The oc curve_of_attribute_acceptance_plans

The OC Curve of Attribute Acceptance Plans
The Operating Characteristic (OC) curve describes the probability of accepting a lot as a function
of the lot’s quality. Figure 1 shows a typical OC Curve.

Operating Characteristic Curve

100.0%

80.0%
Probability of acceptance, Pa

60.0%

40.0%

20.0%

0.0%
0.0% 10.0% 20.0% 30.0% 40.0% 50.0%
Percent nonconform ing, p

Figure 1 Typical Operating Characteristic (OC) Curve

The Shape of the OC Curve
The first thing to notice about the OC curve in Figure 1 is the shape; the curve is not a straight
line. Notice the roughly “S” shape. As the lot percent nonconforming increases, the probability
of acceptance decreases, just as you would expect.

Historically, acceptance sampling is part of the process between a part’s producer and consumer.
To help determine the quality of a process (or lot) the producer or consumer can take a sample
instead of inspecting the full lot. Sampling reduces costs, because one needs to inspect or test
fewer items than looking at the whole lot.

Sampling is based on the idea that the lots come from a process that has a certain
nonconformance rate (but there is another view described below). The concept is that the
consumer will accept all the producer’s lots as long as the process percent nonconforming is
below a prescribed level. This produces the, so called, ideal OC curve shown in Figure 2.

When the process percent nonconforming is below the prescribed level, 4.0% in this example,
the probability of acceptance is 100%. For quality worse than this level, higher than 4%, the
probability of acceptance immediately drops to 0%. The dividing line between 100% and 0%
acceptance is called the Acceptable Quality Level (AQL).

Copyright © 2009 by Ombu Enterprises, LLC

The OC Curve of Attribute Acceptance Plans Page 1 of 7

The original idea for sampling takes a simple random sample, of n units, from the lot. If the
number of nonconforming items is below a prescribed number, called the acceptance number and
denoted c, we accept the lot. If the sample contains more nonconforming items, we reject the lot.


100.0%

80.0%

60.0%

40.0%

20.0%

0.0%
0.0% 2.0% 4.0% 6.0% 8.0% 10.0% 12.0% 14.0% 16.0% 18.0% 20.0%
Percent nonconforming, p

Figure 2 Ideal OC Curve

The only way to realize the ideal OC curve is 100% inspection. With sampling, we can come
close. In general, as the sample size increases, keeping the acceptance number proportional, the
OC curve approaches the ideal, as shown in Figure 3.

Similarly, as the acceptance number, c, gets larger for a given sample size, n, the OC curve
approaches the ideal. Figure 4 illustrates the relationship.

Some Specific Points on the OC Curve
Because sampling doesn’t allow the ideal OC curve, we need to consider certain risks. The first
risk is that the consumer will reject a lot that satisfies the established conditions, i.e., the process
quality is acceptable, but, by the luck of the draw, there are too many nonconforming items in the
sample. This is called the producer’s risk, and is denoted by the Greek letter α.

The second risk is that the consumer will accept a lot that doesn’t meet the conditions, i.e., by the
luck of the draw there are not many nonconforming items in the sample, so the lot is accepted.
This is the consumer’s risk and is denoted by the Greek letter β.

The literature contains a variety of typical values for α and β, but common values are 5% and
10%. When we locate these values on the OC curve, expressed in terms of probability of
acceptance, we actually locate 1 – α.




100.0%
n= 50, c=1

80.0%
n=100, c=2
60.0%
n=200, c=4
40.0%

20.0%

0.0%
0.0% 2.0% 4.0% 6.0% 8.0% 10.0% 12.0% 14.0%

Figure 3 As n Increases the OC Curve Approaches the Ideal


100.0%

80.0% n=100, c=2

60.0% n=100, c=1

40.0% n=100, c=0

20.0%

0.0%
0.0% 2.0% 4.0% 6.0% 8.0% 10.0% 12.0% 14.0%

Figure 4 As c Increases, for Fixed n, the OC Curve Approaches the Ideal



These points correspond to specific values of lot quality and they have a variety of names. The
point associated with 1 – α is often called the Acceptable Quality Limit or AQL. This is not
necessarily the same AQL used to describe the ideal OC curve. For an α of 5% this means a
process operating at the AQL will have 95% of its lots accepted by the sampling plan.

Similarly, the point associated with β is often called, in contrast, the Rejectable Quality Limit or
RQL. A process operating at the RQL will have 5% of its lots accepted by the sampling plan.

Lastly, some authors consider the process quality where the lots have a 50% probability of
acceptance. This is called the Indifference Quality Limit or IQL.

Figure 5 illustrates these points.


100.0%
1-α

80.0%

60.0%
50.0%
40.0%

20.0%
β
0.0%
AQL IQL RQL
0.0% 10.0% 20.0% 30.0% 40.0% 50.0%

Figure 5 Specific Points on the OC Curve

A Stream of Lots and the Binomial Distribution
We described the OC curve in terms of a process that produces a series of lots. This leads us to
recognize that the underlying distribution is the binomial. In the binomial distribution, there are
two possible outcomes. The items in the sample are either conforming or nonconforming. In
addition, the probably of selecting a nonconforming item doesn’t change as a result of the
sample. Since we are sampling from a process, the potentially infinite number of items is not
impacted by taking the sample.

