2015.11.30.

Industrial Quality
Assurance
Topic: Factorial Experiment
Design

Dr. Gyula Varga

associate professor

I. Task solving for the form of exponential

function
Task 2:
Let us determine how the technological parameters
(v cutting speed,
f feed rate,
a depth of cut)
effect on tool life (T) of a turning tool having P20 insert!
Solvation: Let approximate the tool life in the form of
T=Cvfa (1)
where:
C unknown constant
, ,  unknown exponent

Let the logarithm of Equation (1):

ln T=ln C +  ln v +  ln f +  ln a (2)

2/46 Industrial Quality Assurance

1
 2015.11.30.

Let denote with ỹ= lnT, ß0=lnC, ß1=, ß2=, ß3=,

so Equation (2) will be: ~
y  β 0  β1~
x1  β 2 ~
x 2  β3~
x3
Using the transformational formula
 2(ln~x i  ln~x if ) 
x i   ~ ~  1
 ln x if  ln x ia 
We get the polinom:
y=b0+b1x1+b2x2+b3x3

Task is determination of the constants (b0; b1; b2; b3)

of the polinom!

3/46 Industrial Quality Assurance

Levels of „variation interval” and

number of experimental set up.

„variation interval” Cutting speed: Feed rate: Depth of cut:

and
number of ~ ~ ~
x 3  a[mm]
x1  v[m/min] x 2  f[mm/rev]
experimental setup

~
x 1 ln~x 1 ~
x2 ln~
x2
~
x3 ln~
x3
Zero level x i  x io  0 141 4,949 0,125 -2,079 1,25 0,223

Variation
interval Δxi  1 85 4,443 0,075 -2,590 0,75 - 0,288

Lower x i  x ia  1
56 4,025 0,050 -2,996 0,50 - 0,693
level
Upper level x i  x if  1
226 5,421 0,200 -1,609 2,00 0,693
4/46 Industrial Quality Assurance

2
 2015.11.30.

Measuring results:
Experimental
setup X  [x ki ]nf Y  [y kl ]nm Y  [ y k ]n
[n]

x0 x1 x2 x3 T1 T2 T3 lnT1 lnT2 lnT3

yk y  lnTk

1 + 95 156 132 4,554 5,050 4,883 127,7 4,83

- - -
2 + + - - 25 31 23 3,219 3,434 3,135 26,33 3,26

3 + 135 129 85 4,905 4,860 4,443 116,33 4,74

- + -
4 + + + - 14 16 22 2,639 2,773 3,091 17,33 2,83

5 + 162 264 185 5,088 5,576 5,220 203,67 5,29

- - +
6 + + - + 143 215 170 4,963 5,371 5,136 176 5,16

7 + 10 8 12 2,303 2,079 2,485 10 2,29

- + +
8 + + + + 124 68 45 4,820 4,220 3,807 79 4,28
5/46 Industrial Quality Assurance

n 8

 ln y k
4,83  3,26  4,74  2,83  5,29  5,16  2,29  4,28
b0  k 1
  4,09
n 8
n 8

x k1 ln y k
 4,83  3,26  4,74  2,83  5,29  5,16  2,29  4,28
b1  k 1
  0,20
n 8

n 8

x k2 ln y k
 4,83  3,26  4,74  2,83  5,29  5,16  2,29  4,28
k 1
b2    0,55
n 8

n 8

x k3 ln y k
 4,83  3,26  4,74  2,83  5,29  5,16  2,29  4,28
b3  k 1
  0,17
n 8

4,09 
 0,20
B  [b i ] n   
So  0,55
 
0,17 

6/46 Industrial Quality Assurance

3
 2015.11.30.

Function in between Tool life and technological data is:

y = 4,09 - 0,20x1 - 0,55x2 + 0,17x3 (3)

Let transform back Equation (3) into the system of natural dimensions.

