Quality Design and Control

Prof. Pradip Kumar Ray

Department of Industrial and Systems Engineering
Indian Institute of Technology, Kharagpur

Lecture – 17
Statistical Process Control-I (Contd.)

So, related to statistical process control we have several topics to discuss.

(Refer Slide Time: 00:21)

The statistical concepts knowledge in statistical concepts is a necessity and while we

discuss this topic that is the fundamentals of process control and a process design and
control. So, let me first talk about the statistical concepts. There are certain other
concepts you must have who first have an understanding before we go on discussing the
control charting and all other the tools and techniques in statistical process control.
(Refer Slide Time: 01:03)

The yesterday or in the last class we mentioned about the measures of dispersion and we
discussed already the first two measures that is the range and the variance of the standard
deviation. The third one that is the interquartile range is also is used in many cases. So,
the interquartile range is essentially we refer to the lower quartile that is a Q 1 and how
do you define a Q 1 is the value such that one-fourth of the observations fall below 8 and
three-fourth fall above it. So, that is the definition of Q 1. The middle quartile is the
median half the observations fall below it and half above it.

Similarly, we have the third quartile that is the value and such that the three-fourth of the
observations fall below it and one four above it I think it is clear to you. So, the
interquartile range is given by the IQR is Q 3 minus Q 1. So, this is the traditional
practice. To find the IQR what you have to do the data are to be ranked in ascending
order that is the first step. The first quartile is located at the rank of 0.25 into n plus 1
where n is the number of data points in the sample say 100. So, 100 is suppose the
sample size. So, the third quartile we located at rank 0.75 into n plus 1.
(Refer Slide Time: 02:57)

So, what is the first quartile? That means, the 0.25 n plus 1 and what is the third quartile
that is 0.75 n plus 1. So, here in as example, you have trying to observations these are the
values like 2.2, 2.5, 1.8 and so on. So, how do you determine the location of Q 1 and Q 3
that is straight we are applying the formula that is 0.25 into 21, 21 is n plus 1 twenty plus
1 that is 5.25 and similarly location of Q 3 is 0.75 into 21 that is 15.75. So, how do you
do? That means, you erase the data in the ascending order is it; that means, the first data
point is 1.7 and the last one that is 20th one that is 2.6.

So, at this location that is 1.825 you have the value; that means, the rank is 5.25; that
means, this is the rank there is a fifth rank between fifth rank and the sixth rank. So, what
is the quartile one that is 1.825 it is clear. And similarly as this is the location of Q 3 is
the rank is 15.75 that means, almost you know it is near to 16. So, the corresponding
value you can compute that is 2.275. So, this simple rule we follow.
(Refer Slide Time: 04:26)

Obviously, you know now you can determine the interquartile range right it is a measure
of dispersion. So, how do you do that? That means, Q 1 1.8 to 5 we have already
measured and Q 3 is 2.275. So, interquartile range is the difference that is 0.45
millimeter. So, this is the relative frequency and this is versus the data value. So, 25
percent is of the points the data points are less than say Q 1 and 75 percent of the data
points are less than Q 3, is it ok. So, this is what are you (Refer Time: 05:05) range.

(Refer Slide Time: 05:06)

Now, we are referring to the probability distributions. You know there are you are
supposed to use various kinds of tools and techniques and these tools and techniques
many such tools and techniques being used in quality design and control and you know
based on certain distributional assumption right. We have already mentioned that you
need to deal with the random variable either discrete type or continuous type in quality
control and improvement purposes.

So, what do you try to do? That means, against a particular say random variable which
you are dealing with in a given case you must be higher of what kind of you know the
probability distributions it may have. So, now, this just I will discuss briefly some of the
important probability distributions and their characteristics which are commonly used in
a quality control and design. So, this probability distributions are grouped under two
categories the first one is a discrete distribution; that means, the corresponding you know
that say random variable is of discrete type. So, these distributions deal with those
random variables that can take on a finite or countably infinite number of variables
essentially they can assume the integer values. Several discrete distributions have
applications in quality control. So, I will just name them, like the first one is the
hypergeometric distribution ok.

