CONTROL CHART

“BASICS”

South Pacific TCE

May 2012

©2011 Hewlett-Packard Development Company, L.P.

The information contained herein is subject to change without notice
 Workshop Objective

• Provide an entry level review of control charts, covering:

– Their purpose and why we use them
– How to select the appropriate control chart
– How to create them using Minitab and the MS Excel SPC version
– How to interpret results of the control chart
– Why the results of the chart may not look right and what you can do to fix the issue

2 HP Confidential
What is a control chart and
why bother?
• Was created by Walter Shewhart (1891 – 1967), also known as I Chart of Cycle Time
the Shewhart chart 16
UCL=14.46

• A better title than ‘control chart’ would be ‘learning charts’ as 12

they help us to learn about our process and determine what

Individual Value
actions should take place 8
_
X=6.30

• They help us separate common cause variation and 4

special cause variation easily – this is their main 0

purpose LCL=-1.86

1 3 5 7 9 11 13 15

• Will appear similar to a run chart however with the addition of Observation

an upper and lower control limit, along with a centre line

• Great for viewing trends and showing before and after
performance, especially changes to the mean and variation
• Unlike run charts, there are different types of control charts
depending on what type of data you have

• Helps us determine if our changes

are resulting in improvements!

3 HP Confidential
Control Charts at a glance

ADVANTAGES DISADVANTAGES
• The best tool for identifying special and common • Needs additional calculation of upper and lower control
limits if Minitab or MS Excel spreadsheet is not being used
cause variation, along with process stability for the – this can take some time
past 80+ years
• Reasonable amount of coaching is required to determine
• By separating special and common cause variation, appropriate control chart to use, and then how to interpret
they can reduce tampering to a system which often results appropriately
leads to greater variation, not less
• Sometimes a run chart will suffice
• Can be used to view performance over time as well • Appropriate subgrouping strategies need to be considered
as on a subgroup basis (comparing people, when creating the chart
products, regions etc.)
• Does not contain specification limits

• Many myths about control charts exist, with some more

prominent ones being you need ‘normal’ data to start with
– these myths create unnecessary confusion

4 HP Confidential
What will be covered in this
presentation?
HP Lean Six Sigma
o Where the calculations for control limits come from
o The decision tree for selecting the right control chart
o Attribute & Continuous data and why it’s important
o Practical application of I/X charts
o Practical application of P charts
o Creation of these charts using both Minitab and MS Excel SPC workbook
o Brief review of ‘special’ cause vs ‘common’ cause variation and why it’s important to know
the difference
o The rules for identifying special causes
o When to revise control limits

5 HP Confidential
Anyone for math?
Shewhart called the control limits "three-sigma" control limits and gave a general
formula to calculate the limits for any statistic. I Chart of Cycle Time
16
UCL=14.46

Let S be the statistic to be charted, then 12

Individual Value
the centerline: CL = us
8
_
X=6.30

the upper control limit: UL = us + 3 * σs 0

the lower control limit: LL = us - 3 * σs

LCL=-1.86

1 3 5 7 9 11 13 15
Observation

σ is often referred to as standard deviation incorrectly- see below

Type of Control Chart Definition of σ Explanation

Individual Chart (X chart) σ – MRbar / 1.128 Average of Moving Ranges divided by bias correction factor

Xbar and R chart σ - Rbar / d2 Average of within subgroup ranges divided by bias correction factor

Xbar and S Chart Weighted Average of within subgroup standard deviations divided by
σ - Sbar/ c4
bias correction factor Aren’t you glad we’ve
P chart ___________ From binomial distribution: square root of center line = average percent got tools to calculate
σ-
√ pbar*(100–pbar)
_____________
defective
these for us 
√n

C chart ____ From Poisson distribution: square root of center line – average number
σ - √ cbar of defects

U chart ___ __ From Poisson distribution: square root of center line – average rate
σ - √ ubar ) ∕ √ n

6 HP Confidential
The Decision Tree of Control Charts
Choosing the right control chart >> revised
decision tree TYPE OF DATA

