STAT8010

Applied Statistical Techniques

Process Performance
Introduction to Control Charts
 Prevention vs Detection
Historically…
– Manufacturing builds product.
– QC inspects (and accepts/rejects) product.
– This is a detection strategy which is wasteful.

A prevention strategy prevents bad products being built in

the first place.
– Requires an understanding of the elements of Statistical Process
Control.

Building bad parts costs the same (or even more)

as building good parts.
2
 Variation

Is this just something that is part of the process?

3
4
 Causes of Variability
• Chance causes or common causes
– Numerous small causes of variability
inherent to a system or process
– Operate randomly or by chance
– Once a reason for a common cause of
variation has been found, it should be treated
as special and removed.

• Assignable or special causes of

variability
– Have relatively large effects on the process
and are not inherent to it
– The circumstances or factors that cause this
kind of variability can be identified
 Causes of Variation

A stable process is in a state of statistical control and has only common causes of variability acting on it
A process is said to be out of statistical control if one or more special causes are operating on it
 Process Actions
• If special cause is detected:
– Action to eliminate the cause can generally be taken by those close to the
process or operation.

• If only common cause is present:

– The present and future capability of the process is known.
– Reducing variation requires action on the system. This is a conscious
decision taken by management.

Any attempts to adjust processes and treat common causes as special causes
will tend to increase variability, not reduce it. 7
 vs.

Control Capability
Stability over time. Ability of a stable
process to meet
specifications.

8
 Control vs Capability

Capable (Do we meet specifications?)

Yes No
Process is predictable but there is too
In Control (Is the Yes Desired Condition much variation to consistently meet
process stable?) requirements
Not desired, but may be
No acceptable in some Chaos!
circumstances
Variation over time
 Control Charts
Control charts can be used to monitor process quality over
time to (ie. Process could be the manufacturing of an item):
1. Assess process stability = variability over time
2. Assess process capability
3. Aid in process improvement
 Theory of Control Charts
Control charts plot time on the x-axis, and outcomes on the y-
axis. Bands demarcate degree of variation from mean or
expected process performance:
Based on the fact that chance
variation follows known patterns or
statistical distributions.
These distributions are used to
estimate the probability of individual
measurements or sample statistics
by chance alone.
 Theory of Control Charts
• Center Line (CL) of a Control Chart

– Represents the mean of the process in

control.  
• Upper and Lower Control Limit
(UCL & LCL) of a Control Chart
– Lines that represent the limits to the
variation of a process places at +3𝜎
above and -3𝜎 below the mean
– An observation above the UCL or below
the LCL suggests an uncontrolled or
special-cause variation.
 Types of Error in Control Charts

Actual State of Process

Only Common Special Causes

Cause

Out of Control False Alarm Correct Decision

Control Chart
Indicators
In Control Correct Decision Failure to Detect

14
 Out of Control Signals and Patterns
A process is considered out of control if any of the following occur: 
1. A point falls above the upper control limit or below the lower control limit.
2. Two out of three consecutive points fall above the +2 limits or two out of
three consecutive points fall below the -2 limits.
3. Four out of five consecutive points fall above the +1 limit or four out of five
consecutive points fall below the -1 limit.
4. Eight or more consecutive points lie above the center line or eight or more
consecutive points lie below the center line.
5. Eight or more consecutive points move upward in value or eight or more
consecutive points move downward in value.
Out of Control Signals and Patterns
 Out of Control Signals and Patterns
Trends, ascending or
descending:
Due to gradual wear of process
component or deterioration of
process parameters (chemical
process)

Mixture Pattern: points in the

outer bands of the control limit
window and few near the centre
line. Seen as overlapping
distributions of data. Possible
causes is “overcontrol” from
operators making adjustments.
 Out of Control Signals and Patterns
Cyclic pattern: possible causes
are operator fatigue,
environmental shifts like
temperature or humidity, shift
changes, rotation of workers or
machines, etc.

