DR.

ALVIN’S PUBLICATIONS

D ES I GN O F
E XPE R I M E N TS WI TH
M I NI TA B
DR. ALVIN ANG

1 | PAGE

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 CONTENTS
I. Introduction ......................................................................................................... 5
II. Simple Full Factorial Experiment (2 Levels)............................................................ 7
A. Creation of Experiment ...............................................................................................8
1. Display Available Designs ...............................................................................................................9
2. Create Factorial Designs .................................................................................................................9
3. Factors ..........................................................................................................................................10
4. Options .........................................................................................................................................10
B. Running the Experiments & Getting the Results (Response) ....................................... 12
C. Analyzing the Results ................................................................................................ 13
1. Pareto Chart .................................................................................................................................14
2. Normal Plot ..................................................................................................................................14
3. Factorial Plot ................................................................................................................................15
D. Predicting Results ..................................................................................................... 17
III. Definitions of DOE........................................................................................... 18
A. Factors / Levels / Responses ...................................................................................... 18
B. Run / Randomization ................................................................................................ 19
C. Blocking .................................................................................................................... 21
D. Replicate vs Repeat ................................................................................................... 23
E. What if many of the effects are significant? ............................................................... 24
F. Corner Point / Center Point ....................................................................................... 25
G. Full vs Fractional Factorials ........................................................................................ 26
H. Resolution and Confounding ..................................................................................... 29
1. Resolution III ................................................................................................................................31
2. Resolution IV ................................................................................................................................31
3. Resolution V .................................................................................................................................32
I. Screening (Plackett Burman Design) .......................................................................... 33
1. Using Assistant .............................................................................................................................34
2. Screening While Creating Factorial Design ..................................................................................36

IV. Response Surface Methods (>2 Levels) ............................................................ 37

A. Central Composite Design (CCD) ................................................................................ 37
1. Creating the Experiment ..............................................................................................................38
2. Explaining the PT Type .................................................................................................................40
a) Blocking within CCD .................................................................................................................41

2 | PAGE

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 3. Analyzing the CCD Experiment .....................................................................................................44
a) Factorial Plot ............................................................................................................................44
b) Pareto Chart and Model Equation ...........................................................................................47
c) Surface Plot ..............................................................................................................................49
B. Box-Behnken Design ................................................................................................. 50
1. Creating the BB Experiment .........................................................................................................51

V. Mixture Designs ................................................................................................. 52

A. What is a Mixture Design .......................................................................................... 53
B. MIxture Experiment Example .................................................................................... 55
1. Creating the Mixture Experiment ................................................................................................57
2. Predicting the Mixture Experiment ..............................................................................................61

VI. Taguchi Method ............................................................................................. 62

A. Concept .................................................................................................................... 62
B. Taguchi’s Optimization .............................................................................................. 64
C. Meaning of Signal to Noise Ratio (S/N) ...................................................................... 65
1. Larger is Better .............................................................................................................................66
2. Nominal is Best ............................................................................................................................66
3. Nominal is Best (default) ..............................................................................................................66
4. Smaller is Better ...........................................................................................................................67
D. Example: Minimizing the Response (Smaller is Better) ............................................... 68
a) Minitab Solution ......................................................................................................................69
(1) Creating the Worksheet ..................................................................................................69
(2) Defining the Factors ........................................................................................................72
(3) Analyzing the Responses .................................................................................................73
b) Minitab Output ........................................................................................................................79
c) Predicting Taguchi Results .......................................................................................................80
E. Summary of Taguchi’s Example ................................................................................. 83
1. Nominal Response (Nominal is Best) ...........................................................................................83
2. Largest Response (Larger is Better) .............................................................................................84

VII. DOE FAQs ....................................................................................................... 85

A. What if many / all factors become significant? ........................................................... 85
B. What if there's missing data? .................................................................................... 85
C. Must we use all factors? ............................................................................................ 85
D. What if I can't replicate all runs? ............................................................................... 85
E. Must I randomize all trials? ....................................................................................... 85
VIII. Successful Steps to Implement DOE ................................................................. 86