(Refer Slide Time: 06:46)

So, the hypergeometric distribution you know there is lot of the applications and
particularly you know by an application related to say acceptance sampling when you
deal with isolated single lot. So, the distributional assumption is hypergeometric.

So, hypermetric geometric this is the probability mass function, this is the probability
mass function and you have so, this is basically you know sigma or the px is given by D
c x N minus D, n minus X and this is the equations you use an X could be; that means,
the number of is number like say when you deal with say the number of nonconforming
units in a sample. So, it may assume several any one of these integer values and in n and
D; that means, this is n not m this is n n into n comma D. So, maximum could be n and
the in the X values and n is within or either it could be n or it could be D between D and
n. So, this is the formula use and, the number of nonconforming items in the populations
the size of the population capital n is the size of the population small n is the size of the
sample combination of D items taken at taken at a time and number of nonconforming
items in the sample. So, this is the hyper geometric distribution.

(Refer Slide Time: 08:38)

So obviously, what I am asking you to do like say for any say the distribution you come
across there are there are parameters. So, what will be the expression with these
parameters, particularly you know mean and standard deviation. So, that always you
need to compute. Like say given you know the probability must functions for the
hypergeometric distribution. So, what will be the expression for, what will be the
expression for you know the mean and the standard deviation? So, that always you can
compute.

Similarly we come across binomial distribution and binomial distribution this is n c x or

n c r, p to the power r into 1 minus p to the power n power r. So, the value of r could be 0
1 2 up to n, n is a sample size then in many cases we assume that that the random
variable is Poisson distributed. So, this is say the probability mass function of our
Poisson distribution lambda to the power e to the power minus lambda, lambda to power
X into factorial X this should be lambda to the power X by factorial X is it ok, where
lambda is equals to the mean or average number of events that happen over the specified
product volume or the time period.

(Refer Slide Time: 10:14)

Now, these are the 3 distributions we come across frequently one is the hyper metric
hypergeometric distribution, the second one is the binomial distribution and the third one
is the Poisson distribution. In many cases what we do I make I make on the certain
conditions, you may go for Poisson approximation to binomial distribution. So, when we
refer to these cases will have these assumptions or these approximations. So, later on will
take off numerical problems and where in many cases you go for Poisson approximation
for binomial distributions.

As far as continuous distributions are concerned we have many, but here at this point in
time I will be referring to the normal distributions. The most widely used distributions in
the theory of statistical quality control is the normal distribution. So, this is the
probability density functions you are all aware of. So, so this is one and there these there
are two parameters we have, one is the sigma and the second one is the mu. So, these are
the mu is the mean and sigma is the standard deviation.

So, this is the typical you know of the distribution f X versus X and this is a symmetric
distribution and with mean mu and the standard deviation sigma, is it ok.

(Refer Slide Time: 11:46)

Now, what we try to do? That means, we standardize the normal distributions and we
refer to the standard normal distributions where X is transformed with Z and Z is
essentially you know is nothing, but the difference of value of a particular say, particular
value of the random variable from the mean and these difference is expressed in the
standard deviation unit.

So, the difference is X minus mu. So, mu is one of the parameters. So, X minus mu these
difference expressed in the standard deviation is to this way you know the Z has been
defined and so obviously, the probability distribution function, probability density
function it is not distribution function density function is phi z is 1 minus e to the power
minus z by 2, is it ok, where Z may live anywhere between minus infinity to plus
infinity. So, here; obviously, with this transformation, this normal distribution we will
have a mean of 0 and variance of 1, is it ok. So, normally this is specified as Z you know
N 0 1 and we say that Z has a standard normal distribution, is it ok. So, X is transformed
into Z.

(Refer Slide Time: 13:23)

This is an example just as the simple examples I have taken the length of a machine part
is known to have a normal distribution with mean of 100 millimeter and the standard
deviation of 2 mm is it. So, what proportion of the parts will be above 103.3 mm?

