1

UNIT 7: STATISTICAL
PROCESS CONTROL

Control Charts for Variables

Control Charts for Attributes
 2

7.1. INTRODUCTION
 3

FOUNDATIONS

• In any production process, a certain amount of natural variability will always exist.

A process that is operating with only inherent causes of variation present is said
to be in statistical control.

• The output of a process can be modified or deteriorated by improperly adjusted or

controlled machines, operator errors, or defective raw material (assignable causes

of variation). It is said to be an out-of-control process.

• The mean and the standard deviation of the process can shift to out-of-control

values due to assignable causes. In that case, the process output will not conform
to requirements.
 4

FOUNDATIONS

• No process is truly stable forever, and, eventually, assignable causes will occur.

• A major objective of statistical process control is to quickly detect the occurrence of

assignable causes of process shifts, so that a corrective action may be undertaken.

• The control chart is an on-line process-monitoring technique widely used for this

purpose.

• The control chart may also provide information useful in improving the process. The

goal of statistical process control is reducing variability as much as possible.

5
 6

CONTROL CHART
• The control chart is a graphical display of a characteristic measured from a sample, e.g.

sample average, VS the sample number or time.

• The center line represents the average

value of the quality characteristic

corresponding to the in-control state.

• As long as the points plot within the

control limits (upper and lower lines),

the process is assumed to be in control,
and no action is necessary.
 7

CONTROL CHART

• If a point plots outside of the control limits, or the points behave in a systematic or

nonrandom manner, is interpreted that the process is out of control, and the
assignable causes for this behavior should be found and eliminated. If so, the
process performance can be improved.

• If the process is in control, all the plotted points should have a random pattern.

• There is a close connection with hypothesis testing: if the sample average 𝑥̅

exceeds either control limit, we conclude that the process mean is out of control,
which is equal to some value of the process mean μ1 ≠ μ0 (i.e. rejecting the
hypothesis of process mean equal to μ0).
 8

EXAMPLE
Semiconductor manufacturing: an important measure during the last step is the flow
width of the resist. Samples of size n = 5 are take every hour. The sample average 𝒙 is
plotted in order to monitor the process

mean.

If the process mean is 1.5, the

standard deviation of the sample
average 𝑥̅ is:
 9

EXAMPLE
• If the process is in control, 100(1-α)% of the sample means 𝑥̅ would fall between 1.5 +

Zα/2(0.0671) and 1.5 - Zα/2(0.0671). If we choose Zα/2 = 3:

• It is equivalent to setting up the critical region for testing the hypothesis about the

process mean, whose standard deviation is known (0.15):

 10

EXAMPLE
 11

GENERAL MODEL
• Let w be a sample statistic that measures some quality characteristic of interest,

whose mean and standard deviation are μw and σw, respectively. Then the control limit
become:

• The “distance” L of the control limits from the center line is expressed in standard

deviation units.
 12

TYPES OF CONTROL CHARTS

• Control Charts for Variables: If the quality characteristic can be measured and

expressed as a number on a continuous scale of measurement, it is called a variable.

In such cases, the characteristic is described with a measure of central tendency and
a measure of variability.

• Control Charts for Attributes: we may count the number of nonconforming units

(whether or not it possesses certain attributes), or the number of nonconformities

(defects) appearing on a unit of product.

• The design of the control chart consists in the selection of the sample size, control

limits, and frequency of sampling.

 13

CHOICE OF CONTROL LIMITS

• By moving the control limits farther from the center line, we decrease the risk of

a type I error, that is, the risk of a point falling beyond the control limits, indicating
an out-of-control condition when no assignable cause is present.

• However, widening the control limits will also increase the risk of a type II error,

i.e. the risk of a point falling between the control limits when the process is really
out of control.

• If we move the control limits closer to the center line, the opposite effect is obtained:

The risk of type I error is increased, while the risk of type II error is decreased.
 14

CHOICE OF CONTROL LIMITS

• In the previous example, where 3-sigma control limits were used, if we assume

that the flow width is normally distributed, we find from the standard normal table
that the probability of type I error is 0.0027 (and the probability that an in-control
process will exceed the three-sigma limits in one direction is 0.00135).

• Another option is to choose the type I error probability and calculate the

corresponding control limit. If we specified a 0.001 type I error probability in one

direction, then the appropriate multiple of the standard deviation would be 3.09.
They are called 0.001 probability limits:
 15

CHOICE OF CONTROL LIMITS

• We could add additional limits, which are called warning limits (usually at 2-sigma

distance):

• When probability limits are used, warning limits are

0.025 limits.

