Quality and Performance
Chapter 3
Part 2: SPC
Acceptance Sampling
The application of statistical techniques to determine
whether a quantity of material should be accepted
or rejected based on the inspection or test of a
sample
Acceptable Quality Level (AQL): the proportion of
defective items that are acceptable to the buyer
Less expensive to test a sample, rather than all of
the items
Risk: the sample may not be representative of the
batch
2/34
Acceptance Sampling Procedure
1. Take a random sample from a large quantity of
items, and test the sample for various quality
measures
2. Accept the batch if the number of defects in the
sample is below a threshold value
3. If the sample does not pass:
1. Inspect all items in the batch, repair or
replace defectives, OR
2. Return the entire batch to the supplier
Acceptance sampling is one way to ensure quality in the supply chain
3/34
1
Notice how the buyer’s
specifications become the
supplier’s targets.
4/34
Statistical Process Control (SPC)
Statistical process control is the application of
statistical techniques to determine whether a
process is delivering what the customer wants.
SPC measures changes in a process by analyzing a
sample of the output
SPC can answer the following questions:
Does output conform to design?
Is there systematic deviation?
Can we identify and eliminate the causes of variation?
5/34
Performance Measures
Variables: Service or product characteristics that can
be measured, such as weight, length, volume, or
time.
Length of time to deliver a package
Diameter of a piston
Weight of a box of cereal
Attributes: Service or product characteristics that
can be counted.
Number of late deliveries
Number of defective pistons
Number of cereal boxes that are under the
required weight
6/34
2
Sampling
Sampling plan: A plan that specifies a
sample size, the time between
successive samples, and decision rules
that determine when action should be
taken.
Sample size: A quantity of randomly
selected observations of process
outputs.
7/34
Sample Means and
the Process Distribution
Sample statistics have their own distribution, which
we call a sampling distribution.
8/34
Sampling Distributions
The sample mean is the sum of the observations divided
by the total number of observations
where
xi = observation of a quality characteristic (such as time)
n = total number of observations
= mean
9/34
3
Sampling Distributions
The range is the difference between the largest observation
in a sample and the smallest. The standard deviation is the
square root of the variance of a distribution. An estimate of
the process standard deviation based on a sample is given by
where
σ = standard deviation of a sample
xi = observation of a quality characteristic (such as time)
n = total number of observations
= mean 10/34
Process Distributions
o A process distribution can be characterized by its
location, spread, and shape.
o Location is measured by the mean of the
distribution and spread is measured by the range or
standard deviation.
o The shape of process distributions can be
characterized as either symmetric or skewed.
o A symmetric distribution has the same number of
observations above and below the mean.
o A skewed distribution has a greater number of
observations either above or below the mean. 11/34
Causes of Variation
Common causes: random and unavoidable.
Tool imprecision
Weather conditions
Traffic congestion
Assignable causes: can be identified and
eliminated.
Tool wear
Machine out of calibration
Human error
12/34
4
Common Causes of Variation
Random and unavoidable
Consider the distribution of observations
Mean
Spread: dispersion about the mean
Range: difference between largest
and smallest observation
Standard deviation
Shape
Symmetric
Skewed 13/34
Common Causes
Standard Deviation/
Mean Spread
This is variation that
we accept as
unavoidable.
14/34
Assignable Causes of Variation
Can be identified and eliminated.
Process is “out of control” when there
are assignable causes of variation that
have not been eliminated.
This is variation that we can do
something about!
15/34
5
16/34
17/34
18/34
6
19/34
20/34
21/34
7
22/34
23/34
24/34
8
Performance Measures
Variables: measurable product characteristics
Product: physical measurements
Service: time
Precise information about problem
May be difficult to collect and analyze
Examples include size of table legs; length of delivery time
Attributes: countable product characteristics
Number of defects
Proportion of defective products
Easier to collect, provide less information
Examples include number of defective table legs; number of
late deliveries
Consider the tradeoff between the cost of finding/avoiding quality
problems and the cost of not achieving quality.
Often both performance measures are used.
25/34
How much data do we need?
Consider costs of inspection vs. costs of
failure.
Complete inspection is advisable when the
cost of failure is very high
Nuclear fuel tubes
Pacemakers
26/34
Decisions about SPC
What are the managerial decisions
involved in using Statistical Process
Control?
For a brief discussion of some of these
issues, click here.
27/34
9
Control Charts
Show distribution of observations in relation
to target and limits.
Center line = target value.
UCL = upper control limit
LCL = lower control limit
Process is in control if observations from a
random sample fall between the two limits.
28/34
29/34
30/34
10
31/34
32/34
Using Control Charts for
Process Improvement
1. Take a random sample and measure the
process
2. Plot statistics: when problems are indicated,
seek the assignable cause
3. Eliminate the cause if it is a problem,
incorporate the cause it is an improvement
4. Repeat the cycle periodically
33/34
11
Possible Errors
Type I: random error is considered due to assignable
causes.
Type II: out-of-control process is considered in control.
Trade-off between these two types of errors and amount
of data, control limits.
Decisions about control limits:
tighter limits more expensive to control
Depends on costs of control vs. costs of failure.
34/34
12