So, what do you need to do? Let X denote the length of the part. So, mu 100 sigma is
equals to 2. So, the standardized value of 103.3 corresponding to say X 1 103.3 minus
100 divided by 2 because sigma is 2 and so it is 1.65. So, what you refer to that means,
area under the curve beyond 1.65. So, this is the probability the Z is greater than 1.65.
So, here it is 1.65. So, if you refer to the standard normal table. So, you will find the area
under the curve beyond Z equals to 1.65 is 0.0495; that means, 4.95 percent is the areas.
So, this way we calculate we refer to the standard normal distribution.
(Refer Slide Time: 14:37)

So, what proportion of the output will be between 98.5 and 102, what you try to do
against 98.5? You calculate corresponding Z 1 is it when the mean and the standard
deviation is known. So, this is 1.00 and similarly against 98.5 what is Z 2? That means,
98.5 minus 100 divided by 2; that means, minus 0.75.

So, now what you try to do? That means, area you know area under the curve between
say Z equals to 1.0 and Z equals to minus 0.75. So, minus 0.75 is somewhere here and
1.0 Z equals to 1.0 is here, is it ok. So, the area is this one is it that you have to compute.
So, the required you were referred to the standard normal distribution the corresponding
table and the required probability equals this 0.6147, is it ok. So, this value you get by
referring to the standard normal distribution. Similarly this, this area also you get and
when you subtract these 2 areas from 1. So, you get the value of 0.6147, it is clear. So,
you know how to use this term like standard normal distribution table.
(Refer Slide Time: 16:05)

What proportion of the parts will be shorter than 96.5? So that means, against 96.5 again
you calculate its Z value that is minus 1.75 with reference to 100 that is the mean and the
standard deviation is 2. So, this area again you refer to the standard normal distribution
table. So, against a value of Z of minus 1.75, area under the curve beyond this value of 2
minus infinity that is 0.0401. So, 0401 that means, 4.01 percentage of the points they will
be shorter than 96.5 all the parts ok. So, these are the simple examples.

(Refer Slide Time: 16:55)

Similarly it is important not many of the parts exceeds the desired length; obviously. If a
manager stipulates that no more than 5 percent of the part should be oversized what
specification limit should be recommended. So, here the problem is that X must should
be greater than equals to the probability is 0.05. So, this is 0.95 less than equals to A. So,
corresponding is a Z value that is 1.645. So, that you get and you need to determine X 1.
So, this is known, this is known, this is also given and. So, what you need to do? The
what will be the corresponding value of X 1.

So, therefore, this way we say that this way you determine the value of X 1 and we say
that a should be set at 103.29 to achieve the desired stipulation is it. So, this is another
example.

(Refer Slide Time: 17:50)

Now, under continuous distributions say exponential distribution also is used and we will
distribution particularly when you try to model the reliability function of a component;
and what is reliability? Reliability as you know this is one of the dimensions of the
quality.

So, this is the probability of the mass function probability, probability density function
for the Weibull distribution and this is exponential distribution is a particular case of
Weibull distribution. So, Weibull distribution is having 3 parameters that is you know the
gamma, gamma is referred to as the location parameter, beta is the shape parameter and
alpha is scale parameters. So, these we come across.
(Refer Slide Time: 18:41)

Now, we also must have an idea about the sampling because you know when you go for
statistical process control through control charting. So, why do you construct the control
chart what you need to do; that means, from the process of the from the population you
have to create a sample; obviously, while you create a sample; that means, you must have
a thorough idea about the sampling procedures.

So, just I will be referring to the concepts what is the basic concepts in sampling. So, a
sampling design is the description of the procedure by which the observations in a
sample are to be chosen it is very simple. So, what is a simple random sample? One of
the most widely used sampling designs in quality control is the simple random sample.
So, random number tables you refer to random number tables compute or computer
generated random numbers can be used to draw a simple random sample. What is a
stratified random sample sometimes the population from which the samples are selected
is heterogeneous; that means, suppose there is in the population of for the persons. So,
there could be you know the male population there could be female population is just one
example.

And you have to, you know you have to collect the sample from the population. So, and
it must be representative. So, what you try to do? That means, you collect one sample
from say the male population and another sample from the female population. So, this is
just a simple example. So, this is referred to as a stratified random sample. So, for
instance consider the output from two operators who are known to differ greatly in their
performance same example.