• If one or more points fall between the warning limits and

the control limits, we could increase the sampling

frequency and/or the sample size so that more
information about the process can be obtained quickly.
 16

SAMPLE SIZE AND SAMPLING FREQUENCY

• In general, larger samples will make it easier to detect small shifts in the process

mean.

• When choosing the sample size, we

must keep in mind the size of the shift

that we are trying to detect.

• Taking large samples very frequently is

usually not economically feasible.

Curva de Operación Característica.
 17

SAMPLE SIZE AND SAMPLING FREQUENCY

• ARL (Average Run Length) is the average number of points that must be plotted

before a point indicates an out-of-control condition. If p is the probability that any

point exceeds the control limits:

• For the example with 3-sigma limits, the value ARL0 (when the process is in

control) is:

• The drawback is that the standard deviation of the run length is very large (in the

example, the standard deviation of ARL0 is also 370).

 18

SAMPLE SIZE AND SAMPLING FREQUENCY

• If the process goes out of control, ARL1 points out the number of samples to detect

the process shift. If n=5 and the process mean shifts to 1.725, the probability of 𝑥̅
falling between the control limits is approximately 0.35.:

1 1
ARL1    1.54
p 1  0.35

• If we want to reduce the time needed to detect the out-of-control condition, we

could sample more frequently or increase the sample size.

 19

ANALYSIS OF PATTERNS
• A control chart may also indicate an out-of-control condition when the plotted points

exhibit some nonrandom pattern of behavior. (If the process were in control, all points
should appear to be random).

• We note that 19 of 25 points plot below the

center line, while only 6 of them plot above.

This control chart also has an unusually
long run up (five points in a row increase in
magnitude) and an unusually long run
down.
 20

ANALYSIS OF PATTERNS • A cyclic behavior may indicate a problem with

the process, such as operator fatigue, raw
material deliveries, heat buildup, and so on.

• Pattern recognition can improve with

the inclusion of 1-sigma lines, thus

splitting the chart into 6 zones.
 21

ANALYSIS OF PATTERNS
The pattern is nonrandom and the process is out of control if:
1. One or more points outside of the control limits.

2. Two of three consecutive points outside the two-sigma warning limits but still inside the control
limits.
3. Four of five consecutive points beyond the one-sigma limits.
• Warning: when using several
4. A run of eight consecutive points on one side of the center line.
decision rules simultaneously,
5. Six points in a row steadily increasing or decreasing.
the overall type I error (false
6. Fifteen points in a row in zone C. alarm probability) increases:
7. Fourteen points in a row alternating up and down.

8. Eight points in a row on both sides of the center line with none in zone C.

9. An unusual or nonrandom pattern in the data.

10. One or more points near a warning or control limit.

 22

PHASE I & PHASE II

• Phase I: set of process data is gathered and analyzed in a retrospective analysis,

constructing trial control limits. Points that are outside the control limits are
investigated, looking for potential assignable causes and making an effort to
eliminate them. Points outside the control limits are then excluded and a new set of
revised control limits calculated. After several cycles, the process is stabilized.

• Phase II: we use the control chart to monitor the process by comparing the

sample statistic for each successive sample as it is drawn from the process to the
control limits.
 23

7.2. CONTROL CHARTS

FOR VARIABLES
 24

INTRODUCTION
• A single measurable quality characteristic (can be expressed in terms of a numerical

measurement), such as a dimension, weight, or volume, is called a variable.

• it is usually necessary to monitor both the mean value of the characteristic (sample

average 𝑥̅ ) and its variability (range R or standard deviation s).

• Shifts in the process mean or standard deviation can result in a high fraction of

nonconforming products (process output falling out of the specification limits).

 25

STATISTICAL BASIS

• The sample average 𝒙 is:

• We know that 𝑥̅ is normally distributed (central limit theorem) with mean μ and

standard deviation σ ̅ σ⁄ 𝑛. If σ is known, the probability is 1 – α that any sample

will fall between (confidence interval):

• 3-sigma limits: quantile Zα/2 is replaced by 3.

