Statistical Process Control and

Life Cycle of a Trend

Steven S Prevette
Fellow, CQE American Society for
Quality

-1-
Aim of this Session

Appreciation for use of numerical data

Overview of Statistical Process Control
Apply teachings from Dr. Shewhart, Dr. Deming, and
Dr. Ackof
Application of Trending Life Cycle of a Trend
Make better management decisions / assist in
better management decisions
Introduction to a SPC Colour Coded Dashboard

-2-
 My Introduction

Steve Prevette is a practitioner of Performance Measurement using

the Theory and Tools developed and promoted by Dr. Walter Shewhart
and Dr. W. Edwards Deming. He currently works as a statistician for
Fluor Corporation at the US DOE Savannah River Site and supports
three other Fluor nuclear sites, including Canadian Nuclear
Laboratories. He supports performance indicators in a broad
spectrum of subjects, including contractor assurance, safety, security,
environment, operations, maintenance, and quality. Steve and his
team provide customers with more than 4,700 charts and files per
month. Yes, Steve has a control chart of the number of charts made
each month.
Steve has published articles both with the American Society for
Quality, and the American Society of Safety Engineers. Lead to
Succeed for the American Society of Qualityappeared in the
September 2005 issue of Quality Progress and Charting Safety
Performance by ASSE in its Professional Safety magazine in May
2006. He has provided Dr. Demings Red Bead Experiment for more
than 2,500 persons, and leads a monthly training session in
performance indicators. He is a Fellow of the American Society for
Quality and an ASQ Certified Quality Engineer. Steve holds a B.S. in
Civil Engineering from Virginia Tech, and an M.S. degree in Operations
Research from the Naval Postgraduate School.

-3-
 Statistical Process
Control
A way of:
presenting data on a chart
determining if you have a
trend
determining if you are
stable
determining the capability
of your process
It is also a way of thinking -4-
What is a Trend?
Many procedures and policies call for
trending to be performed, for trends to
be identified
Websters Dictionary:
to extend in a general direction,
a general movement,
to veer in a new direction,
to show a tendency,
to become deflected
For our purposes: a changing condition

-5-
Canadian Nuclear
Laboratories
The Chalk River Site (near Pembroke ON)
has a long history of nuclear research and
development
In 2015, the site transitioned from a
Canadian Government site to a
Government Owned / Contractor Operated
site
Fluor Corporation assigned me to help
with performance metrics improvements
at CNL
A trending procedure that I had authored
was already in place at CNL, easing the
transition -6-
Lessons of the Red Bead
Experiment
Dr. Demings Red Bead
https://www.youtube.com/playl
Experiment illustrates the ist?list=PL8E522DD542C4CA69
importance of
understanding when we
have random noise.
Rewards, punishments,
ranking of the workers,
feedback to the workers
had no efect on the results
of the process.
The process was stable and Or Google the following:
needed to be changed! Red Beads Prevette

-7-
Importance of Trending
Actions taken to improve a process are diferent,
depending on whether or not the process is stable

Attempting to explain or correct for individual

datum point changes in a stable process will not
improve performance. This is a Type I error.
You will be making a mountain out of a molehill

Missing initial indication of a change (and missing

the opportunity to determine the cause of the
change). This is a Type II error.
You will allow the molehill to grow into a mountain

-8-
Management Theory:
The Theory of Variation
SPECIAL CAUSE VARIATION
If a statistically significant trend occurs, find the
special cause of this trend. Use this information to
correct or reinforce these special causes.

COMMON CAUSE VARIATION

If no trends exist, you must look at the long run
performance of the process and fundamentally
change the process in order to improve the process.

-9-
The Losses

1. Explaining results of a
random, stable process,
reacting to each individual
result
2. Missing the detection of a
significant change in the
data, and the underlying
process
-10-
Construction of the Control
Chart
Plot the actual data, and preferably at least 25 points
Calculate an average of the data over an initial baseline interval
Calculate the corresponding UCL/LCL set at 3 standard deviations from the
Average Line. UCL
16

14
Average Line Performance Data
12

10

4
LCL
2

23
1

11

13

15

17

19

21

25

27

29
Use a set of Rules to detect Trends

-11-
Generating and Using Control
Charts
This session is only intended as a broad introduction
to SPC
If you are in need of more detailed information, I wrote
a trending procedure for use at the U.S. Hanford Site
and it was posted to the Internet
You may find a copy of it by googling Generating and
Using Control Charts or HNF-4931
One location is
cci.drexel.edu/faculty/gbooker/Reference/technical/co
ntrol_cht.doc

-12-
 Definition of a Trend (at CNL)
1. One point more than 3 sigma from center line.
2. Nine points in a row on same side of the center line.
3. Six points in a row, all increasing or all decreasing.
4. Fourteen points in a row, alternating up or down.
5. Two of 3 points more than 2 sigma from the center line
(same side).
6. Four of five points more than 1 sigma from center line
(same side).
7. Fifteen points in a row within 1 sigma of center line
(either side).
8. Eight points in a row more than 1 sigma from center
line (either side).