COUNT OR CLASSIFICATION CONTINUOUS

(attribute date) (variable data)
Number of YES/NO
accidents per month, Ontime/late
number of scratches delivery, ok the first
on a surface
COUNT CLASSIFICATION time/rework,
ok/excessive Number of minutes
scratches early or late,
length in cm of
If large each scratch
expected) e.g. # complaints/month in

Incidences or Defectives or
subgroup size
nonconformities non conforming units
If rare data (centralised and

>> 1000
complaints centre

Fixed Variable Subgroup Fixed Variable

Fixed Variable
subgroup subgroup size subgroup subgroup
opportunity opportunity
size size of 1 size size
When we
have a large
variation
within the
range go to
X-bar and X-bar and an XbarS
C Chart U Chart NP Chart P Chart X Chart
R chart S chart chart

Average &
Number of Incidences Number of Percent Individual Average &
standard
incidences per unit defectives defective measurement range
deviation

8 HP Confidential
Two basic types of data
Noting the presence or absence of some characteristic or attribute in each
of the units in the group under consideration, and counting how many units
do (or do not) possess the quality attribute, or how many such events occur
Attribute in the unit, group, or area. Essentially, a category measure.
Data

Measuring and recording the numerical magnitude of a quality

characteristic for each of the units in the group under consideration. Data
Continuous in it’s current state not belonging to a classification or category measure.
Data
(aka. variables
data)

Note: The type of data determines the type of control chart used to record and
analyze the data

9 HP Confidential
Two Types of Attribute Data

10
Types of Data
Classification-Count-Continuous

Data Type Quality characteristic Recorded data

Classification Delivery Performance On-time/late delivery

Rework OK the first time/rework
Scratches OK/excessive scratches

Count Changes Number of changes/design

Accidents Number of accidents/month
Scratches Scratches/surface
Defects

Continuous Time Minutes early or late

Weight Grams using a laboratory scale
Scratches Length in cm of each scratch
 Lightning Strikes
•Number of strikes in city limits Count
•Percent of buildings hit per month Classification
•Current (amps) per bolt of lightning Continuous

12
Labor Strikes
•Number of labor strikes in Paris Count Or Continuous?
•Percent of people striking Classification
•Number of days on strike Continuous

13
 Rational Sub-grouping Strategy
• Organize data within subgroups in a way that is likely to give the greatest chance for the
data in each subgroup to be alike and greatest chance for data between subgroups to be
different
• The aim is to include only common causes of variation within a subgroup with special
causes of variation occurring between subgroups
• Usually best to see data over time

If you have a batch processing process that results in the same performance for each process in a batch,
do not treat this as a subgroup – failure to do this will result in misleading control limits as there is no
variation within the subgroup. Treat each batch as one measurement/data point

Learning Objective:
– “Obviously, the ultimate object is not only to detect trouble but also to find it, and such discovery naturally
involves classification. The engineer who is successful in dividing his data into rational subgroups based
on upon rational hypotheses is therefore inherently better off in the long run than the one who is not thus
successful” Dr. Shewhart, 1931
Types of Control Charts – The X Chart

X Chart
Control Chart for Individual
Measurements
• The control chart for individual data is useful when:
– There is no rational way to organize the data into
subgroups
– Measures of performance of the process can only be
obtained infrequently
– The variation at any one time (within a subgroup) is
insignificant relative to the between subgroup variation

– It is the Swiss army knife of control charts! It is often used as

the default chart when results for other control charts appear
misleading
Individual Control Charts

ADVANTAGES DISADVANTAGES
• An extension of the familiar run chart • Since there is only one measurement in each
“subgroup,” there is no opportunity to focus on
• No calculations are required when plotting different sources of variation through
a data point subgrouping

• Plotting is done each time a measurement is • Large amounts of data (100 + data points) will
made, providing fast feedback produce a chart that’s hard to analyse

• Capability can be evaluated from the

control limits

17 HP Confidential
Practical application of an
Individual Control Chart

Scenario
• GB/BB needs to identify special and common cause variation for:

A. TAT to return a spare part to HP

B. The amount of time it takes to spend creating and or approving a quote

C. Forecast accuracy dollars difference

D. CPRD over the past 2 years

18 HP Confidential
Example: X Chart

A process change took place in November 10 – special cause variation has taken place giving us a high degree of
belief that our change has resulted in an improvement!