When the producer presents lots for acceptance, they often come from a process that is operating
at some quality level, i.e., the process produces a certain percentage of nonconforming items.
The probability of obtaining a specified number of nonconforming items, Pr(x), from a sample of
n items with percent nonconforming, denoted p, is given by the binomial distribution.



⎛n⎞
Pr( x) = ⎜ ⎟ p x (1 − p )n − x , x = 0,1, … , n
⎜ x⎟
⎝ ⎠

In a single sample plan we accept the lot if the number of nonconforming items is c or less. This
means we are interesting in the probability of 0, 1, …, c items. We write this as

c
⎛n⎞
Pr( x ≤ c) = ∑ ⎜ i ⎟ p (1 − p )
⎜ ⎟
⎝ ⎠
i =0
x i−x

The probability of accepting the lot is the probability that there are c or fewer nonconforming
items in the sample. This is the equation above, and is what we plot as the OC curve.

The Isolated Lot And The Hypergeometric Distribution
The binomial distribution applies when we consider lots coming from an ongoing production
process. Sometimes we consider isolated lots, or we are interested in a specific lot. In these
cases, we need to realize that taking the sample, because we sample without replacement,
changes the probability of the next item in the sample. In these cases, we need the
hypergeometric distribution.

⎛ d ⎞⎛ N − d ⎞
⎜ ⎟⎜
⎜ x ⎟⎜ n − x ⎟
⎟
Pr (x ) = ⎝ ⎠⎝ ⎠
⎛N⎞
⎜ ⎟
⎜n⎟
⎝ ⎠

Here, N is the lot size, n is the sample size, and d is the number of nonconforming items in the
lot.

If we are interested in determining the probability of c or fewer nonconforming items in the
sample then we write:

⎛ d ⎞⎛ N − d ⎞
c
⎜ ⎟⎜
⎜ i ⎟⎜ n − i ⎟
⎟
Pr (x ≤ c ) = ∑ ⎝ ⎠⎝
⎛N⎞
⎠
i =0 ⎜ ⎟
⎜n⎟
⎝ ⎠

We can use this equation to draw the OC curve for the isolated lot.

Often, the β risk is applied to each lot, instead of the stream of lots. In these cases, the quality
level corresponding to a probability of acceptance equal to β is called the Lot Tolerance Percent
Defective (LTPD).

Single and Double Sample Plans
The material above discusses sampling plans in which we draw one sample from the lot. This is
called a single sample plan. We describe the plan by a set of parameters:
n is the sample size,


c is the maximum number of nonconforming items allowed for acceptance, and
r is the minimum number of nonconforming items allowed for rejection.

In a single sample plan r and c differ by 1.

In contrast, there are double sampling plans in which we take the first sample and make one of
three decisions: accept, reject, or take a second sample. If we take the second sample, we then
make an accept/reject decision.

As described above the set of parameters used to describe a double sample plan are:
ni is the ith sample size,
ci is the maximum number of nonconforming items allowed for acceptance on the ith
sample, and
ri is the minimum number of nonconforming items allowed for rejection on the ith
sample.

For example, a single sample plan may be:

n = 20, c = 2, r = 3.

A double sample plan may be:

n1 = 20, c1 = 1, r1 = 4
n2 = 20, c2 = 4, r2 = 5

In this example, if we had 2 nonconforming items on the first sample, we would draw the second
sample. In total, we would have sampled 40 items.

We can calculate the probability of acceptance, the information we need to define the OC curve,
by the following equation.

ri −1
Pr (xi ≤ ci ) + ∑ Pr(x
i = ci +1
1 = i ) × Pr (x2 ≤ c2 − i )

The probabilities are, as described above, calculated using either the binomial or hypergeometric
distributions.

The c=0 Sampling Plans
Many practitioners are concerned that traditional lot acceptance sampling plans allow
nonconforming items in the sample. For example, the single sample plan n=20, c=2, r=3 allows
as many as two nonconforming items in the sample.

One solution is the use of plans that don’t allow any nonconforming items. One example of plan
is n=20, c=0, r=1. The consumer would reject the lot if any nonconforming items appeared in
the sample.



Sampling plans with c=0 don’t have the same kind of OC curve discussed above. Instead of the
classic “S” shape, that starts to approximate the ideal curve, c=0 OC curves drop off sharply
without the bend. Figure 6 shows the OC curves for these two plans.

Notice how quickly the c=0 plan drops off. The figure also has dashed horizontal lines at 5% and
95% probability of acceptance. The AQL and RQL for these plans are listed below.

c=0 c=2
AQL 0.26% 4.21%
RQL 13.9% 28.3%

It is easy to see that c=0 plan will accept many fewer lots than the corresponding c=2 plan. If
your process cannot tolerate even a few nonconforming units, c=0 plans may be a good
approach. However, recognize that lot rejection incurs a transaction cost, that may be high. The
selection is a c=0 plan is certainly an economic decision.


100.0%

80.0%
n=20, c=2

60.0%
n=20, c=0
40.0%

20.0%

0.0%
0.0% 5.0% 10.0% 15.0% 20.0% 25.0% 30.0% 35.0% 40.0% 45.0% 50.0%

Figure 6 OC Curve comparison Showing c=0 Effect



The oc curve_of_attribute_acceptance_plans

More Related Content

What's hot (16)

Viewers also liked (20)

Similar to The oc curve_of_attribute_acceptance_plans (14)

More from Ankit Katiyar (20)

The oc curve_of_attribute_acceptance_plans