(Refer Slide Time: 20:39)

So, you have to collect sample; that means, from one operator and similarly for the
second operator. So, I will elaborate it later on the stratified sampling method is widely
used in the field of you know the quality control. So, rather than randomly selecting a
sample from their combined output, a random sample is selected from the output for the
each operator this way both operators are fairly represented.

So, we can determine whether there are significant differences between them, is it ok. So,
whatever the quality characteristics you are referring to. So, stratified random sample is
obtained by separating the elements of the population into 2 into non overlapping distinct
groups called strata is it ok, like one group is female another group is the male is it. So,
the data sets for one operator and the data sets for the second operator. So, these are
called different there are as the strata and then selecting a sample random sample from
simple random sample from each stratum.
(Refer Slide Time: 21:56)

So, similarly we have a cluster sample the cluster is formed and many a time no you
come across such situations the clusters are formed is it ok. So, what do you, many a
time what you do the entire population is grouped into several clusters based on
curtained the characteristics. So, if a plant has plants throughout the southern eastern
United States you may not be feasible to sample from each plant is it ok. Since like say,
cluster are then defined say one for each plant and some of the clusters are then
randomly chosen say 3 of the 5 clusters, is it ok.

(Refer Slide Time: 22:38)

So, this is the concept of sampling. Now, before I refer to certain fundamentals you know
of the statistical process control or the techniques. So, I am again referring to the
prevention based quality control this diagram as we took as I have already told you that
said that the first stage in the evolution of the quality system is inspection and the next
one is the quality control.

So, whatever you know the statistical process control what we are talking about we
assume that we are in a stage in a condition where prevention based quality control you
know is a necessity and this control chart which you are going to use for statistical
process control; obviously, you know the system should be the prevention based quality
control. So, you know what is it; that means, you have several kinds of resources, you
have the process and then as a worker you had as an operator the process what you need
to do; that means, you need to go for measuring the quality characteristics and if
adjustment you need to have in the measurements you must be able to do the adjustments
is it. And ultimately the output should be you know the (Refer Time: 24:11) to you and
so that you get a satisfied customer. So, this is in this prevention mode, prevention based
quality control the control chart is to be used.

(Refer Slide Time: 24:19)

Now, I will just briefly tell you as that why control charts what is a control chart. So,
what is a control chart? Control chart is a graphical tool for monitoring the activity of an
ongoing process. So, it is an ongoing process; that means, through control by using
control chart what you can do require definitely you can go for online real time control.

Control charts are sometimes referred to as the Shewhart control chart because Walter A
Shewhart, he actually the first you know way back in 1920s he introduced the concept of
control charting and the subsequently his concept of control charting was implemented in
several well known organizations throughout the world.

Now, how do you construct a control chart? It is very simple, you have x axis, you have
the y axis. Now, supposing for a given quality characteristic you need to you know the
construct the control chart. So, what you try to do you need to collect the values of the
quality characteristics from the process the values of the quality characteristics are
plotted along the vertical axis; that means, the y axis and the horizontal axis represents
the samples or the subgroups as I have already pointed out that the samples which you
draw from the population or from the process there are certain rules there are certain
norms later on will discuss this one. So, and these the samples you have to draw as the
process is on; that means, the time order is to be mentioned it means that if you say it is
the first sample; that means, the first sample is drawn first if you say is the second
sample; that means, the second sample is drawn next like this. So, trying order is to be
mentioned.

So, this you know the x axis represents the samples from which the quality characteristic
is found; that means, instead of measuring or all the units in the population you just rise
sample and within the sample you have you know a number of values and those values
are to be collected and then you get its the average value. So, say the average represents
a particular sample quality characteristic. So, these average is calculated based on the
number of observations in the sample, that means the sample size and these
characteristics are then plotted in order of time in which the samples are taken this is the
general way of constructing the control chart. So, it is very very simple. So, examples of
quality characteristics like average length, average diameter, average tensile strength,
average resistance and so on and so forth.
(Refer Slide Time: 27:23)

So, this is the typical control chart. So, this is actually the y axis, this is the x axis and in
a typical control chart you have a central line and you have the upper control limit, you
have the lower control limit is it and usually they are at equal distance from say the
center line. So, on one side you have the upper control limit on the other side there is the
same distance you place the lower control limits.