19 HP Confidential
Continuous Subgroup Control Charts
X-bar and R/S Charts

ADVANTAGES DISADVANTAGES
• Where volumes are high, an X chart is not • Large volumes within a subgroup can cause very
as beneficial as a X-bar R or X-bar S narrow limits and in these cases an X chart must
be employed instead
• The common approach at HP is to group
data in weeks, months, quarters etc.
anyway

• Variation from one subgroup to the next is

easily analyzed, be it range, variation or
standard deviation

20 HP Confidential
Examples:
X-bar and R/S Control Charts

Scenario
Grouping any of the continuous types of data based on:

• Hour

• Day

• Week

• Month

• Batch deployment

21 HP Confidential
Example: X bar S Chart

The X bar S puts us in a good position to do analysis on trends – in this case, it’s TAT from 1 week to
the next. We can see some special cause variation which may lead to a new or increased
understanding of the system/process

22 HP Confidential
 Types of Control Charts – The P Chart

25-15
Attribute Control Charts (C, U, NP & P)

ADVANTAGES DISADVANTAGES
• Can relate to one of the more common • Calculations of control limits are more complex if
metrics used within HP: a % that Minitab or MS Excel SPC not being used
meets/does not meet a set target, therefore
• As subgroup size increases the variation
data is normally readily available
between values generally decreases which can
result in very narrow control limits. For
subgroup sizes of 1000+ an X chart should be
substituted for a P chart

• Control limits may appear ‘unreasonable’ in

achieving ‘stable’ performance if the % defects
are too high compared to the subgroup size

• The default of an X chart needs to be used when

control limits are too narrow

24 HP Confidential
Types of Attribute Control Charts

Type of
Statistic Subgroup
Chart Name Attribute
Charted size
Data
Number of
NP Chart Classification nonconforming Constant
units

% of
P Chart Classification nonconforming May vary
units

Number of
C Chart Count Constant
incidences

Incidences per
U Chart Count May vary
unit

25
Practical application of a
Attribute Control Chart

Scenario
• GB/BB needs to identify special and common cause variation for:

A. % of renewals meeting the set TAT = P / NP chart

B. % of products returned within 5 days = P / NP chart

C. Support calls logged = C chart

D. # information missing per approval document = U chart

26 HP Confidential
Example:
Attribute Control Charts (C, U, NP & P)

A process change took place in August – the special cause variation has given us a high
degree of belief that our change has resulted in an improvement!

27 HP Confidential
Control Charts

To create in the SPC MS Excel sheet…..

Determine which chart is required>Choose appropriate control
chart worksheet>Enter data into cells>Reset ‘Start & End Row
data’>Determine your calculation of control limits using ‘Recalc’
column>If data points need to be excluded enter in ‘1’ into ‘Exclude’ column

The SPC workbook can be found here:

http://teams1.sharepoint.hp.com/teams/SPBITraining/Tools/Forms/AllItems.asp
x?RootFolder=%2fteams%2fSPBITraining%2fTools%2fControl%20Charts&Folder
CTID=&View=%7bC98D926A%2d6E49%2d4D4A%2d9F87%2d7886960B556
6%7d

Do not save over the version on the share point 

28 HP Confidential
Control Charts

To create in Minitab…..
Simple: Determine which chart is required using Minitab Assistant
Control charts or other methods>Stat>Control Chart

Separate control chart into different stages of project: Same as

above>Chart Options>Stages>In ‘define stages’ enter in the column that
captures the different stages

Minitab homepage with connection instructions found here:

http://enhanced1.sharepoint.hp.com/teams/edw_bi/GBIorgs/PlatformArchitect
ure/minitab/default.aspx
Minitab defaults to select only special cause rule #1 = any data point outside the control limits. To change
this go to Options>Tests. We recommend you use the rules listed in this slide deck