xi  2
ln~ x if 
x i  ln~
1
ln~
x if  ln~
x ia

ln~
x 1  5,421 ln~
x1  5,421  0,698 ln~
x1  4,723
x1  2 1  
5,421  4,025 0,698 0,698

ln~
x 2  (1,609) ln~
x 2  1,609  0,694 ln~
x 2  2,303
x2  2 1  
(1,609)  (2,996) 0,694 0,694

ln~
x 3  0,693 ln~
x 3  0,693  0,693 ln~x3
x3  2 1  
0,693  (0,693) 0,693 0,693

7/46 Industrial Quality Assurance

0,20 0,55 0,17 ~

ln~
y  4,09  (ln~
x 1  4,723)  (ln~
x 2  2,303)  lnx 3
0,698 0,694 0,693

ln~
y  4,09  0,287ln~
x1  1,353  0,793ln~
x 2  1,825  0,245ln~
x3

lnT  3,618  0,287lnv  0,793lnf  0,245lna

e 3,618  a 0,245 37,263  a 0,245

T  0,287 0,793  0,287 0,793 , min
v f v f
Verification with basic level:
37,263  1,25 0,245 37,263  1,056
T   49,465 min
1410,287
 0,125 0,793
4,138  0,192

8/46 Industrial Quality Assurance

4
 2015.11.30.

Fractional (partial) experiment design

In the full factorial experiment design the number of experimental setup
greatly increases the number of coefficients to be determined in the
linear model.
The full factorial experiment design greatly redundant related to the
experimental setup.
It is evident the idea, that let reduce the number of experimental setup in
that way, we can lost only that type of information which is not substantional
very much to construction of linear model.

This type of experiment design is called to:

fractional (partial) experiment design.

9/46 Industrial Quality Assurance

Fractional (partial) experiment design

Reduction of number of experimental setup

Let start from the sipliest form, from the 22 full factorial
experiment design, which design matrix is:

10/46 Industrial Quality Assurance

5
 2015.11.30.

Dependent
Number of variable
Factors
experimental setup
yk

x0 x1 x2 x1x2 (x3)

1 +1 -1 -1 +1
~y
1

2 +1 +1 -1 -1
~y
2

~y
3 +1 -1 +1 -1 3

4 +1 +1 +1 +1 ~y
4
11/46 Industrial Quality Assurance

We have to determine, four parameters and the shape of the result is:
y=b0+b1x1+b2x2+b12x1x2
non full second order equation.

If we have experient related that, wether the process can be described

with lenear model over the chosen variational interval, than
it is enough to determine three parameters (b0, b1, b2).

In this case an unused degree of freedom remained, which give us the

possibility
to involve an additional factor into the linear model,
without
increasing the number of experimental setup.

In the design matrix we leave the first three column of the factors without
any change (x0, x1, x2) and the x1x2 product column vector will belong to
the new factor x3.

12/46 Industrial Quality Assurance

6
 2015.11.30.

If the effect of any factor depends on the level of any other

factors, in that case we say, in between the two factors there is an
interaction.
In practice it is distinguished two, three, etc. factorial interactions.

However here the coefficients of the polinoms will not be independent

ones, as it were in the case of 2f type full factorial experiment design.

The mixtures:

b1  (ß1+ß23), b2  (ß2+ß13), b3  (ß3+ß12)

This does not creates any problem in our case, as we assumed linear
model, so the number of interactions is zero.
So, it is possible to examine the effect of three factors with four
experimental setup instead of eight (n=23=8) .

13/46 Industrial Quality Assurance

Law (thumb):
For reduction the number of experiments we have to order to
the new factor that column vector of the matrix which belongs
to the interactions that can be abandoned.

In this case the signs of the column determine the value of the new vector in
the experimental setup.

14/46 Industrial Quality Assurance

7
 2015.11.30.

2. Fractional replication

In this case, when four experimental setup is applied for the estimation of
three factor, we apply only the half of the 23 type factorial experimental
design.
We say, that we applied subdouble (half) replication.
If we put the x3 variable equal with -x1x2 - which can be used similarly well –
than we get the other half of the 23 type plan.
In this case the determined coefficients will be the collective effects
estimations as follows:

b1  (ß1-ß23), b2  (ß2-ß13), b3  (ß3-ß12) (2)

15/46 Industrial Quality Assurance

If we realize both subdoubled (half) replication, than we we can get separate

estimation for
the linear effects and
the interactions
as, in the 23 type full factorial experiments.
The combination of these two subdoubled (half) replications gives the
full factorial experiments.

With the same contexture increasing further the number of the involved
factors, in case of four factorial experiments the matrix of 23=8 experiments is
the half replication of the 24 type full factorial experiments.

The matrix of 8 experimental setup is the 1/4 replication of the 25 type

full factorial experiments, and 1/8 replication of the 26 type full factorial
experiments.

