Appunti Statistica per la sperimentazione e le previsioni in ambito tecnologico (Parte 2)

EXAMPLE 23

FULL FACTORIAL DESIGN first interactions

aster second auber

interaction

~

↓

thuree columns

to

related

Vector

three effects

main

P 2

= 1 dequee

1)

(p-11(p-1)(p tweedom

ABC = > of

=

-

two levels

(p-1) x(q-1) begwee greedom

1

AB of

= >

=

two

into

diwide

We block and

=> => ABC

AB + 1

1 =

= -

two

table besign

Limited guastional equal

in

to factorial design

quel I -ABd

- 1 =

- I

AB 1m

=

= In

+1 I tuials

ABG 4

+ =

= that

- This

named is would

DEFINING CONTRAST a

: to build

usegue

is design

faction factorial

to

desporting quee

Design

gastorial black cowespent

imum every

max

interaction to level ABe

of

a

I

L the

I lose possibility

to-Be =>

A equal

is 2m6

instead in estimate

to ABS

equal

block is

A destant

because is

ABe

#

to Be block

in eace

odvantages I

if esse

: to

lave

/

AB2 make

1

=+

effect

fuials tuials

only far

,

27128x127 e

Four trals linstead

example = > of

So

: many the

I reduce number

=>

(3_16

I tuials

tials

make guartien of

can a 23

23-1

this tuials

I

this

In quational

=> case use _

factorial design

I black

EFFECT When quational I

CONFOUNDING and

* if

= use use

to tuials

estimate fact

It the

bue extract and

I

B0 is

A + some

.

divibe block

in m o re

to this tor

only

several

veston effect

covesmond not

column

= column

in

one case a

a ,

,

two

verten cavespand effects .

Hp BG negligible

is mull

A >

Be

+ du

=

degrees levels

/2

effects

estimate three

greedom black only

of your

3

=> can

we interactions

Hluree two and

effects

main effects

main one

o %

If effect

SSto negligible main

null

=> 100 4-5

or

= [0 1) not relevant and

p-value

=> , not negligible

null au significant

>

=

01

.

os 0

0 ,

V

m u m m u n

mum 1

33

0 , 0 10

1 1 0 85 -O

,

. elevant =

null au

negligible

A I

significant

not

EXAMPLE

ELECTRIC PLANNING

EXPERIMENTAL

WELDING -

Kogofletis 1009) design

matuix

In this data

and the

example and

Wynn of exp

,

, to

table experiment

illustrate the best

find conditions

Inext whose is

purpose

an two

welking plates the

by

variable

electrical The given

is

from

of

You respanse

. Kg/mm'

by

the measured

effectiveness pauts

weeded There

mechanical gauce

you .

, I

two method nuelveating

at levels single

considered all weaving

gactos

a

a re : ,

,

, 1

.

to

peleated

/ %

pelveated

sombitions not

D 100

, data

Experimental Matrix observational

and . I

matuix DoE estimate

of cam

to

cowesponding factors

-

V ↓

effects

factaus

- du L effects

- main main

du some

effects and 2-3

interactions

T

to obtain S

tuvals starting from * -4

factos lave

↑ :

(

we

= %"Guational

otuials e

27 -

2x- P /am defining

lot contrast

=> wing of

- a

Wevem/comaiden Ba

I A

ABS

=+ =

B Ad

=

C AB

= patterm/structure

I confonding

B&

A

ABG =>

= -

- =

B Ad

= - AB

C PATTERN

ALLAS

CALLED

= - mo

13-1 design

understand

to

situation

With confonding

not

complex

=> is

move easy

so

hatteum/structure to interactiem

because consider the

have maximum

we .

How ...?

solve

cam

we

M e

two

& because NAB

For

=> example

FRACTIONAL FACTORIAL

Lisadvantage effects

I

When compounding

lave

factorial Design I

geaction gull

make of

- I to

several

each effects estimate

solumn vesta able main

measures a

am

.

I BC

effect pde mull

is

if if negligible

a

suppose au

cam .

* EXAMPLE

WELDING the

My is

response to

have

I

that

stwengle to

after

not welding

two

the

Divide lastwe .

27

Is satured

vestous

lave las

Design

= degwee gueeform

column

We f of

s

a no

.

extimate the

sufficient to vestos

DoE

included

because in solumn

G

eurou my

,

lave

I degree greedom

and I of . to estimate

I bequee each

assigning gueedom

=> BC.

&,

A

AB

of

on e

am ,

to the I

set that

Another investigate

must

identify fastaus

is

aim of .

I interested interaction

first auder

in

only

am .

vector a effect

to

The column tor

study

mot stuby

used but

is

=> main - am

a

interaction AxB

.

Table become

D column s

>

1

3 my

=

. FH

A Be DE

+ + +

+ . . . negligibled

the h ave

interactions >06

The P-value

m u st to

Assignment the

between lave considering

column vestous be Jame DoE .

RESOLUTION effect

CRITERIA main

We comaiden R- 2

-1 = with

I have design effects

all

resolution

Weven confonded

main

of I

a are

interaction

auber

first upneu

dur .

interaction

first

2 is auter du upper .

If I lave RI 13 interaction

auber

second upper

au

R

I V =

If I

lave 4

1 =

-

= This fliz interaction