Workshop

on

Design Of Experiments
&

Statistical Quality Control

(Basic Concepts and Project Discussion)

On 20th January 2018

At

Mannur
Total Quality Management (TQM)

Design
Of
Experiments

DOE – Basic Concepts

Total Quality Management (TQM)

Basic Concepts

1.1 Factor and Response

The quality of a component as measured by its dimension , geometrical tolerances surface finish ,
etc. at fine boring or milling or any other machine may depend upon the stock to be removed ,
number of cuts, feed rate, rpm., coolant quality , viscosity , flow rate , and a host of relevant factors.
We call dimension , ovality , taper , surface finish as the dependent variables or 'Responses '.
Stock , number of cuts , feed rate , rpm. , coolant quality, etc. which can be varied at will resulting in
variation in response are called independent variables or ' Factors '.
There could be a large number of factors affecting one or more responses.

1.2 Levels and Types of Factors

In an experiment to improve the surface finish of a component one may choose three materials.
These three materials are considered as three levels of the factor "Material". This factor is
qualitative.
One may try four feed rates which can be measured. They will be referred to as the four levels of the
factor "Feed Rate". The factor is termed as quantitative.

1.3 Treatment / Treatment combination / Factor combination

A combination of various factor levels is called treatment/ treatment combination/ factor
combination
For example, one type of material, specific feed rate, specific rpm. is called a treatment / treatment
combination / factor combination.

1.4 Cause-Effect Diagram (CED)

Cause-effect diagram or fish bone diagram is a methodical way of writing down various causes
related to machine, material, method, measurement, man and environment which are likely to
influence the response or cause variation in the response. CED will enable to view the problem in its
totality. It provides a list of several factors or several improvement theories for verification through a
statistically designed experiment.

MAN MILEU /ENVIRONMENT MACHINE

New Time Tool
Operator Life
Location Operating
Poor Job Condition
Knowledge Temperature s PROBLEM

Feed Calibration EFFECT

Rate Alloys
Lubricant
Supplier I
Process
Inspectors Suppliers
Parameters
Supplier II
METHOD MEASUREMENT MATERIAL

DOE – Basic Concepts

Total Quality Management (TQM)

1.5 Design of Experiment

The analysis of any data is dictated by the manner in which the data are collected. Design of
experiment is then a plan for collection of data on response(s) when the chosen factors are varied in
a prescribed way.
The data arising out of design of experiment can be a basis for multivariate analysis.

1.6 Use of Statistically Designed Experiments

There are many process variables, variation in which will affect product quality .
There are many product and process quality characteristics, variation in which can contribute to the
variation in the assembly parameters.
Statistically designed experiments are useful in
 Identifying variables which are really important / dominant
 Providing quantitative estimates on how various factors affect product quality
 Providing information on interactions
 Providing a direction for varying various factors for further improvement
 Choosing optimum combination of various factors

1.7 Advantage of Statistically Designed Experiment

The advantage of using a statistically designed experiment is that the data arising out of such
experiments can be subjected to analysis of variance.

1.8 Experimental Error

The response can vary from one product to another although manufactured using the same factor
combination. This variation can arise due to slight variation in the levels of factors under
investigation or due to variation in the background conditions (factors not considered for the
experimentation). This variation among response values for a given set of factor combination is
called experimental error.

1.9 Tests of Hypotheses

In our effort for continuous improvement, it is common to seek improvement in yield, reducing
rejections, reduction of variance, etc.

The use of hypothesis testing to verify experimental results contain enough information to cast
doubt on conventional wisdom.

1.10 Analysis of Variance

Analysis of variance is a technique by which the total variation observed in the response can be
subdivided into its meaningful components
 Variation due to factors under study
 Variation due to interactions
 Variation due to experimental error
Anova provides tests of significance, magnitude of various effects and their importance.

DOE – Basic Concepts

Total Quality Management (TQM)

1.11 Principles of Experimental Design

Replication
Randomization and
Local Control (balancing, blocking & grouping of experimental units)
are principles of Experimental Design.

REPLICATION : conducting repeated trials for the factor combination constitutes replication .
Multiple readings do not necessarily constitute replication . It provides an estimate of experimental
error , this will be used for testing significant effects. It also enables us to obtain a more precise
estimate of the average effect of any factors.

RANDOMIZATION : is a method to eliminate the bias on the part of the experimenter.