Some type of the fractional factorial experiment design, the number of experi
mental setup, compared by the full factorial experiments, are summarised in
Table 3.2.

16/46 Industrial Quality Assurance

8
 2015.11.30.

Table 3.2.
Number of Type of full factorial Type of fractional factorial Number of
factors experiment design experiment design experimental setup
2f Replication 2f-r n=2f n=2f-r
3 23 1/2 23-1 8 4
4 24 1/2 24-1 16 8
5 25 1/4=1/22 25-2 32 8
6 26 1/8=1/23 26-3 64 8
7 27 1/16=1/24 27-4 128 8
8 28 1/16=1/24 28-4 256 16
9 29 1/32=1/25 29-5 512 16
10 210 1/64=1/26 210-6 1024 16
11 211 1/128=1/27 211-7 2048 16
12 212 1/256=1/28 212-8 4096 16
13 213 1/512=1/29 213-9 8192 16
14 214 1/1024=1/210 214-10 16384 16
15 215 1/2048=1/211 215-11 32768 16
5 25 1/2 25-1 32 16
6 26 1/4 26-2 64 16
7 27 1/8 27-3 128 16

17/46 Industrial Quality Assurance

Generally say:

In those type of fractional experiments

where r pieces linear effects are substituted
instead of the place of interactions, the number of interactions
can be calculated by the

2f-r
formula.

18/46 Industrial Quality Assurance

9
 2015.11.30.

3. Choosing of subdoubled replication.

Generating relations and determining contrasts
It can be seen that the fractional factorial experiment plans
claim much less experiments, as full factorial experiments.

For establishment of mixing effects we have to determine the so called

determining contrast.

This is the product of columns, of which all the elements are

or (+1) or (-1).

In this way we get the both side of determining contrast.

In order to determine the mixing we have to multiply both side of determinin
g contrast with the given effect.

E.g. in composition of 23-1 type subdouble replication, let the value of

x3 =+x1x2.

19/46 Industrial Quality Assurance

No of x3=x1x2 No of x3=-x1x2
exp. exp.

x1 x2 x3 x 1x 2x 3 x1 x2 x3 x 1x 2x 3

1 - - + + 1 - - - -1

2 + - - + 2 + - + -1

3 - + - + 3 - + + -1

4 + + + + 4 + + - -1

20/46 Industrial Quality Assurance

10
 2015.11.30.

If we multiply the element of the x1x2x3 column than, the product of

x1x2x3 is +1 in all cases.
This is the determining contrast: 1=x1x2x3.

In order to determine, what is the effect of the variable x1 how it mix,

let multiply both side of the determining contrast by x1.

x1 = x12x2x3 = x2x3, as xi2 = 1 always true.

To the variable x2 is x2 = x1x22x3 = x1x3,

To the variable x3 is x3 = x1x2x32 = x1x2.

This mean that the coefficients of the linear equations will be the

b1  β1  β 23 
 estimations.
b 2  β 2  β13 1
b 3  β 3  β12 

21/46 Industrial Quality Assurance

If the value of x3 is –x1x2, than the determining contrast is –1 = x1x2x3.

Executing of multiplications we get the formulas written earlier.

The b1  β1  β 23  and b1  β1  β 23  relations,

 
b 2  β 2  β13 1 b 2  β 2  β13 2
b 3  β 3  β12  b 3  β 3  β12 

which show, that what effects are mixing with a given factor, we call it
generating relations.

The effectiveness of replications depends on the mixing system.

Generally those replications are the most effective ones, where the
linear effects are mixing with the highest order interactions.

We have to endeavour for choosing them.

22/46 Industrial Quality Assurance

11
 2015.11.30.

Task solving for fractional experimet design

Task 1:
Let us determine the common estimations of linear effects and
interactions (generating relations) of response functions of experiment whi
ch are executed by the 24-1 type subdouble (half) replications!

