Quality Design and Control

Prof. Pradip Kumar Ray

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

Lecture – 38
Acceptance Sampling-II (Contd.)

Keywords: Variable Sampling Plans for Controlling Lot Proportion Nonconforming

I will be discussing the type two variable sampling plans.

(Refer Slide Time: 00:26)

And the first type already we have discussed, now we are going to discuss the variable
sampling plans used for controlling lot proportion nonconforming and, I will be dealing
with several number of numerical exercises.
(Refer Slide Time: 00:45)

Not only the process average, you also need to control the proportion nonconforming
and, it is not that it is related to the attributes data, many a time know whenever we say
the proportion nonconforming we say, we might may assume that it is related to the
attributes data. So, you may also in the defined proportion nonconforming with respect to
a quality characteristic for which the variables data are collected, the exact deviation
from the standard, from the nominal value, if any, you can determine and accordingly
you can determine this proportion nonconforming related to a particular lot.

So, given the quality characteristic for which the sample values are obtained, the
proportion nonconforming can be estimated.
(Refer Slide Time: 02:15)

Now, so, let us talk about the variable sampling plan or controlling lot proportion
nonconforming when the quality characteristic is of single specification limit type. So,
for a single specification limit type quality characteristic, there are many such quality
characteristics you come across. Now, it is assumed that that this quality characteristic is
normally distributed with standard deviation known, like in previous cases, we have
assumed.

The proportion nonconforming can be computed with a procedure as follows. So, when
the normality assumption holds s, proportion nonconforming you can calculate. Suppose
you say lower specification limit type quality characteristics. The lower specification
limit is specified as L. So, you compute Z L. that is actually the standard normal deviate.

So, .

is a sample average and L is the lower specification limit, σ is the process standard
deviation which is known.
(Refer Slide Time: 04:01)

When upper specification limit type quality characteristic is considered, so, how do you

compute ZU against the standard normal deviate that is U is the upper

specification. The area under the portion of the standard normal curve that falls outside
ZL or ZU is the estimate of proportion nonconforming.

You are indexing your plan with respect to the proportion nonconforming. How do you
define the good quality? When the proportion nonconforming is very less. How do you
define a bad quality? when the proportion nonconforming is very high. Now it is

observed that as the distance between L or U and increases, the value of ZU will be
more and if the value of ZU or ZL is more, the corresponding area under the portion of the
standard normal curve that falls outside ZL or ZU will be less. So, that is why the
proportion nonconforming decreases,

That if the distance between L and or between U and increases, the proportion
nonconforming decreases.
(Refer Slide Time: 06:16)

Now, with this basic understanding, we refer to the methods for acceptance sampling
when you come across or when you are dealing with the variables data and your
objective is to the control proportion nonconforming and, we will be referring to a
particular quality characteristic.

Now, in this context when your objective is to control the proportion nonconforming,
there may be two methods as suggested by the researchers and the practitioners in this
field and these two methods are known as Form-1 method or it is also referred to as k-
method and the other method is Form-2 method or it is also referred to as the M-method.

It is to be noted that these are not different from each other, it is not with the Form-1, you
get one sorts of result and from the Form-2, you get another sort of results and then you
have to make a choice, it is not like this. First you apply the Form-1 method and you get
a decision; now at the second stage you can modify the Form-1 method with the Form-2
method and you apply the Form-2 method and again you take a decision.

If you follow the follow the rules or follow the methods properly, you will find that
whatever decisions you have taken under Form-1, same decision you will be taking
under Form-2 also. Form-2 method you can use for verification purpose.

In a given situation both the methods may be used, but decision regarding lot acceptance
or lot rejection should be the same.
(Refer Slide Time: 09:23)

How do you use the Form-1 method? First Z L or ZU is computed with respect to a sample
drawn from the lot. As soon as you compute Z L, it is a single specification unit case and
more specifically, it is the lower specification limit case whereas, if you compute Z U, it is
an upper specification limit case. So now, you compute ZL and ZU- the formulas are
known to you.

Next, compare ZL or ZU with a critical value k. The critical value is known, while you
have designed a sampling plan, ultimately you need to know the what is the sample size
and what is the value of k. So, these are the two parameters values in Form-1 method.