3 | PAGE

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
IX. Obstacles to DOE ............................................................................................ 87
A. Resistance Inertia ..................................................................................................... 87
B. Expensive Cost .......................................................................................................... 87
C. Lack of Management Support .................................................................................... 87
D. Lack of Training ......................................................................................................... 87
E. Wrong Thinking that DOE is only used for Manufacturing .......................................... 87
Appendix .................................................................................................................. 88
A. Selected Factorial Designs ......................................................................................... 88
1. 2^2 Full Factorial ..........................................................................................................................88
2. 2^3 Full Factorial ..........................................................................................................................89
3. 2^3 Half Fractional Factorial ........................................................................................................90
F. Selected Placket Burman Designs .............................................................................. 91
G. Selected Taguchi Designs........................................................................................... 93
H. Selected Mixture Designs .......................................................................................... 97
Bibliography ............................................................................................................. 98
About Dr. Alvin Ang .................................................................................................. 99

4 | PAGE

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 I. I N T ROD U C T I ON

There are 4 types of DOEs in Minitab.

5 | PAGE

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
• Factorial (2 Levels)

o Full Factorial

o Fractional Factorial

• Response Surface Method (RSM) (>2 levels)

o Central Composite Design (CCD)

o Box-Behnken Design

• Mixture

o Simplex Centroid

o Simplex Lattice

o Extreme Vertices

• Taguchi

o Larger is Better

o Nominal is Best

o Nominal is Best (Default)

o Smaller is Better

6 | PAGE

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 I I . S I M P L E F U L L FA C T O R I A L E X P E R I M E N T ( 2 L E V E L S )

Go to https://sigmazone.com/catapult-grid/

• We will test the following:

o Release Angle: 140 / 180

o Firing Angle: 110 (since this is constant, we will ignore it)

o Cup Elevation: 220 / 280

o Pin Elevation: 120 / 180

o Bungee Position: 120 / 180

7 | PAGE

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
• 4 Factors 2 Levels, this is known as a 24 Factorial Design DOE.

• Total number of runs = 24 = 16 runs

• Now go to Minitab

A. CREATION OF EXPERIMENT

8 | PAGE

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
1. DISPLAY AVAILABLE DESIGNS

2. CREATE FACTORIAL DESIGNS

9 | PAGE

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
3. FACTORS

4. OPTIONS

10 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 11 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 B. RUNNING THE EXPERIMENTS & GETTING THE RESULTS (RESPONSE)

• Go to https://sigmazone.com/catapult-grid/

12 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 C. ANALYZING THE RESULTS

13 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 1. PARETO CHART

• We see that the Main Effects are: A, B, C and D.

• Anything that surpasses 36.8 is an important effect.

• We see that all other interaction effects fall below 36.8 (thus unimportant)

2. NORMAL PLOT

• Normal plot shows us the same thing.

14 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
3. FACTORIAL PLOT

15 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
If there is a overlap / cross of the lines, there is interaction between the Factors.

However, we see that all 4 Factors have no relationship with each other.

16 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 D. PREDICTING RESULTS

17 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 III. DEFINITIONS OF DOE

A. FACTORS / LEVELS / RESPONSES

• Earlier, we did this test:

o Release Angle: 140 / 180 → LOW / HI

o Firing Angle: 110 (since this is constant, we will ignore it)

o Cup Elevation: 220 / 280 → LOW / HI

o Pin Elevation: 120 / 180 → LOW / HI

o Bungee Position: 120 / 180 → LOW / HI

• 4 Factors 2 Levels, this is known as a 24 Factorial Design DOE.

• Total number of runs = 24 = 16 runs

18 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 B. RUN / RANDOMIZATION

19 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
• Randomization is important to protect against uncontrolled and/or unknown influences of
variables that are not part of the experiment.

• In order to minimize this risk of unknown influence, experimenters randomly assign the order of
testing to improve the chances of averaging out this bias or distortion of the responses related to
the factor(s) under study.

• If you are unable to randomize due to physical or cost constraints, we will need to do Blocking.

20 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 C. BLOCKING

• If you cannot randomize due to lack of resources, example, not enough raw materials to do so
many runs, one may block.

• 1 Block = 1 Batch.

• Possibly, 1 Block = 1 Day. Which means, running an entire experiment in 1 day = 1 block.

• Running the experiment again the next day will be 2 blocks → Block 2.

• Blocking minimizes the risk of the nuisance-factor batches creating excessive estimates of the
inherent variation.

• For example, you want to test the quality of a new printing press.

• However, press arrangement takes several hours and can only be done four times a day.

• Because the design of the experiment requires at least eight runs, you need at least two days to
test the press.