LOCAL CONTROL : refers to balancing , blocking , and grouping .
Balancing consists in having the same number of experimental units for comparing various effects .
It makes design more efficient.
Blocking consists in the allocation of experimental units to blocks in such a way that the units within
a block are relatively homogeneous while the blocks would differ from themselves.
Grouping is placing of a set of homogeneous set of experimental units into groups of equal size for
ultimate allocation to various treatments under consideration.
Sometimes, we keep the current operating conditions and their results as a reference point for
comparison when we perturb various factors and study the response

1.12 One Factor Approach

While a number of factors are said to have an effect on response, the conventional approach has
been to vary one factor, keeping all the other factors constant. This approach is very inefficient. It
consumes more time and, interactions if present , cannot be examined . The effect of one factor can
be cited when all others are held constant . It requires a large number of trials (compared to a
factorial experiment).

1.13 Full Factorial Experiment

When each of several factors are varied a factorial experiment is the totality of all combinations.
let A1 , A2 , be the two levels of factor A and
let B1 , B2 , be the two levels of factor B .
Then a full factorial experiment will consist in conducting the following trials / factor combinations.
A1 B1 , A1 B2 , A2 B1 , A2 B2 .
This is called a 22 Factorial Experiment
23 factorial experiment consists in having 3 factors each at two levels in eight trials.
Similarly 23 , 3n, 4n , 2m x 3n , etc. can be defined .

1.14 Advantages of a full factorial experiment

As against one factor approach, inductive value is more, as a factorial experiment gives information
on main effects and interaction effects.
The various levels of one factor will act as replications for other factors and the factorial effects can
be more precisely estimated.
A factorial experiment gives information about interactions , if present .
In the absence of interactions factorial effects can be estimated precisely .

DOE – Basic Concepts

Total Quality Management (TQM)

The effect of one factor is seen when all other factors are varying .

DOE – Basic Concepts

Total Quality Management (TQM)

1.15 Interaction
When the effect of one factor is different at different levels of the other factors , we say that the two
are interacting . Consider the following case when yield is dependent on temperature and time .

Temperature Time
4 hours 2 hours
1150 80% 70%
1000 70% 60%

The effect of time is to increase the yield by 10 % when time is changed from 2 hrs to 4 hrs
irrespective of temperature . The two factors are said to be not interacting.

DOE – Basic Concepts

Total Quality Management (TQM)

Statistical
Quality
Control

SQC – Basic Concepts

Total Quality Management (TQM)

Basic Concepts

2.1 Data : Data are of 2 types

DATA

MEASUREMENT DATA ATTRIBUTE DATA

Diameter Errors in documents or

Bearing noise drawings,
Amps Defects in a product or
rpm assembly,
Length Defectives in a sample
Width Omissions,
Feed rate Absentees,
Temperature Spares consumed in a period
Moisture Breakdowns
Cycle time
Document processing time
Product life
Pressure

Geometrical Dimensions
Clearnce
Taper
Ovality
air gap
Characteristic that can be
Characteristic that can be counted.
measured on a continuous
scale.

2.2 Chance Causes & Assignable Causes

Chance Causes : Slight variation in quality characteristic or process parameter is inevitable.
Production can slightly vary from shift to shift depending on slight variation in process parameters
Variation due to chance causes cannot be economically eliminated.

Assignable causes :
• A shift in the setting level of a process parameter
• A bad condition of the equipment
• Use of a machine or tool beyond its useful life
• Trend in production / quality
are some examples.

Variation due to assignable causes needs to be reduced by eliminating the assignable causes.

SQC – Basic Concepts

Total Quality Management (TQM)

2.3 Average and Standard Deviation :

There is a tendency for the observations to cluster around a central value. This corresponds to the
level / average at which the quality characteristic is set.
Variation around this level is due to chance causes. Standard deviation is a measure for this
variation.
Let X1 , X2 , X3 , .............. , Xn be a set of n observations.
X1  X 2  - - - - - -  X n
Sample Average  X 
n

1
Sample Standard Deviation  s  ( X  X ) 2
n -1
Let us consider data on Bore Diameter for 100 components
Bore Diameter
11.995 11.999 11.998 11.998 Summary Statistics
11.999 11.996 11.996 11.993
11.99 11.99 12 11.996 LSL 11.992
12 11.994 11.994 11.992 USL 12
11.999 11.995 11.991 11.996 Average 11.99505
11.995 11.992 11.996 11.992 Std Dev 0.002893
11.995 12 11.999 11.994
Min 11.989
11.994 11.994 11.996 11.997
Max 12.001
11.994 11.995 11.998 11.997
11.992 11.995 11.993 11.993
11.999 11.994 11.994 11.995
11.998 11.996 11.995 11.994
11.998 11.993 11.99 11.991
11.997 11.997 12.001 11.995
11.997 11.991 11.995 11.999
11.993 11.991 12 11.995
11.991 11.996 11.99 11.995
11.995 11.996 11.995 11.999
11.997 11.993 11.993 11.994
11.993 11.998 11.998 11.991
11.989 11.996 11.991 11.998 Symbols :
11.993 11.995 11.992 11.996 Average Standard Deviation
11.99 11.997 11.996 11.992 Population  
12.001 11.997 12.001 11.995 Sample s
11.992 11.994 11.994 11.992
Sample average, X = 11.99505 is an estimate of the average bore diameter of the lot / population
of components from which this sample is drawn. We use the symbol  for the population parameter
( average Bore Diameter of the lot.)
Similarly, Sample standard deviation, s = 0.00289 is an estimate of population standard deviation, .