First is with the , with the L, with the σ, you calculate ZL or ZU. So, while you
compute ZU, you must know what the value of U is. So, either ZL or ZU you compute, and
you check whether ZL or ZU is greater than or equals to k.

So, if this condition holds, then the lot is accepted, otherwise, the lot is rejected.
(Refer Slide Time: 11:19)

Under Form-2 method, we proceed to the next step to compute the normal variate value
against L or U in an unbiased manner and when we compute Z U, it may be a biased
estimate, this concept of bias can be explained from different prospective.

So, if you want to avoid this bias and you have to recommend an unbiased estimate for
ZL or ZU against L or U, you need to compute QL or QU, as the case may be, and this Q
you compute by modifying the value of ZL,

Now, these estimates are considered unbiased with minimum variance, n is the sample
size.
(Refer Slide Time: 13:18)

Now, an estimate of proportion nonconforming, is obtained by finding the area on the

normal curve that falls outside QL or QU. So, again you refer to the standard normal table
because the normality assumption holds. So, refer to the standard normal table and you
just get the area under the normal curve that is falling outside QL or QU and this area is

referred to as the proportion nonconforming and the notation is .

Now, compare with a maximum allowable proportion nonconforming, denoted by M.

So, now, we have defined what is M. M is maximum allowable proportion
nonconforming.

So, if you find that this proportion nonconforming that is the actual proportion
nonconforming is greater than the maximum allowable proportion nonconforming so,
this is not a desirable situation; obviously, the lot is rejected otherwise, the lot is
accepted. So, the decision rule is known to you. You need to know the value of n by
using its derived expressions like you do for the k.
(Refer Slide Time: 15:10)

So, now, you have understood what the Form-1 method is as well as the Form-2 method.
So, Form-1 method is also referred to as the k method and the Form-2 method is also
known as M method.

Now, you have to design the sampling plan, you must be able to the derive the
expressions for the sample size as well as the other parameter values of a given variable
sampling plan where the objective is to control the proportion nonconforming. So,
essentially it is an exercise for determination of n, k and M.

Proportion of nonconforming indicating good quality p 1. So, this you must be able to
specify, whether it is 1 % or 0.5 % or 0.05 %. So, that is considered to be in a good
quality in terms of proportion nonconforming for which the producer’s risk is α. So, if
the value of p1 is very less, what you will find that the producer’s risk α also will be very
less. So, may be 0.01 or may be 0.001.

Whether if it is 0.02, this value could be just 5 % or 0.04 or 0.05 and, if the proportion
nonconforming increases; obviously, it is indicative of bad quality. So, that value you
must be able to specify like 0.04 or 0.05 or 0.06; that means, 6 % is considered to be very
bad quality for which the consumer’s risk is β, even if the quality is extremely poor you
might say that there is a high probability of acceptance. So, it is avoidable, whether it is
avoidable or not, that is the different issue, but right now this is the case you cannot
avoid it. So, it may be as high as 10 %.
Assuming normal distribution for x and the lower specification limit L, process mean µ 1
for proportion nonconforming p1 and process mean µ2 for proportion nonconforming p2.
Suppose, these are known, then you refer to the following figure.

(Refer Slide Time: 18:25)

So, you are getting the distribution of x, this is the distribution of x. When the
distribution is here with standard deviation of σ.

Suppose, this is a lower specification limit case. So, the value; means, this area is beyond
L; this is the proportion nonconforming. So, against the proportion nonconforming p 1,
what is the corresponding value of mean that is µ1. Now, there is a shifting in the process.
So, now, µ1 has become µ2 or in other words p1 has become p2. So, if p1 becomes p2, from
the good quality to bad quality, what is the corresponding value of µ 2. So, that also you
can notice, this entire area is considered to be the proportion nonconforming, p a may be
0.01. Now it has become 0.08 or 0.10.
(Refer Slide Time: 19:45)

So, now corresponding to µ1, you calculate z1- the standard normal deviate and
corresponding to µ2, you calculate the standard normal deviate z2.

