

o Q.

CO

o o

S

o o

w to o> ;_ *

to

i i Mr

0

G>

CM

O

0)

o

ii

Q O

5 "5

01

i

o

p

DESIGN

OF

EXPERIMENTS

Pre-experiment

Reduction

of

defective

parts

is

thegoal oftheproject.

The

questionof which

defect

to tackleneededto

be

resolved

by

brainstorming.

Brainstorm

The

first

step

ofthe

DOE

process

is brainstorming.

For

the

brainstorming

session

a

group

of experts

consisting

oftwoproduction_{supervisors, the}most experienced

line

operator, maintenance, the

department's

quality

coordinator, a chemical_engineer, andtwo

other

anodizing

experts attended.

Project Objective

Many

objectiveswereto

be

accomplished

during

theteammeeting.

The primary

goal ofthe

meeting

wastoagreeona project

objective,

andthisgoal was met.

The

project objective

is

to test the

hypothesis

that smut

is

a

function

of six

factors:

anodize

temperature,

seal

temperature,

hot

DI

rinse

temperature, free

sulfuric acid concentration

in

Besponse Variable

(Output)

A

combination of

factors

was

involved in

deciding

ontheproject_{objective, the}

customer complaint

data,

scrap/rework

data,

operator

interaction,

andteammembers

consensus.

Before

the team meeting,

lost

contactappearedto

be

theobvious choiceto

base

theexperiment on

because

ofthescrap/rework

data

and pareto analysisthat showed

25

occurrences

during

the tenweek

interval

of

data

collection.

However,

therewere

severalreasons

why

thiswas not selected

by

the team.

First

of

all,

lost

contact

is

a

problemthat

is

well understood.

Basically

it is

a

racking

problem

due

to theracks

themselves,

the operators, orthecontacts as shown

in

previous section

(Figure

4).

Secondly,

therewereno

lost

contact partsthatslippedthrough to thecustomer

in

1994.

The

problemselected was

smut,

the second

largest

problem.

Because

smut

formation

depends

upon

many

factors

such_{as, chemistries,}

times,

and

temperatures,

the

learning

potential was

immense.

With

the

complexity

ofthesmut problemcomes

many

conflicting

opinions, that need resolution.

Furthermore,

it

wastheproblem withthemost

occurrences of

dissatisfied

customers.

Six

of sixteen

(37.5%)

complaints were received

due

tosmut

in

1994,

meaning

therewere six occasionswherecustomers received

smutty

parts.

Also,

operators admittedthat

they

were

having

a

large

smutting

problem,

thatwas

not recordedonthe

data

log

sheets.

For

_{these reasons,} smutwas chosenastheresponse

Factors

(inputs)

In

the

brainstorming

session,

the

key

parameters oftheanodize process which

may

be

involved

in

producing

smutwere

determined

as

follows:

Table 7. Possible Variables

that

Cause Smut

(Brainstorm)

Seal

Time

Temperature

Concentration

of

fluoride

Concentration

of nickel

Contamination

. pH

Age/activity

Anodize

Time

Temperature

Sulfuric

acid concentration

Aluminum

concentration

Current

density

Black

Dye

Time

Temperature

Material

Alloy

Composition

Table

7

wasthenarranged

into

constants and variables

for

theexperiment and

is

shown

below in

table

8.

Table

8.

Experiment Variables

and

Constants

(First

Draft)

Variables

Constants

Seal

temperature

Material

Anodize

temperature

All

times

Hot DI

rinsetemperature

Black

dye

tankconditions

Sulfuric

acid concentration

Current

density

Fluoride

concentration pH

hot

DI

rinse

pH of theseal

Contamination

Age/activity

ofthe seal

Aluminum

concentration

in

the anodizetank

Concluding

theresults ofthe

meeting,

a

full

project objective was

developed.

The

hypothesis

to

be

tested

is

thatsmut

is

a

function

of six

factors:

anodize

temperature,

free

sulfuric acid concentration

in

theanodize

tank,

seal

temperature,

seal

pH,

fluoride

concentration

in

theseal

tank,

andthe temperature

in

the

distilled

water rinse afterthe

seal.

The material,

thenumber of

parts,

andthe

levels

of

input

were

going

to

be

Revelations Since

the

Brainstorming:

Material

Four

questions

pertaining

to thematerial/partsselectionneededto

be

addressed so

thatmaterial could

be

ordered

in

time toruntheexperiment.

They

were:

1)

which

aluminum

alloy,

2)

what

form:

couponsor actual production_parts,

3)

how

many

pieces,

and

4)

how

fast

cantheparts

be

obtained.

A 6000

series

aluminum,

6061,

was chosenasthematerial

because

it is

the

"purest"

aluminumanodized atthecompany.

An

actual production part was chosen over

coupons

because

of unknowntempersthatareunrepresentative of production parts.

A

windowpart was selected

because

1)

it is

madeof

6061

_aluminum,

2)

smut problemsare

occurring

with_{these parts,}

3)

high

production

lots

are run ofthese

parts,

4)

thereexists

1000

partsavailable

free

ofcharge.

Furthermore,

the

theory

wasthat thechosenparts

were sensitiveto theprocess

indicating

whentheprocess wasoutofcontrol.

Therefore,

if

the smutproblemcould

be

solved

for

these

parts,

then

it

could

be

resolved

for

any

other

parts aswell.

Number of

Parts

Calculations

showedthatat

least

onerack of

seventy

twoparts neededto

be

produced per runto

be

successful.

The

calculation was

based

ontheminimum a