SQC – Basic Concepts

Total Quality Management (TQM)

2.4 Pattern of Variability : Histogram and Normal Distribution

Histogram
The data range has been divided into 14 class intervals and tally marks put and the clas
frequency and relative frequency given.

Class Relative
Class Interval Tally Marks Frequency Frequency
> 11.984 0 0
11.984 - 11.986 0 0
11.986 - 11.988 0 0
11.988 - 11.99 |||| | 6 0.06
11.99 - 11.992 |||| |||| |||| 15 0.15
11.992 - 11.994 |||| |||| |||| |||| | 21 0.21
11.994 - 11.996 |||| |||| |||| |||| |||| ||| 28 0.28
11.996 - 11.998 |||| |||| |||| | 16 0.16
11.998 - 12 |||| |||| | 11 0.11
12 - 12.002 ||| 3 0.03
12.002 - 12.004 0 0
12.004 - 12.006 0 0
12.006 - 12.008 0 0
12.008 - 12.01 0 0

13% below LSL.

3% above USL.
Total non-conformance 16%.

SQC – Basic Concepts

Total Quality Management (TQM)

Normal Distribution
It has been empirically verified that most of the quality characteristics under the influence
of chance causes will have a normal pattern of variability (bell shape) with certain mean ()
and standard deviation ()

Properties of normal distribution :

1. Normal distribution is completely designated by two parameters  and .

2. Normal curve is bell shaped.
3. Normal distribution is symmetric around .
4. Normal distribution extends from - to + , but for all practical purposes the region  + 3
and  - 3 covers almost the entire part.
5. The area under the normal curve is one.
6.  ± 1 will contain 68.26%,  ± 2 will contain 95.44% and  ± 3 will contain 99.73% of the
observations.

SQC – Basic Concepts

Total Quality Management (TQM)

HISTOGRAM Interpretation

LSL USL LSL USL

T T

LSL USL LSL USL LSL USL

T T T

LSL USL LSL USL LSL USL

T T T

LSL USL

SQC – Basic Concepts

Total Quality Management (TQM)

LSL USL

LSL USL

LSL USL

SQC – Basic Concepts

Total Quality Management (TQM)

HISTOGRAM Examples
100 readings of Bore Diameter from two shifts.

Both distributions are similar with respect to average and variance.

100 readings of Bore Diameter from 2 spindles

Both distributions have different averages and variances.

SQC – Basic Concepts

Total Quality Management (TQM)

Distribution of Averages
Depending upon what items of the population are included in the sample , the sample
average keeps on varying from one sample to another. The frequency distribution of
averages follows along with the frequency distribution of individuals.

Tally Marks Tally Marks

Class Interval Averages Population
> 11.984
11.984 - 11.986
11.986 - 11.988
11.988 - 11.99 |||| |
11.99 - 11.992 |||| |||| ||||
11.992 - 11.994 |||| || |||| |||| |||| |||| |
11.994 - 11.996 |||| |||| ||| |||| |||| |||| |||| |||| |||
11.996 - 11.998 |||| |||| |||| |||| |
11.998 - 12 | |||| |||| |
12 - 12.002 |||
12.002 - 12.004
12.004 - 12.006
12.006 - 12.008
12.008 - 12.01

In-fact the sample averages follow a normal distribution with the same  , but  given by /n,
where n is the sample size. (eg. Bore Dia)

 We must allow the averages to vary to some extent.

 The extent to which we allow the averages to vary is explained in tests of significance.

SQC – Basic Concepts

Total Quality Management (TQM)

2.5 Line Chart

Visual aid to see the conformance quality and trends and shifts if any.

2.6 Average chart / X chart

X chart is used to detect a shift in the level.

We use central line at X for statistical regularity – whether the system is under the influence of
chance causes. We use  for technical control.
Lower and upper control limits are given by µ ± A2 R , Where A2 is read off from table of control
charts constants for the sample size n.

SQC – Basic Concepts

Total Quality Management (TQM)

2.7 Range Chart

R-chart is used to detect increase or decrease in variability.

We use central line at R .

Lower and upper control limits are given by D3 R and D4 R respectively.

SQC – Basic Concepts