(Refer Slide Time: 20:34)

Now, you consider the distribution of as shown in the following figure. You have

these parameter that is k. Now, against µ1, you have the distribution of .
If for the quality is extremely good, the probability of non-acceptance is very high. So,
this is basically called the producer’s risk, whether when the quality deteriorates to µ 2,
with respect to this k, now the area under the curve which is within k, this value is
greater than k.

So, obviously, there is a possibility of its acceptance even if the quality deteriorates. So,

we are considering the distribution of .

(Refer Slide Time: 22:02)

Under Form-1, the lot is accepted if

Now, we have two expressions, one particular value of µ is µ L, L is fixed, that is the lot
specification unit case and σ is the processed under deviation. So, the probability we
have computed that is z1, the standard normal deviate. So, ultimately this becomes
(Refer Slide Time: 23:50)

So, you have the two equations and you have two unknowns and these two independent
simultaneous equations you have. Now, solving these two equations simultaneously, we
get
You follow the decision rule when k is known. So, z α, zβ, z1, z2, you compute, you get the
expressions for n. So, sample size is known, either you use these expressions. Now, these
expressions when you use, you try to meet the α conditions, but you do not bothering
about meeting the β conditions whereas, if you use these alternative expression for
computing k, for determining the value of k; obviously, you are trying to meet the β
conditions, but you are not bothering about the α conditions.

(Refer Slide Time: 25:36)

So, under Form-2, you need to estimate the value of M. We know the expressions for k.
Now, if you opt for Form-2; obviously, you must have an estimate for the value of n. An
estimate of n is

Once n is known, follow the decision rule. For a given problem both the methods may be
used, Form-2 may be used for verification purpose.

However, the decision regarding lot acceptance or lot rejection must be same in both the
methods.
(Refer Slide Time: 26:30)

Now, here is an example. In the manufacture of heavy duty utility bags for household
use. the lower specification limit for the carrying load is 100 kg. So, it is a lower
specification limit case. We want to accept lots that are 2 % nonconforming with a
probability of 0.92. For lots that are 12 % nonconforming, it is a very bad quality, the
desired probability of acceptance is still 0.10. So, this is basically the consumer’s risk.
The standard deviation of the carrying load is known to be 8 kg, this is the standard
deviation known case. Find a variable sampling plan and explain its operation.

(Refer Slide Time: 27:22)

So, here we have L = 100, p 1 = 0.02, α = 0.08, p2 = 0.12 is a bad quality, 12 %
nonconforming, still the value of β is very high, that is the consumer’s risk is 10 % and σ
is 8.

So, from the standard normal distribution table, we get z 1 against p1 that is 2.055, z2
against p2 that is 0.175, zα against α = 0.08, you get a value of 1.405 and z β against the
value of β, the tail area, that is 1.282, you are referring to the standard normal table. You
use the formula and you get a value of n that is 9.323, and you just make it the next high
integer that is 10 and if you want to satisfy the value of α = 0.08; there are two the
formulas for k. So, if you use the first formula, if you want to meet the condition of α
exactly, then the value of k is this- 1.611.

(Refer Slide Time: 28:48)

You can refer to the other formula for k, if you want to meet the consumer’s risk
condition exactly, you can calculate the values of k using both the equations, there will
be two values and you may also take the average. So, that is the third option you can also
adopt.

So, we select a sample of 10 bags and find the sample average of the carrying load to be
110 kg, we can use either Form-1 or Form-2; using Form-1, you can calculate Z L as this-
1.25. Since ZL ≤ k, the decision based on Form-1is to reject the lot.
(Refer Slide Time: 29:34)

And similarly, you calculate QL, you are moving to Form-2. So, you calculate 1.32 and
the corresponding the proportion that falls outside the standard normal value that is
0.0934 and then you calculate M that is

And against this value of 1.70, you get an area of 0.0446, 4.46 is the maximum allowable
proportion nonconforming; you are getting a value of say 9.34 percent.

So, since > M the decision is to reject the lot.

(Refer Slide Time: 30:35)

So, for selection of variable sampling plans, standardize sampling plans you may use like
you do for attribute sampling plans and there are standards, MIL-STD-414, an equivalent
standard is ANSI/ASQC Z1.9 or equivalent standard is ISO 3951.

(Refer Slide Time: 31:00)