Solvation:
If we execute the 24 type full factorial experiment design, than
the number of experiments is
n = 24 = 16 .
If we apply 24-1 type subdouble (half) relations in our experiments, than
n = 24-1 = 23 = 8 experimental setup is enough.
In the case of the 24-1 type subdouble (half) relations 8 solutions can
exist.
Let us denote these with upper case letters of the alphabet:
A~x4= x1x2 E~x4= x1x3
B~x4=-x1x2 F~x4=-x1x3
C~x4= x2x3 G~x4= x1x2x3
D~x4=-x2x3 H~x4=-x1x2x3
23/46 Industrial Quality Assurance

No of x1 x2 x3 x4 x 1x 2 x 1x 3 x 2x 3 x 1x 2x 3 x 1 x 2x 3x 4
exp.
1 - - - - +
2 + - - - -
3 - + - - -
4 + + - - +
5 - - + - +
6 + - + - -
7 - + + - -
8 + + + - +
9 - - - + +
10 + - - + -
11 - + - + -
12 + + - + +
13 - - + + +
14 + - + + -
15 - + + + -
16 + + + + +
Industrial Quality Assurance
24/46

12
 2015.11.30.

No of x1 x2 x3 x4 x 1x 2 x 1x 3 x 2x 3 x 1x 2x 3 x 1 x 2x 3x 4
exp.
1 - - - - B+
2 + - - - A-
3 - + - - A-
4 + + - - B+
5 - - + - B+
6 + - + - A-
7 - + + - A-
8 + + + - B+
9 - - - + A+
10 + - - + B-
11 - + - + B-
12 + + - + A+
13 - - + + A+
14 + - + + B-
15 - + + + B-
16 + + + + A+
Industrial Quality Assurance
25/46

No of x1 x2 x3 x4 x 1x 2 x 1x 3 x 2x 3 x 1x 2x 3 x 1 x 2x 3x 4
exp.
1 - - - - B+ F+ D+ G- +
2 + - - - A- E- D+ H+ -
3 - + - - A- F+ C- H+ -
4 + + - - B+ E- C- G- +
5 - - + - B+ E- C- H+ -
6 + - + - A- F+ C- G- +
7 - + + - A- E- D+ G- +
8 + + + - B+ F+ D+ H+ -
9 - - - + A+ E+ C+ H- -
10 + - - + B- F- C+ G+ +
11 - + - + B- E+ D- G+ +
12 + + - + A+ F- D- H- -
13 - - + + A+ F- D- G+ +
14 + - + + B- E+ D- H- -
15 - + + + B- F- C+ H- -
16 + + + + A+
Industrial Quality Assurance
E+ C+ G+ +
26/46

13
 2015.11.30.

It can be seen in Table, that there can be 8 solutions for

the 24-1 type half replication from the experimental setup in
Table 2.
Sign of the 24-1 Number of half replication of experimental
type half setups in the Table of the 2 type full
replication factorial experimental design
A: x4=x1x2 2, 3, 6, 7, 9, 12, 13, 16
B: x4=-x1x2 1, 4, 5, 8, 10, 11, 14, 15
C: x4=x2x3 3, 4, 5, 6, 9, 10, 15, 16
D: x4=-x2x3 1, 2, 7, 8, 11, 12, 13, 14
E: x4=x1x3 2, 4, 5, 7, 9, 11, 14, 16
F: x4=-x1x3 1, 3, 6, 8, 10, 12, 13, 15
G: x4=x1x2x3 1, 4, 6, 7, 10, 11, 13, 16
H: x4=-x1x27/46
2x3 2, 3, 5,Quality
Industrial 8, 9, 12, 14, 15
Assurance

The dissolving power of these replications are different.

In the replications
A – F 3 – 3 factors are playing role in the determining contract,
while in replications
G – H these number are 4 – 4.
These last two replication have the maximum dissolving power.

Let us choose the replication signed G.

The determining contract of this is 1=x1x2x3x4.
The common estimations are determined by the next formulas :
x1 = x12x2x3x4 = x2x3x4  b1 (ß1+ß234)
x2 = x1x22x3x4 = x1x3x4  b2 (ß2+ß134)
x3 = x1x2x32x4 = x1x2x4  b3 (ß3+ß124)
x4 = x1x2x3x42 = x1x2x3  b4 (ß4+ß123)
x1x2 = x12x22x3x4 = x3x4  b12 (ß12+ß34)
x1x3 = x12x2x32x4 = x2x4  b13 (ß13+ß24)
x1x4 = x12x2x3x42 = x2x3  b14 (ß14+ß23)

28/46 Industrial Quality Assurance

14
 2015.11.30.

x1 = x12x2x4 = x2x4 b1 (ß1+ß24)

x2 = x1x22x4 = x1x4 b2 (ß2+ß14)
x3 = x1x2x3x4 = x3 b3 (ß3)
x4 = x1x2x42 = x1x2 b4 (ß4+ß12)
x1x2 = x12x22x4 = x4 b12 (ß12+ß4)
x1x3 = x12x2x3x4 = x234 b13 (ß13+ß234)
x2x3 = x1x22x3x4 = x134 b23 (ß23+ß134)
x1x4 = x12x2x42 = x2 b14 (ß14+ß2)

29/46 Industrial Quality Assurance

This type of mixturing give possibility for estimate

linear effects common with the three factorial interactions
and the
two factoral ones among each other.