Stratification : points cluster

around center line. Possibly due
to incorrectly calculated control
limits or collecting samples that
are too dispersed in variation.
 Types of Control Charts
• Control charts can be divided into two categories that are
determined by the type of measurements used to monitor
a process:
1. Attribute control charts (measures discrete variables)
2. Variables control charts (measures continuous
variables)
Decision Tree for Control charts

2 <= n < 10
The Elements of a Process Management System
• What is meant by a Process Control System?
• Variation: Special or Common cause?
• Actions: Local or System?
• Control and Capability
• The Process Improvement Cycle
• Control Charts: Tools for control and
improvement.

21
22
 STAT8010
Applied Statistical Techniques

Control Charts for Variable Data

 Today’s Learning Outcomes

Construct and use variables control charts for process

management.
Apply the principles of rational subgrouping.
 Variables Control Charts
• Variables control charts can be used when a quality characteristic is being
measured on a continuous variable
• Usually used in pairs
– Variation chart -monitors short term variation and answers the question “Is the variation
within subgroups constant?”
– Location chart (Data chart)–monitors changes in the average value of the process and is
a visualization of the longer term variation.
 X bar and R Charts
•Probably
  the most commonly used chart

• R Chart – Subgroup range (R) is plotted which monitors process

variability
• chart – Subgroup average ( is plotted which monitors the central
tendency of the process
• We typically need to select and evaluate at least 25 samples of
size n
 X bar and R Charts
•Probably
  the most commonly used chart

• R Chart – Subgroup range (R) is plotted which monitors process

variability
• chart – Subgroup average ( is plotted which monitors the central
tendency of the process
• We typically need to select and evaluate at least 25 samples of
size n
 Range (R) Chart
•Centerline
  =
= individual sample or subgroup range for subgroup i
k = number of samples or subgroups

  UCL (R)
  LCL (R)
 X bar Chart
•Centerline
  = Process Average (𝜇) = Grand Mean of the
Sample Means =
= sample mean for subgroup i
k = number of samples or subgroups used

  UCL ()
  LCL ()
 X bar and S Charts
• For subgroup sizes n < 10 , the range provides a
reasonable estimate with greater simplicity and
efficiency than the standard deviation
• For samples of size n=10 or more, the X bar and s
charts can be used
• s chart examines the variability of the process
instead of the R chart
 s Charts
•  
= sample standard deviation for sample or subgroup i
k = number of subgroups

  UCL (s)
  LCL (s)
 X bar Chart (using s)
•Centerline
  = Process Average (𝜇) = Grand Mean of the
Sample Means =
= sample mean for subgroup i
k = number of samples or subgroups used

  UCL ()
  LCL ()
 Control Charts for Individuals (X charts)
• When it is not practical or possible to collect data on subgroups
– Low volume operations
– Lengthy intervals between output units
– When a single measurement represents work output for an entire time period
– Evaluation of the quality characteristic requires destruction of the item or unit and testing samples
greater than n = 1 might be prohibitively expensive

• Consider each individual observation as a “subgroup” of size n=1

• When there is only one observation in each subgroup, there will be no variation within
the subgroups and must estimate process variability another way
1. Evaluate with moving range
 Individual Value (X) Chart with the Moving Range
Estimates
•   process variability by considering the differences between consecutive
observations

Moving Range = = - |

estimates the process variability and is used as the centre line of the moving range chart

Note that the calculation of MR considers the absolute value of the difference between consecutive measurements
 Moving Range Chart
•UCL
  (MR)
LCL (MR)

If the chart shows the process to be out of control, we should

investigate and eliminate the special causes that can be identified

If the moving range charts shows the process to be in control, we can

be confident that will provide a good estimate of 𝜎 to proceed with
construction of the trial limits for the X chart
 Individual Value (X) Chart
Centerline
•   =
is the ith individual value
k is the number of individual values available for calculating trial limits
UCL(X) = + 2.66
LCL(X) = - 2.66

To estimate Sigma using Moving Range Average:

S = / 1.128
 Guideline on Control Chart Interpretation
• Process stability is reflected in the relatively constant variation exhibited in
Control Charts. If a process is stable, the likelihood of a point falling outside the
band between +/- 3σ is so small that such an occurrence is taken as a signal of
a special cause of variation. In other words, something abnormal is occurring
within your process. However, even though all the points fall inside the control
limits, special cause variation may be at work. The presence of unusual
patterns can be evidence that your process is not in statistical control. Such
patterns are more likely to occur when one or more special causes is present.
• Using control limits which are less than 3 sigma from the centerline may trigger
a hunt for special causes when the process is already stable.
 When to change Control limits?
1. When removing out-of-control data points. When a special cause has been
identified and removed while you are working to achieve process stability.
2. When replacing trial limits. When a process has just started up, or has
changed, you may want to calculate control limits using only the limited data
available. Once you have 20 or 30 groups of 4 or 5 measurements without a
signal, you can use the limits to monitor future performance.
3. When there are changes in the process. When there are indications that
your process has changed, it is necessary to re-compute the control limits
based on data collected since the change occurred.
 Guidelines for Sample Selection
• Small frequent samples will usually be more effective than large infrequent
samples
• Use small frequent samples whenever the process is a high-volume one and
units of output are produced rapidly
– Important to detect shifts as soon as possible after they occur
– The longer the interval between samples, the greater the risk that a large number of
output units will be unacceptable in terms of quality
• Use small frequent samples whenever many special causes are possible
• Increase the frequency of sampling whenever a special cause is suspected
 Guidelines for Sampling
• Small problems and minor variations at the beginning of a process
tend to become magnified as the process progresses
– Monitoring toward the beginning of a process will generally be of greater
benefit than waiting to monitor at the end
• Sample at points where the likelihood that a problem will occur is
greatest or where special causes are more likely to occur
• Sample where the potential for cost reduction is greatest should a
special cause occur
• Sample where is it feasible and safe to do so
 Rational Subgroups
• Subgroup data allow us to monitor two types of variability:
• Inherent random variability of the process
– Estimated by the variability within subgroups
– Monitored by the variability portion of the chart
• Variability in process average 
– Between-group variability reflects process change
– Monitored by the central-tendency portion of the chart
• If a process is in a state of statistical control, variability within subgroups will reflect only
inherent variability of the process and will be equal to the between-group variability
 Rational Subgroups
Two basic strategies to select subgroups:
1. Select units produced as closely together in time as possible
– Will minimize variation within subgroups
– Will increase chances of detecting between-group variability indicating process shifts
2. Randomly sample continuously throughout production or the sampling
interval
– Sample will then be representative of all units produced during the interval in which the
sample was selected
– Unfortunately, maximizes within-sample variation and is not as effective when it is
important to detect shifts in process average
 Frequency of Sampling
• The more frequently samples are drawn, the more sensitive the
chart will be to the presence of special causes and the more
quickly a shift in process average will be detected
• Need to balance sample size and frequency of sampling against
budgetary requirements, time and the costs of failing to detect a
shift in the process
 Homework

Compute the control limit for the x and s control charts

 Homework
• Ch 15: 15.3.3, 15.3.5, 15.4.1
 Homework

Compute the control limit for the x and s control charts

 STAT8010
Applied Statistical Techniques

Unit 6
Control Charts for Attribute Data
 Today’s Learning Outcomes
Chapter 6
– Attribute Control Charts
• np and p charts
• c and u charts
Decision Tree for Control charts

2 <= n < 10
 Attribute Control Charts
1. Counts of Nonconforming Items
charts:
a) Number of nonconforming charts
(np charts);
b) Fraction or proportion of
nonconforming charts (p Charts);
2. Area of Opportunity Charts:
c) Number of nonconformities per unit
(c charts);
d) Nonconformities per unit charts (u
charts).
Attribute Control Chart Formulas
 Attributes Control Charts
• Use when samples or subgroups of process output are evaluated and:
– the number of nonconforming items in a sample are counted or;
– the number of nonconformities per item are counted
• Number Nonconforming (np ) Chart
– # of nonconforming items in a sample is expected to be the proportion of
nonconforming items, p, times the size of the sample, n
• Proportion Nonconforming (p) Chart
– Proportion of nonconforming items in samples instead of number
– Use in cases where unequal sample sizes are used
 Assumptions for the Use of the np Chart
1. There are only two possible outcomes for the event: An item must be
found to be either conforming or nonconforming. No intermediate
values are possible.
2. The probability, p, of a nonconforming item is constant
3. The quality of successive items are independent
4. All samples contain the same number of items, n.