Please note that Minitab requires data to be sorted in order of subgroups if you’re
using a subgroup based control chart

29 HP Confidential
Common vs Special Causes

Common Special
• Those causes inherent in the process over • Those causes not part of the process all the time
time, affect everyone working in the process, or do not affect everyone, but arise because of
and affect all outcomes of the process specific circumstances

• Chance cause • Assignable cause

• Stable process – no matter how great • Unstable process

the variation
• Process not in statistical control
• Process in statistical control
• Can be good or bad – positive special cause
variation when detected is often our best trigger
that our change has resulted in an improvement

30 HP Confidential
How to get started

Select a process measure or

quality characteristic

Under 12
data points?
Use a run chart

Develop appropriate
control chart

Convert to rational Narrow control Convert to X

subgroup limits? chart

Is process
Stable? Common
Causes only?

31 HP Confidential
Using the concepts of variation to guide
improvement
Matching Action to Variation
Select a process measure or
quality characteristic Type of Action
Look for What Take action Study all the Make changes
was different based on the data to the process
between reported
individual difference
points

Develop appropriate Waste time Increase Gain a better Reduce

variation understanding variation
control chart

Common Cause
of the system

Type of Variation
   
Gain useful Reduce Lose time in Loss of
information variation responding to productivity,
the problem may increase
Special Cause

variation

Is process
stable
   
32 HP Confidential
Balancing the Mistakes Made in Attempts to
Improve
ACTUAL SITUATION

When NO Special When a Special Cause is

Cause is occurring in occurring in System
ACTION System
Take action on individual -$
outcome (treat special) +$
(Mistake #1)
Treat outcome as part of -$
system; work on changing +$
the system (treat common) (Mistake #2)

The above table explains the foundation for the value of the control chart – it states that
if we action and address each individual data point in isolation however it is part of common
cause variation we are very likely to increase our performance variation which can result in
unnecessary costs to the organization. The same result can occur if we treat special causes with
a common cause, system based solution.

Studies show that the vast majority of variation is due to common cause variation, leading us to
understand that systemic changes rather than specific, case based solutions are more likely to
help us achieve our targets.
Evaluating processes
Stable Unstable
Process Process

No special causes exist and

Capable Special causes exist, but the
the process meets or exceeds
process meets customer
customer expectations
Acceptable to the customer specifications*

Ideal State

Incapable
No special causes exist, but Special causes exist and the
the process does not meet process does not meet
Unacceptable
customer expectations customer specifications
to the customer

* Because the process is unstable, there is no guarantee that it

will be acceptable in the future.

34 HP Confidential
Rules for Determining a Special Cause

Minitab allows you to select the rules you want to test for which will be identified with a red mark. In ‘I
Chart Options’ go to the ‘Test’ tab and tick the appropriate rules.

Different industries can apply different special cause rules – the above rules are universally accepted at
HP.
Guidelines for Applying the Special Cause
Rules
A point exactly on a limit is not considered outside the limit (rule #1)

A point exactly on the centerline does not cancel or count towards a run (the run in rule # 2 is also commonly
called a shift)

Ties between two consecutive points do not cancel or add to a trend (rule #3)

When control charts have varying limits (due to varying numbers of measurements within subgroups) rule #3 is
not appropriate

When there is not a lower or upper limit on one side of the center line (for example, on a range chart with less
than seven measures in a subgroup or on a P-chart with 100% as a possible result for the process), rules #1 and
#4 do not apply to the side missing the limit
When do we revise control limits?