As it can be seen the correct choose of factorial replications

claim a lot of endurance and work, but
by decreasing of number of experimental setup this work is
pay morefold off.

There is no other way.

In the case of use of fractional experimental design we punctually have
to know the mixing system, and we have to see clearly what kind of
information we loose.

30/46 Industrial Quality Assurance

15
 2015.11.30.

Task 2:
Let us determine how the Határozzuk meg, hogyan befolyásolják az R18
 clearance angle,
 rake angle and the
s width of margin
have effect on the tool life of a milling cutter
(material HSS, diameter:  22 )

Solvation:

Variables: ,  and s.
When the use of full factorial experiment design the number of
experimental setup is n = 23 = 8.
All the experiment should multiply with three times.
So the total number of experiments is 24 , however the experiments
should created with special grinding.

31/46 Industrial Quality Assurance

This requires a lot of time and work, and its cost is extremely high.
That is why we decide the application of fractional experiment design
.
In this way we have to do n = 23-1 =4 experiments and intend to examine
12 pices milling cutter.

At our decasion we supposed that the effect of ,  and s on tool life of

milling cutter can be described adequately by linear model.

The level of experimental setup, the measurement results, and the

constants of the polinoms is summarised in Table 3.

32/46 Industrial Quality Assurance

16
 2015.11.30.

3. táblázat
Variation interval variációs Clearance Rake Width of
intervallum and level of angle
experimental setup ~ angle
~ margin
x1  α  x 2  γ ~
x 3  s(mm)

Zero level 14 15 0,05

xixi,0=0 Δ~
xi
Variation
Δxi  1 4 6 0,03
interval
Lower 10 9 0,02
level xixi,a=-1
Upper 18 21 0,08
level xixi,f=+1

33/46 Industrial Quality Assurance

Let us search the response function in the form of:

T=ß0+ß1+ß2+ß3s, that is .
~
y  β 0  β1 ~
x1  β 2 ~
x 2  β3~
x3
Knowing the basic levels the variational intervals can be calculated,
which can be found in Table 4.

After transformation we get the transformed values of lower and up

per level of the factors. These can be found in tha Table too.

The searched polinom in the transformed system is:

y=b0+b1x1+b2x2+b3x3.
The design of experiments with the factor levels X=[xki]nf can be
found in Table 4.

34/46 Industrial Quality Assurance

17
 2015.11.30.

Table 4

Number of Y  [ y k ]n
X=[xki]nf Y=[ykl]nm
experimental
setup

x0 x1 x2 (x1x2)x3 yk1[min] yk2[min] yk3[min] y k [min]

1 + - - + 26,70 15,38 12,25 18,11
2 + + - - 43,85 47,47 40,90 44,07
3 + - + - 43,30 32,00 29,25 34,85
4 + + + + 30,75 29,50 38,15 32,80

Knowing of average values of measured results we calculate the the coef

ficients of the polinom by solving on normal equations.

35/46 Industrial Quality Assurance

y k
18,11  44,07  34,85  32,80
b0  k 1
  32,46
n 4
4

x k1  yk
 18,11  44,07  34,85  32,80
b1  k 1
  5,98
n 4
4

x k2  yk
 18,11  44,07  34,85  32,80
b2  k 1
  1,37
n 4
4

x k3  yk
18,11  44,07  34,85  32,80
b3  k 1
  7,00
n 4

36/46 Industrial Quality Assurance

18
 2015.11.30.

We transform back the response function into the system of natural

dimensions.

~
x1  14 ~
x  15 ~
x  0,05
T  y  3,46  5,98   1,37  2 7 3
4 6 0,03

T=149,8+1,5+0,23-233s.
The linear effects are mixing with the following two factorial interactions:

b1(ß1+ß23)
b2(ß2+ß13)
b3(ß3+ß12)

The mixing system can be verified on the base of Task 1.

37/46 Industrial Quality Assurance

19