• To distinguish between any block effect (incidental differences between days) and effects because
of the experimental factors (temperature, humidity, and press operator), you must include the
block (day) in the designed experiment. You should randomize run order within blocks.

21 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 22 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 D. REPLICATE VS REPEAT

• Replication = Performing more than one trial of each run

• (Completely NEW SETUP each trial).

• A Replicate is an Independent and Random application of the run, including the setup.

• Repeat = a Repetition of a run WITHOUT going through a NEW SETUP.

23 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 E. WHAT IF MANY OF THE EFFECTS ARE SIGNIFICANT?

• Check → Was the setup for each trial really randomly replicated?

• If not, the responses are repeats, not replicates.

• Replication assumes that each trial is an independent and random performance of the process,
specifically including any process setup.

• If not, the experimenter has repeats that may be neither independent nor random.

• The estimate of s (standard deviation), using repeats will be much smaller than the actual
inherent variation of the process.

• This will make most effects appear significant, when in reality they have not been randomly
replicated to properly estimate the inherent variation.

24 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 F. CORNER POINT / CENTER POINT

• There are 4 Corner Points:

o (-1, -1) / (-1, 1) / (1, -1) / (1, 1) → Each is 1 Run

• The Centre Point is (0,0).

• Currently, we don’t use it.

• We will explore Centre Point in the next section on Response Surface Design (RSM): Central
Composite Design (CCD).

25 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 G. FULL VS FRACTIONAL FACTORIALS

• Full Factorial means All Runs

• For example, 23 → 2 Levels 3 Factors = Total 8 Runs

• But if you lack resources, are you able to do it in just 4 Runs?

• Yes, but its labelled as Resolution III Design (see picture above, in Red → 4 Run, 3 Factors).

26 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
• You can see that the number of Runs increase dramatically as number of Factors increase!

• Thus we need Fractional Factorial to cut down the number of Runs (we can do Screening to
Screen out the unimportant Runs as shown later).

• Similarly, in the previous experiment (Create Factorial Designs) we did a 24 (2 Levels 4

Factors) which we have the option of running the Full Experiment (=16 runs… in which we
did…), or we could just do 8 runs.

27 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
• If we had chosen 8 Runs, this would be a resolution IV (see picture above, in Yellow)

28 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 H. RESOLUTION AND CONFOUNDING

• So long as you do not Run Full Factorials (which means, Run all possible combinations i.e.
perform ALL experiments), you will experience Confounding.

• Presume we have the above: 8 runs cut down to 4 runs….

• Notice that C and AB have the same effects! They are Confounded!

29 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
• Notice that B and AC have the same effects! They are Confounded!

• Notice that A and BC have the same effects! They are Confounded!

30 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
• Since they have the same effects, Confounding means that you cannot tell whether the
Response is due to

o A or BC?

o B or AC?

o C or AB?

1. RESOLUTION III

2. RESOLUTION IV

31 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 3. RESOLUTION V

• Confounding should be Avoided because we cannot differentiate which Factor is affecting

the Response.

• However, due to limitations of Resources, we need to Screen out the unimportant Runs.

32 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 I. SCREENING (PLACKETT BURMAN DESIGN)

• We already know (from Section: Full vs Fractional Factorials) that that the number of Runs
increase dramatically as number of Factors increase (>5 Factors)!

• Thus we need Fractional Factorial to cut down the number of Runs

• Screening helps to Screen out unimportant Runs and cut it down to a Fractional Factorial.

• In this manuscript, we consider 2 ways of Screening:

o Using Assistant

o Screening while Creating the Factorial Design

• Note that in this Manuscript we only consider Plackett Burman Screening Design.

• And if we use Assistant option in Minitab, we are actually doing Plackett Burman Screening.

• Plackett Burman designs are only Resolution III experiments AND 6 or more factors.

• It only identifies Main effects and ignore Interaction effects

• In other words, Minitab only performs Screening when we have >6 Factors.

• Else, Minitab will auto-generate the Fractional Factorial Options for us to choose from (while we
create the Factorial Design).

33 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
1. USING ASSISTANT

Note that Assistant Screening = Plackett Burman Screening

34 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
• C3 (CenterPt or PtType) stores the point type.

• If you create a 2-level design, Minitab names this column CenterPt.

• If you create a Plackett-Burman or general full factorial design, Minitab names this column
PtType.