To obtain a good estimate of p (the probability of the process producing a

nonconforming item), we need to evaluate at least 20 to 25 samples or
subgroups and count the nonconforming items in each.
 Calculating Control Limits for np Charts
𝑘

 ∑
  𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑛𝑜𝑛𝑐𝑜𝑛𝑓𝑜𝑟𝑚𝑖𝑛𝑔 𝑖𝑡𝑒𝑚𝑠 𝑖𝑛 𝑎𝑙𝑙 𝑠𝑢𝑏𝑔𝑟𝑜𝑢𝑝𝑠
𝑖 =1
𝑋 𝑖

´𝑝 = 𝑘

∑ 𝑛𝑖
𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑖𝑡𝑒𝑚 𝑖𝑛 𝑎𝑙𝑙 𝑠𝑢𝑏𝑔𝑟𝑜𝑢𝑝𝑠 𝑖=1

  UCL 𝜎 =√ 𝑛𝑝 (1 − 𝑝)
 

  LCL   s
 Creating the np Chart
• The number of nonconforming items is plotted on the chart in the order in
which the samples were collected
• Plot all of the zones to determine if any violations of the rules for out of
control signals are present
– Minitab does not plot the zones but instead lists any of the points that violate zone rules
• If causes of out of control points can be identified and eliminated from the
process, then the data can be removed from the calculation of the limits and
new control limits calculated. Do not remove the data point completely.
• Limits can be used until a change in the process occurs either by design or
by a shift in process level
 Example – np chart
File [Link]

The example concerns the manufacture of gauze sponges. In 32 samples (k) of 600 sponges
(n=600) , 685 nonconforming sponges are found.
Total number of sponges inspected = n * k = 600 * 32 = 19200

Probability = Total nonconforming sponges / Total number inspected = 685/19200 then p̅ = 0.0357

CL = n * p̅ = 600 * 0.0357 =21.406

3√(n*p*(1-p) = 3√(600*0.0357*0.9643) = 13.628

Let’s try in Excel!

 Example – np chart in Excel
Manual Calculations:
1. Calculation the proportion nonconforming if
this is unknown.
2. Calculate the Centre Line = n * proportion
nonconforming
3. Calculation 3* Standard deviation = 3 *
√n*p*(1-p)
4. Calculate UCL and LCL
5. Copy-Paste Numbers in columns for CL, UCL
and LCL
6. Use Chart feature to generate Line Graph
 Example – np chart
NP Chart of Nonconforming Sponges
45

40

35

30

25

20

15

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32

Nconfrm CL UCL LCL

 np chart - Exercise
• File [Link]
 p Charts for Equal Subgroup Sizes
  𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑛𝑜𝑛𝑐𝑜𝑛𝑓𝑜𝑟𝑚𝑖𝑛𝑔 𝑖𝑡𝑒𝑚𝑠 𝑖𝑛 𝑎𝑙𝑙 𝑠𝑢𝑏𝑔𝑟𝑜𝑢𝑝𝑠
´𝑝 =
𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑖𝑡𝑒𝑚 𝑖𝑛 𝑎𝑙𝑙 𝑠𝑢𝑏𝑔𝑟𝑜𝑢𝑝𝑠
 =

 
UCL
  ´𝑝 (1 − ´𝑝 )
 
LCL
𝑠 𝑝=
√ 𝑛

p chart will be similar to the np chart but the points plotted represent the proportion
nonconforming in the samples and not the number nonconforming
 p Charts for Unequal Subgroup Sizes
• Two
  ways to handle:
• Use different limits for each subgroup
– Most sensitive statistically but requires recalculation of the upper and
lower control limits for each sample
– Approach used by Minitab
• Use , the average value of n, to compute the UCL and LCL
– Computationally simpler
– May result in erroneously accepting or rejecting the hypothesis that the
process is in control
p Charts for Unequal Subgroup Sizes
  ´𝑝 (1 − ´𝑝 )
𝑠𝑝=