Area between the upper this point is to be noted. The area between the upper and the
lower control limits these reflects actually the natural variation of the process whereas,
the area beyond UCL; that means, this area or the area beyond LCL that means, this area
this you know represents actually the variation due to assignable causes. That means, if a
certain sample point falls over here you assume that there is a variation definitely, but
this is within the natural variation. But if the point is somewhere here; that means,
against the fourth sample; that means, you will find that there is a variation because your
target is a centerline, so this much variation. Now this must variation is due to some
assignable causes and something has gone wrong at that point in time while you collect
data for the fourth sample.

So, this is referred to as assignable cause. And whenever you find that the point is the
plotted says the fourth point is plotted say outside say he and all the points are plotted
within the control zone. So, you may assume that the necessary condition for in control
state has been achieved.
(Refer Slide Time: 29:18)

So, by using a control chart you know whether the process is in control or the process
has gone out of control. So, if a point for the plots outside of the control zone we assume
and we assume that the process has gone out of control; that means, out of control state.
And if you find that the point plots within the control region; that means, the area
specified by the upper control limit and lower control limit we assume that the process is
in control or is in control state.

So, what are the benefits? When to take corrective action is it ok, as soon as you find that
the process has gone out of control now you have to take some corrective actions so that
the process is brought back to a state of control is it ok, from the out of control to in
control state. Type of remedial action necessary; that means, when you start using a
control chart against a process; that means, you are continuously using this control chart
you know the behavior of the process; that means, ins and outs of the process you have a
thorough knowledge and, so what kind of remedial action you need to take to bring the
process back to you know the in control state that you will come to know.

When to leave a process alone; that means, if you find that that the process is a
continuing in control state. So, in many cases you do not go for you know the controlling
the process you leave the process alone. So, if you can create such a condition you
assume that my purpose of using a control chart is fulfilled. You know will be will be
discussing the process capability and when you start collect, we start collecting data
through control charting you know later on these data you can use to the measure the
capability of a process, is it ok. So, if you start using a control chart, so based on the
information you collect, you can measure the capability of the process.

Possible means of quality improvement is it ok. So, always you go for it.

(Refer Slide Time: 31:47)

So, there are two types of the variations. So, and what we have mentioned that that the
two types of causes of variation and what we have been stressing that any exercise and
quality is essentially an exercise on controlling the variability. So, you must know that
the causes of variation that there are two types of causes of variation, one is the common
cause these also referred to as the chance cause even note down or the natural cause.
Whereas, the second kind of the causes they refer to the special causes or assignable
causes.

So, control of a process is achieved through the elimination of special causes. An

improvement of a process is accomplished the reduction of the common causes. That
means, common causes actually are a part of the system.
(Refer Slide Time: 32:44)

So, the variability caused by special or assignable causes is something that is not inherent
in the process. That means, for the time being it has it has happened, but if you take some
remedial action obviously, this cause will be will be eliminated and that that the process
has become free of this particular cause.

Deming believes that fifteen percent of the all problems are due to special causes.

(Refer Slide Time: 33:11)

Now, I will just you will take 2 to 3 minutes of time to 2 more minutes to explain the
common causes. The variability due to common or chance causes is something inherent
to a process; that means, you are living with this. So, any systems we are talking about;
that means, that the we try to reduce the variation, but there will be some variation. So, at
any point in time you must know that what is this allowable variation and these variation
is due to the common causes. So, this is the meaning.

And the process operating under a stable system on common causes is said to be in
statistical control we will be using this term this is referred to as the statistical control or
not. So, management alone is responsible for common cause that is a part of the systems
as the management has created the system and Deming believed that about 85 percent of
all problems are due to common causes and hence can be solved by action on the part of
management. This is through his study he concludes.

So, here I conclude, but you know later on will discuss on the common causes of
variation as well as the assignable causes of variation. So, if you can remove the
assignable causes the process becomes in control state and if you can consider the
common causes, conseque orders and you stripe you can eliminate some of the common
causes we say that the process or the quality control system has improved.