Revise control limits when

Improvements The Shewhart The initial control
have be made to chart remains chart has special
the process and unstable for 20 or causes and there
“Trial” limits have
the more subgroups is a desire to use
been calculated
improvements and approaches the calculated
with fewer than
result in to identify and limits for analysis
20 subgroups
special causes remove the special of data to be
on the Shewhart causes have collected in the
chart exhausted future
Presentation Material Credits

Some material in this presentation has been sourced from…

• The Improvement Guide

• Documentation supplied by API: http://www.apiweb.org/API_home_page.htm

• Rogue wave forum share point:

http://teams5.sharepoint.hp.com/teams/tsgBB/Shared%20Documents/Forms/AllItems.
aspx?FolderCTID=&RootFolder=%2fteams%2ftsgBB%2fShared%20Documents%2fTopic
%20Presentations&SortField=Modified&SortDir=Desc&View=%7b4CE541FF%2d1FCD%
2d4CAF%2d9611%2d1330FF037F1F%7d

38 HP Confidential
QUESTIONS?

©2011 Hewlett-Packard Development Company, L.P.

The information contained herein is subject to change without notice
Example: What’s happened here?
Individual Control Chart

This doesn’t look right – we have ‘grouped’ data points and very narrow control limits which means little analysis can be
gained from the chart. The results suggest 2 things:
a. We have a batch process
b. There is either no or very little variation within the batches

The solution here is to either:

a. Use a subgroup if there is variation in the subgroup (X bar R/S)
b. Use an individuals X chart

40 HP Confidential
Example: What’s happened here?
Individual Control Chart Solution

Summarising the batch into 1 data point and using an individuals X chart enables much easier analysis of the process to
look for special and common cause variation.

41 HP Confidential
Example: What’s happened here?
Individual Control Chart

There’s too many observations here to make any analysis worthwhile. In cases where data can be grouped based on
time, use a subgrouping chart.

The control chart is still valuable as we can determine how many points sit outside the 3 sigma limits – this can be
valuable information when it comes to analysis

42 HP Confidential
Example: What’s happened here?
Individual Control Chart solution
Xbar Chart of E2E TAT
20

15
Sample Mean Hours

10 UCL=9.93
_
_
X=7.51

5 LCL=5.10

0
1 3 5 7 9 11 13 15 17 19
Week #'s starting from September
Tests performed with unequal sample sizes

Changing the subgroups to each week makes turns the control chart into a valuable analysis tool.

Here we’re trying to see whether there’s any special cause variation from week to week, an assumption that we had
going in. We do see special cause variation at week 8 therefore time can be invested into determining why the average
TAT’s were much worse that week. Assigning the special cause then increases our understanding of the system.

43 HP Confidential
Example: What’s happened here?
Subgroup Control Chart
Xbar-S Chart of Workday Tat
1

12 1

Sample Mean
10
1
8 1 1
1 1
1 _
UCL=6.86
_
6 X=6.02
LCL=5.17
1 1 1
4 1 1
1 1
1 3 5 7 9 11 13 15 17 19 21 23
Sample

10 1 1

1 1
8
Sample StDev

1
6 1 1 _
UCL=5.762
S=5.163
1 1 LCL=4.565
4 1 1 1
1 1 1 1 1 1
1
2 1
1 3 5 7 9 11 13 15 17 19 21 23
Sample

Tests performed with unequal sample sizes

Very narrow control limits make the chart virtually useless for analysis. Seeing as we’re already working with subgroups,
the most likely reason for these narrow control limits is a very large sample size – in these cases the best option is to
change the chart to an individuals chart.

Remember to refer to your revised control chart decision tree – when the subgroup size reaches 1000 the best option is
an X chart.

44 HP Confidential
Example: What’s happened here?
Subgroup Control Chart solution
I Chart of Mean by Change
Before After
14
1

12 1
Individual Value

10

6 UCL=5.95
_
X=4.40
4
LCL=2.85
2
09 09 10 1 0 10 10 1 0 10 10 10 1 0 10 10 1 0 11 11 11 1 1 11 11 1 1 11 11
o v- e c- an- b- a r- pr- ay - n- ul- g- p- ct- o v- e c - an- b- ar- pr- ay- n- ul- g- p-
N D J Fe M A M Ju J Au Se O N D J Fe M A M Ju J Au Se
Month

By using an X Chart instead we were in a much better position to assess the special cause variation due to a process
change, and increase our degree of belief a change has taken place.

45 HP Confidential