• The codes are: 0 is a Center Point run and 1 is a Corner Point. (we will explain Centre Point
and Corner Point later in RSM – CCD, next section).

35 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
2. SCREENING WHILE CREATING FACTORIAL DESIGN

36 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 I V . R E S P O N S E S U R FA C E M E T H O D S ( > 2 L E V E L S )

• There are two types of Response Surface Methods (RSM)

o Central Composite Design (CCD)

o Box-Behnken

A. CENTRAL COMPOSITE DESIGN (CCD)

37 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
1. CREATING THE EXPERIMENT

38 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
As you can see frrom the above, we have many CCDs….

• Full / Half / Quarter / Eighth….

• We will not go through them

• More explanation here: https://support.minitab.com/en-us/minitab/21/help-and-how-

to/statistical-modeling/doe/supporting-topics/response-surface-designs/summary-of-
central-composite-designs/

39 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 2. EXPLAINING THE PT TYPE

How did we manage to achieve 13 runs?

• 5 Centre Points + 4 Axial Points + 4 Corner Points = 13 Runs in total

40 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 a) Blocking within CCD

• Previously, we saw that 5 Centre Points + 4 Axial Points + 4 Corner Points = 13 Runs in total

• But they were all run in 1 Block. (unblocked)

• Presume now there was Blocking (we wanted to Run the experiment in 2 days or 2 blocks).

• Why do we need 14 runs?

41 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 (this picture shows 14 runs = 7 Corner (Factorial) Runs and 7 Axial Runs)

• 14 Runs is because, when two blocks are required there should be a

o Factorial Block (Block 1 = 7 Factorial Runs) and

o An Axial Block (Block 2 = 7 Axial Runs).

42 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 43 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
3. ANALYZING THE CCD EXPERIMENT

a) Factorial Plot

44 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
• Notice the difference between the chart above vs the one in Section: Factorial Plot (ctrl click
the link).

• The Factorial Plot of a CCD is curved because Minitab generated many points to test per
Factor. E.g. Pressure = 220 / 732 / 114 / 838 / 476…

• If you compare this to a Factorial Plot of a Simple 2 points experiment (shown below)….
You see that its impossible to generate a curve with just 2 points….(high and low)

• Thus, CCD (RSM) is better than a Simple Factorial Experiment because it can detect
curvature (Non-Linear)

45 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
We note that there is a maximum Retention Time when the Pressure is arouund 500 and the
Temperature around 50.

46 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
b) Pareto Chart and Model Equation

47 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
• We note that Pressure is the Key Factor affecting Response Time.

• Temperature is insignificant.

48 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
c) Surface Plot

49 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 B. BOX-BEHNKEN DESIGN

• Note that a 3 Factor 2 Level BB Design minimum is 15 Runs

• Note that BB Design are ALL Centre Point Runs!!!

• In other words, BB is a special subset of the CCD (you can imagine it to be a Fractional
Factorial of a Full Factorial Design).

50 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 1. CREATING THE BB EXPERIMENT

• We shall not carry on with further experiments.

• We shall stop here at only creating the BB experimental design.

• This is because the rest of the steps are similar and can be repeated as above (see section:
CCD: Creating the Experiment (ctrl click the link))

51 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 V. MIXTURE DESIGNS

There are 3 types of Mixture Designs:

• Simplex Centroid

• Simplex Lattice

• Extreme Vertices

52 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 A. WHAT IS A MIXTURE DESIGN

53 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
• Picture above shows a Simplex Centroid (Unaugmented) with 7 Points (7 Runs)

• There are 3 Components / Factors (X1 / X2 / X3)

• X1 + X2 + X3 Must == 100% → They are ingredients (mixture) which make up a total

percentage of 100%.

54 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 B. MIXTURE EXPERIMENT EXAMPLE

• Go https://cusum.mx/en/simuladores/

55 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
The Machine Parameters (Fixed Constants) are:

• Extrusion: 80

• Cooling: 10

• Screw: 60

• Air: 0.5

• Puller: 2

The Raw Material Parameters (Mixture Ingredients) are:

• Polyethylene: 10 ~ 99%

• Additives: 0 ~5%

• Colorants: 1 ~ 4%

• Regrind: 0~75%

56 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 1. CREATING THE MIXTURE EXPERIMENT

An Augmented Simplex Centroid has 10 Points (see above picture).. thus it will produce a table of 10
runs…

57 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 58 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 59 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
• Recall: PtType stores the point type.