 26

STATISTICAL BASIS

• In practice, μ and σ are not known. In that case, the best estimator of μ is the

grand average. Suppose that m samples are available, each containing n

observations on the quality characteristic:

• The range of each sample is the difference between the largest and smallest

observation:

• The average range of m samples:

 27

STATISTICAL BASIS

• Control limits on the 𝒙 chart:

• Control limits on the R chart (monitors the process variability):

 28

STATISTICAL BASIS

• There is relationship between the Range of a sample from a normal distribution and

the standard deviation of that distribution (σ): random variable W = R/σ, called
relative range. The parameters of the distribution of W are a function of the sample
size n (the mean of W is d2). Thus, an unbiased estimator of σ is:

• If we use 𝑥̿ as an estimator of μ and 𝑅 /𝑑 as an estimator of σ:

 29

STATISTICAL BASIS

• Now consider the R chart. The standard deviation of W, say d3, is a known function

of n. Since R = W∙σ, we may estimate the standard deviation of R by:

• The parameters of the R chart with the usual 3-sigma control limits are:
 30

EXAMPLE

• We wish to establish statistical

control of the flow width of the

resist in a semiconductor
manufacturing process using
𝑥̅ and R charts.

• Twenty-five samples, each of

size five wafers, have been

taken when we think the
process is in control.
 31

EXAMPLE

• Control limits for R chart:

• The values for D3 and D4 can be found at the Table of Factors:

 32

EXAMPLE

• Control limits for 𝑥̅ chart :

• The values for A2 can be found at the Table of Factors:

 33

PROCESS CAPABILITY

• The control chart data may be used to describe the capability of the process to

produce units according to the specifications

• In the previous example, specification limits are 1.50±0.50, and the process

standard deviation can be estimated as:

• We may estimate the fraction of

nonconforming units produced as:
 34

PROCESS CAPABILITY

• Another option is the process capability ratio Cp, where the denominator is the

difference between the natural tolerance limits in the process:

• In the previous example, by using σ as an estimate of σ:

• We can also compute the percentage of the specification band that the process
uses up:
 35

CONTROL LIMITS VS SPECIFICATION LIMITS

• The control limits are driven by the

natural variability of the process

(measured by the process standard
deviation σ).

• There is no mathematical or
statistical relationship between the
control limits and specification
limits. The latter are determined
externally.
 36

PHASE I & PHASE II

• Phase I: each of the out-of-control points are examined, looking for an assignable

cause. If an assignable cause is found, the point is discarded and the trial control
limits are recalculated.

• Once a set of reliable control limits is established, we use the control chart for

monitoring future production. This is called phase II control chart usage. When a
point plots out of the control limits, we should investigate if an assignable cause has
occurred at or before that time.
 37

EXAMPLE

Phase II Operation:
twenty additional
samples were
collected after the
control charts were
established.
 38

EXAMPLE: PHASE II

• Samples 43 and 45

plot above the UCL.

• In addition, the
general pattern of
points on the chart
from sample 38
onward is indicative
of a shift in the
process mean.
 39

EXAMPLE: TOLERANCE DIAGRAM

• It is sometimes helpful

to construct a run chart

of the individual
observations in each
sample.

• It may reveal a pattern

in the data, or show that

a particular value was
produced by one or two
unusual observations in
the sample.
 40

INTERPRETATION

• In interpreting patterns on the 𝑥̅ chart, we must first determine whether or not the R

chart is in control. If both charts exhibit a nonrandom pattern, the best strategy is to
eliminate the R chart assignable causes first.

• Cyclic patterns: on the 𝑥̅ chart may result

from systematic environmental changes such

as temperature, operator fatigue, regular
rotation of operators and/or machines. R
charts will sometimes reveal cycles because
of maintenance schedules, operator fatigue,
or tool wear.
 41

INTERPRETATION
• Mixture: very few points near the center

line. It is generated by two overlapping

distributions (parallel machines) or
because of making process
adjustments too often.

• Shift: may result from the introduction

of new workers; changes in methods,

raw materials, or machines; or a change
in either the skill or motivation of the
operators.
 42

INTERPRETATION
• Trend: are usually due to a gradual
deterioration of a tool, or operator fatigue.

• Stratification: tendency for the points to

cluster artificially around the center line. Due

to incorrect calculation of control limits or
when the sampling process collects one or
more units from several different underlying
distributions within each subgroup.
 43

OPERATING-CHARACTERISTIC FUNCTION

• The probability of not detecting a shift from the in-control mean value μ0 to

another value μ1= μ0+ kꞏσ:

• For L=3, n=5 and a shift of 2σ units:

 44

OPERATING-CHARACTERISTIC CURVE
• The OC curve plots the β-risk against the

magnitude of the shift we wish to detect

expressed in standard deviation units (k∙σ) for
various sample sizes n.