 
𝑖
√
𝑛𝑖
𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑖𝑡𝑒𝑚𝑠 𝑢𝑛𝑑𝑒𝑟 𝑐𝑜𝑛𝑠𝑖𝑑𝑒𝑟𝑎𝑡𝑖𝑜𝑛
𝑛´ =
𝑇𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑠𝑢𝑏𝑔𝑟𝑜𝑢𝑝𝑠

 =

 
UCL

 
LCL
 Example – p chart
• File [Link]
• The data represents the number of trucks a recycling
facility and number of trucks overweight on each 20
consecutive business days.
• Generate a p-chart in Excel.
• Generate a p-chart in Minitab using both methods
 Sample Sizes for p Charts and np Charts
• Attribute measures are less sensitive than variable measures
• Samples should be large enough to have a 90% chance of
detecting at least one nonconforming unit in a sample
• Large enough that the average number of nonconformities in a set
of subgroups is at least 2.0
This limits the usefulness of attribute charts! If we improve quality then meeting the above criteria
would imply that we use larger samples – which is contrary to good practice. Why increase inspection if
quality is improving?
Attribute charts tend to have application where organizations are new to SPC techniques and quality is
“low”. However they also have application where the subgroup sizes are naturally large and the analysis
becomes meaningful.
 Area of Opportunity Charts- c and u chart
• Area of Opportunity – unit in which one or more events can
occur
• c chart – use when areas of opportunity are the same size
• u chart – use when dealing with areas of opportunities of varying
sizes
 c Charts

•Centerline
  =
 =

  UCL

  LCL
 Example – c chart
# #
Day Nonconforming Day Nonconforming
The adjacent table represents the 1 1 12 6
number of nonconformances per 2 1 13 9
1000 meters in telephone cable. 3 3 14 11
4 7 15 15
a. Calculate CL, UCL and LCL 5 8 16 8
manually. 6 10 17 3

b. Create a control chart for the data 7 5 18 6

in Minitab 8 13 19 7
9 0 20 4
c. Is the process in statistical control?
10 19 21 9
11 24 22 20
Example – c chart
 Example – c chart
• The process is not in statistical control; three subgroups exceed the UCL. Exclude
subgroups 10, 11 and 22, and re-calculate the control limits. Subgroup 15 will then be out of
control and should also be excluded.
 u Charts
  =

•Centerline
  =
 Example – u chart
A paper mill uses a control chart to monitor the imperfection in
finished rolls of paper. Production output is inspected for 20 days,
and the resulting data are shown. Use these data to set up a control
chart for nonconformities per roll of paper. Does the process appear
to be in statistical control? What center line and control limits would
you recommend for controlling current production?
 Day # Rolls Produced # Imperfections

Example – u chart 1
2
18
18
12
14
3 24 20
4 22 18
5 22 15
6 22 12
7 20 11
8 20 15
9 20 12
10 20 10
11 18 18
12 18 14
13 18 9
14 20 10
15 20 14
16 20 13
17 24 16
18 24 18
19 22 20
20 21 17
Example – u chart
 u Charts
• There is an inverse relationship between the size of the area of
opportunity and the distance between the upper and lower control
limits
– Large subgroups or areas of opportunity will result in narrow limits
– Small subgroups or areas of opportunity will result in more widely spaced limits
– We can deal with this by calculating the exact limits for each subgroup size
which is more statistically sensitive but computationally more complex - Minitab

 
UCL  
LCL
 Attribute vs Variable Control Chart
Type Advantage Disadvantage
Attribute • Several quality characteristics can be considered • Provides little information
jointly. about process performance.
• Relatively simple to implement • Out of control points only
• Does not require complicated gauges and resources after process has changed.
to maintain. • Require larger sample size
• Avoids expensive and time-consuming to detect shift
measurements.
Variable • More useful information about process performance. • Each quality characteristic is
• Directly obtain information about process mean and considered separately.
variation. • More expensive
• Out of control conditions signals potential special • More time-consuming
causes of variation. • Requires resource to
• Signal impending trouble before defectives are maintain measuring system
produced.
• Require smaller sample size to detect process shift.