• The codes are:

o 0 is a Centre Point run and

o 1 is a Corner Point.

o I think -1 is an Axial Point

60 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
2. PREDICTING THE MIXTURE EXPERIMENT

61 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 V I . TA G U C H I M E T H OD

A. CONCEPT

• By right, a Full Factorial for a 3 factor 2 level (23) = 8 runs required.

• Taguchi can reduce the number of runs to only 4.

• This is possible because Taguchi assumes the 3rd column (3rd factor) to be the interaction
effect of factors 1 and 2, thus it can be ignored.

62 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
• Taguchi regards Factor Interaction Effects as Noise (thus they are insignificant and ignored).

• Taguchi only regards Main Factor Effects.

63 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 B. TAGUCHI’S OPTIMIZATION

• Precision relates to the Standard Deviation (SD)

• Target Value relates to the Mean

64 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 C. MEANING OF SIGNAL TO NOISE RATIO (S/N)

• S/N ratio combines the effect of the Mean and Standard Deviation into one value used to gage
the process output.1

• In other words, the larger the S/N ratio, the more robust the process is to Noise.

• As mentioned above,

o Precision relates to the Standard Deviation (Step 1 for Optimization). We use the S/N
ratio to identify those Control Factors that reduce variability.

1 https://www.amazon.com/Practical-Design-Experiments-DOE-Optimizing/dp/0873899245

65 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 o Accuracy relates to the Target Value (Mean) (Step 2 for Optimization). We identify
Control Factors that move the mean to target and have no effect on the S/N ratio.

• There are 4 types of S/N ratio in Minitab: (note that for all types, our objective is to Maximize
the S/N ratio → increase Signal, lower Noise)

1. LARGER IS BETTER

• Objective is to Maximize the Response.

• Example: Identify the Factors that Increase the hardness of the steel alloy.

2. NOMINAL IS BEST

• Objective is to try to hit the Target Response using S/N ratio based on Standard Deviations.

3. NOMINAL IS BEST (DEFAULT)

• Objective is to try to hit the Target Response using S/N ratio based on Means and Standard
Deviations.

• Example: Identify the Factors that allow the manufacture to find the nominal specification.

66 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 4. SMALLER IS BETTER

• Objective is to Minimize the Response.

• Example: Identify the Factors that Reduce the force necessary to open the sealed packaging.

67 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 D. EXAMPLE: MINIMIZING THE RESPONSE (SMALLER IS BETTER)

Time Temp Pressure With Noise 1 With Noise 2 With Noise 3

1.25 250 80 2.9 3 2.8
1.25 260 90 2.4 2.2 2.3
2.5 250 90 3.5 3.6 3.7
2.5 260 80 2.6 2.5 2.7

• This example is taken from (Durivage, 2016)

• A sealing process requires large force to open the packaging.

• 3 Factors, 2 Levels

o Time: 1.25 and 2.5 seconds

o Temperature: 250 and 260 Degree Celsius

o Pressure: 80 and 90 PSI

• By right, a 23 will take total 8 runs, but we will do a L4 (only 4 runs)

• Responses (Three outputs collected under three different noise conditions)

o With Noise 1

o With Noise 2

o With Noise 3

• Objective:

o What are the Factors that help to reduce the Force required to open the package?

o Noise is irremovable, we can only try to Maximize the S/N ratio

68 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
a) Minitab Solution

(1) Creating the Worksheet

69 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 70 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 71 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
(2) Defining the Factors

72 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
(3) Analyzing the Responses

(a) Defining the Responses

73 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 (b) Graphs

74 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 (c) Analyses

75 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 (d) Terms

76 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 (e) Options

77 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 (f) Storage

78 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
b) Minitab Output

79 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
c) Predicting Taguchi Results

80 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 81 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 82 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 E. SUMMARY OF TAGUCHI’S EXAMPLE

With reference to Example: Minimizing the Response (Smaller is Better), we can repeat this example
for:

1. NOMINAL RESPONSE (NOMINAL IS BEST)

Temp Time Pressure Polymer

Level 1 – 1 Lo 125 80 2 0.5
Level 2 –2Mid 150 85 4 1
Level 3 – 3 Hi 200 90 6 1.5

o We want to nominalize the product specifications

o Choose a L9 (4 Factor 3 Levels) Taguchi Experiment

• Produce the table above (where Y1, Y2 and Y3 are the Responses with Noise).