• For the example, if the shift is 1∙σ and n=5, we

have β=0.75. Thus, the probability that the shift

will be detected on the first sample is only 1-
β=0.25. And the expected number of samples
taken before the shift is detected:
 45

ARL (AVERAGE RUN LENGHT)

• Expected number of samples until one point is

plotted out of the limits, which is equal to the inverse

of the probability of a point being out of the limits. For
an in-control and out-control process, respectively:

• There are ARL curves that show the expected

number of samples taken in order to detect a certain

shift in the 𝑥̅ chart.
 46

7.3. CONTROL CHARTS

FOR ATTRIBUTES
 47

INTRODUCTION

• Not all quality characteristics can be expressed with variables data. Some products

could be classified into one of the two categories called conforming or

nonconforming, depending on whether the product meets the requirements on
one or more quality characteristics.

• Attributes data: a control chart for the fraction of nonconforming units could be

established. Alternatively, in some processes we may examine a unit of product and

count defects or nonconformities on the unit.
 48

STATISTICAL BASIS

• The fraction nonconforming is defined as the ratio of the number of

nonconforming items in a population to the total number of items. The items may
have several quality characteristics that are examined simultaneously.

• The probability that any unit will not conform to specifications is p. If a random

sample of n units of product is selected, and if D is the number of nonconforming

units, then D has a binomial distribution with parameters n and p;

• The mean and variance of D are n∙p and n∙p(1-p), respectively.

 49

STATISTICAL BASIS

• The sample fraction nonconforming is defined as the ratio of the number of

nonconforming units in the sample D to the sample size n:

• The mean and variance of 𝒑 can be obtained from variable D (binomial):

• It is called the p chart.

 50

CONTROL CHART
• By using the mean and the variance of the statistic w that measures the quality

characteristic (L=3):

• We will work with statistic 𝒑 (sample fraction nonconforming).

• If the true fraction nonconforming p in the production process is known:

• If the lower control limit turns out to be < 0, we set

LCL = 0, and assume that the control chart only

has an upper control limit.
 51

CONTROL CHART

• When p is unknown, it must be estimated from observed data. The usual procedure

is to select m samples, each of size n. Then if there are Di nonconforming units in

sample i, we compute the fraction nonconforming in the ith sample as:

• The average for all simples estimates

fraction nonconforming p:
• The limits of the

control chart are:

 52

CONTROL CHART

• The control limits previously defined should be regarded as trial control limits

(phase I). Any points that exceed the trial control limits should be investigated. If
assignable causes for these points are discovered, they should be discarded and
new trial control limits determined.

• We would usually be given a standard value of p that represents a desired or target

value for the process fraction nonconforming. If future samples indicate an out-of-
control condition, we must determine whether the process is out of control at the
target p but in control at some other value of p.
 53

• 30 samples of n = 50 cans each were selected at half-hour intervals over a

EXAMPLE three-shift period in which the machine was in continuous operation:
 54

EXAMPLE

• First of all, We construct a phase I control chart using this preliminary data to

determine if the process was in control when these data were collected.

• Using 𝑝 as an estimate of the true process fraction nonconforming, we can now

calculate the control limits:

 55

EXAMPLE
• We note that 2 points plot

above the UCL, so the

process is not in control.
After investigating both
point, assignable causes
are determined.

• Consequently, samples
15 and 23 are eliminated,
and new control limits are
Calculated.
 56

EXAMPLE

• The revised control limits are calculated as:

 57

EXAMPLE
• Analysis of sample 21 does
not produce any reasonable
or logical assignable cause
for this, so we can retain the
point.
• We conclude that the
process is in control at level
p=0.215. The the revised
control limits should be
adopted for monitoring
production (phase II).
 58

EXAMPLE
• After machine adjustments, 24 samples of n = 50 observations each are collected:
 59

EXAMPLE

• Point from sample 41 is below LCL. However, no assignable cause for this out-of-control

signal can be determined.