83 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 2. LARGEST RESPONSE (LARGER IS BETTER)

Preheat Equalize Austenize Temper Quench

Level 1 – Low 1250 1350 1725 750 150
Level 2 – Hi 1450 1450 1775 950 200

o We want to increase the hardness of steel alloy.

o Choose a L8 (5 Factor 2 Levels) Taguchi Experiment

• Produce the following table above (where Y1, Y2 and Y3 are the Responses with Noise).

• The S/N ration should be the same as above.

84 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 VII. D OE FA Q S

A. WHAT IF MANY / ALL FACTORS BECOME SIGNIFICANT?

• There could be outliers or special causes that distort each run.

• Check if there are any unusual values.

• Outliers should be removed or replaced.

B. WHAT IF THERE'S MISSING DATA?

• Replace missing data with average.

C. MUST WE USE ALL FACTORS?

• Yes, try to. it is to your advantage.

• If you don't use them, the unused factors will be called "dummy" factors.

D. WHAT IF I CAN'T REPLICATE ALL RUNS?

• Randomly select runs to be replicated

E. MUST I RANDOMIZE ALL TRIALS?

• No, but it would be best.

• Sometimes it’s not possible to re-setup e.g., setting up a furnace temperature is only done
once.

• You can't randomize and must carry things out in certain order at one go.

• This may lead to confounding the temperature with another unknown factor. (Due to no
randomization).

85 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 VIII. S U C C E S S F U L S T E P S TO I M P L E M E N T D OE

1. Define objectives

2. Assemble a small knowledge team

3. Review all pertinent relevant data

4. Brainstorm to generate potential factors.

a. Be creative and do not accept existing theories without data.

5. Segregate (from the list) those factors that can be controlled vs uncontrollable.

6. Separate the expensive factors vs cheap factors.

a. Always include factors that are cheap, quick and easy to study.

7. For every factor, set levels boldly but NOT carelessly.

a. You need levels to be as wide as possible to force effects out of them.

b. But you must avoid dangerous or unfeasible conditions.

8. Design your study.

a. Narrow down the factors + review the total cost + complexity and control of the
experiment + need for replications.

9. Randomize but if you can't, be aware of drawing false conclusions due to unknown external
influence affecting the DOE.

10. Run the DOE + ensure correct levels + ensure materials are correct and keep good records.

11. Use Graphs to communicate findings.

a. Report conclusions in simple language for the audience, not statistical terminologies.

86 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 I X . OB S TA C L E S TO D OE

A. RESISTANCE INERTIA

• "We have always used OFAT and no time to learn new approach!"

• Answer: You will always get what you have always gotten if you keep doing what you keep
doing!

B. EXPENSIVE COST

• Nothing is free.

• DOE is an investment with a payback.

• Normally, the main cost is not Material Cost nor Processing Cost.

• Its the cost of using People's time.

• Use it carefully.

C. LACK OF MANAGEMENT SUPPORT

• Management must be involved.

• They must be educated and understanding what the objectives are, what the costs will be,
what is expected, and believe the power of DOE.

D. LACK OF TRAINING

• You need educated people who knows DOE techniques to carry it out.

E. WRONG THINKING THAT DOE IS ONLY USED FOR MANUFACTURING

• DOE can be used for any area that can define factors for study.

• Including strategy and sales.

87 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 APPENDIX

A. SELECTED FACTORIAL DESIGNS

1. 2^2 FULL FACTORIAL

88 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
2. 2^3 FULL FACTORIAL

89 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
3. 2^3 HALF FRACTIONAL FACTORIAL

90 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
F. SELECTED PLACKET BURMAN DESIGNS

91 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 92 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 G. SELECTED TAGUCHI DESIGNS

93 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 94 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 95 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 96 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 H. SELECTED MIXTURE DESIGNS

97 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 B I B L I OG R A P H Y

Durivage, M. A. (2016). Pracitcal Design of Experiments (DOE): A Guide for Optimizing Designs and
Processes. Quality Press.

98 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG
 A B O U T D R . A LV I N A N G

Dr. Alvin Ang earned his Ph.D., Masters and Bachelor degrees from NTU, Singapore. He is a
scientist, entrepreneur, as well as a personal/business advisor. More about him at www.AlvinAng.sg.

99 | P A G E

COPYRIGHTED BY DR ALVIN ANG

WWW.ALVINANG.SG