 60

EXAMPLE

• We may formally test the hypothesis that the process fraction nonconforming after

the machine adjustments differs from the process fraction nonconforming in the
preliminary data:

• The test statistic for the hypothesis:

 61

EXAMPLE

• In our example:

• Comparing this to the upper 0.05 point of the standard normal distribution, we find

that Z0 = 7.10 > Zα/2 = 1.645. Consequently, we reject H0 and conclude that there
has been a significant decrease in the process fallout.
 62

EXAMPLE

• It seems logical to
revise the control limits
again, using only the
most recent samples
(numbers 31–54).

• Since the calculated

lower control limit is
less than zero, we
have set LCL = 0.
 63

EXAMPLE • Data for the process during next five shifts:

 64

EXAMPLE • Control chart does not indicate lack of control:

 65

DESIGN OF THE CONTROL CHART

• Various rules have been suggested for the choice of n.

• If p is very small, we should choose n sufficiently large so that we have a high

probability of finding at least one nonconforming unit in the sample. Otherwise, we

the presence of only one nonconforming unit in the sample would indicate an out-
of-control condition.

• For example, if n = 8 and p = 0.01:

• If there is one nonconforming unit in the sample, 𝑝̂ > UCL, which is unreasonable :
 66

DESIGN OF THE CONTROL CHART

• Suppose that p = 0.01, and we want the probability of at least one nonconforming

unit in the sample to be at least 0.95:

P  D  1  0.95  1  P  D  0   0.95  P  D  0   0.05

• Using the Poisson approximation to the binomial, we find that λ = n∙p must exceed

3. Since p = 0.01, this implies that the sample size n should be 300.
 67

DESIGN OF THE CONTROL CHART

• Another option is to choose a sample size large enough to have approximately a

50% chance of detecting a process shift of some specified amount δ.

• We should choose n so that the upper control limit exactly coincides with the

fraction nonconforming in the out-of-control state:

• Suppose that p = 0.01, and we want the probability of detecting a shift to p = 0.05 to

be 0.50:
 68

DESIGN OF THE CONTROL CHART

• If the value of p is small, another criterion is to choose n large enough so that the

control chart will have a positive lower control limit. This force us to investigate
samples containing an unusually small number of nonconforming items.

• If p=0.05 y and 3-Sigma limits are used:

 69

NP CONTROL CHART
• It is also possible to base a control chart on the number nonconforming rather

than the fraction nonconforming:

• If a standard value for p is unavailable, then 𝑝̅ can be used to estimate p.

 70

NP CONTROL CHART: EXAMPLE

• For the previous example, n =

50 and 𝑝̅ 0.213:

• Some practitioners prefer to

use integer values for control

limits on the np chart.
 71

OPERATING-CHARACTERISTIC FUNCTION
• The operating-characteristic (or OC) function is a graphical display of the probability

of incorrectly accepting the hypothesis of statistical control (i.e. type II error)

against the process fraction nonconforming. In other words, it shows the ability to
detect a shift in the process fraction nonconforming from the nominal value 𝑝̅ to
some other value p.

• Since D is a binomial random variable, D must be an integer. If LCL<0, the second

term on the right-hand side of the equation should be dropped.

 72

OPERATING-CHARACTERISTIC FUNCTION
• For a control chart with parameters 𝑝̅ = 0.2, n=50, UCL=0.3697 and LCL=0.0303:

• When p<0.1 (and sample size is large), the Poisson approximation λ = n∙p can be

used. In case p takes a greater value (at least p=0.5 and n>10; for smaller values of p,
a larger sample size is required, provided that nꞏp>4.5), the Normal approximation
could be used (with mean = n∙p and variance = n∙p(1-p)).
 73

OPERATING-CHARACTERISTIC CURVE
• Calculations to generate the OC curve for a control chart with parameters n=50, UCL=0.3697

and LCL=0.0303:
 74

OPERATING-CHARACTERISTIC CURVE
• The OC curve is plotted in the following figure:

• Expected number of samples until one point plots out

of the limits:

1
ARL 
p(sample point plots out of control)

• If the process is in control: • If the process is out of control:

 75

OPERATING-CHARACTERISTIC CURVE
• The probabilities α and β can be calculated directly from the binomial distribution or

read from an OC curve. Consider a process with 𝑝̅ = 0.2, n=50, UCL=0.3697 and
LCL=0.0303. If the process is in control with p = 𝑝̅ , we will have α = 0.0027:

• If the process shifts out of control to p=0.3, β=0.8594:

• Increasing the sample size would result in a smaller value of β and a shorter out-of-

control ARL1.