From Minitab, Western Electric Rules -13-

Why Three Standard
Deviations ?
Many courses incorrectly teach that the
control limits cover 99.7% of the normal
distribution
Not all data are normal, real data can
cause the rate to be as low as 95% (Dr.
Wheeler)
The Tchebychev Inequality states up to
11% can be outside three standard
deviations
We use a suite of rules, and we want to
avoid too many false alarms
-14-
Why Three Standard
Deviations (2)
Dr. Shewhart established 3 standard deviations as
an economic balance between failure to detect
and false alarms.
If you dont believe this, go home and make your
smoke detector more sensitive. Is your house
now more safe?

-15-
Definition of a Trend

-16-
Establishing a Baseline
The goal is to establish one or more
baseline time intervals with no trends
within each interval.
The MW Rule helps to show if a baseline
is likely to be good.
A good baseline detects future trends
without false alarms
If a trend is detected, we dont want it to
be due to too few data points in the
baseline, causing the baseline to have
been inaccurate.
-17-
What Has SPC Done for
Nuclear Sites?
Minimized knee-jerk reactions to the latest
events
Detected changing conditions
Allowed focus upon the system, the
underlying processes
Saved time and money, both in the act of
chart making, and in management actions

SS Prevette 18 -18-
Applying SPC: Life Cycle of a
Trend 0.
Gestation:
Choose a
Measure
4.
Maturity:
Take Data 1. Establish Establish
Baseline
action to stabilizes Expectations,
correct, Routine
reinforce, Monitoring
or stabilize
3. Youth:
Determine 2. Birth:
special A trend is
cause(s) detected

-19-
 Similarity to PDSA Cycle

Act on large Plan a

Stable
scale or Baseline,
change or
Stabilize on
abandon improved test
Review
baseline Pareto,
Histograms
PDSA from The
New Economics,
Deming

Study results, Detect Do carry

what did we Improving out on small
Trend
learn scale

-20-
0. Gestation

Decide what to measure

Develop Operational Definition
Independent data, and do not cumulate
Determine data source(s)
Gather initial data, hopefully at least 25
data points of history

-21-
Gathering Data

Check existing reporting systems.

Many organizations are awash in data, but
the data have never been analyzed or put
to use.
There is an advantage to using existing
data, as it has already been paid for, and
historic data for establishing the baseline
should be available.

-22-
Choosing Reporting Intervals

If a trend developed, how long could you go

without needing to know it? Longer intervals
imply more risk
Need sufficient volume of points (25)
Costs increase as reporting interval
decreases
What is current reporting interval?

-23-
1. Establish the Baseline
Average and Control Limits
The Baseline on a control chart consists of
the average (center) line, a three-standard
deviation Upper Control Limit (UCL) and a
three-standard deviation Lower Control Limit
(LCL).
The Baseline allows us to predict the future,
and evaluate for trends.

-24-
Establishing a Baseline
The goal is to establish one or more baseline
time intervals with no trends within each
interval.
The MW Rule helps to show if a baseline is
likely to be good.
A good baseline detects future trends with
a minimum of false alarms.
If a trend is detected, we dont want it to be
due to too few data points in the baseline,
causing the baseline to have been
inaccurate.
-25-
MW Rule
I developed this rule to meet the needs of government oversight
personnel who wanted a formal rule for when to rebaseline a SPC chart
15 years of experience shows that it works >90% of the time, and for
when it does not work, the SPC process is self-correcting.

3 Changes of
Direction 2

3
1

-26-
Initial Strategy

Use the first 25 points for a trial baseline

If less than 25, use all points, assuming
MW rule is met
Evaluate for trends within this trial
baseline

NOTE: Do not change a baseline unless it is

proven guilty by a Trend

-27-
This is a Trend!
1. One point more than 3 sigma from center line.
2. Nine points in a row on same side of the center
line.
3. Six points in a row, all increasing or all
decreasing.
4. Fourteen points in a row, alternating up or down.
5. Two out of three points more than 2 sigma from
the center line (same side).
6. Four out of five points more than 1 sigma from
center line (same side).
7. Fifteen points in a row within 1 sigma of center
line (either side).
8. Eight points in a row more than 1 sigma from
center line (either side).
-28-
Remove any Trends from
the Trial Baseline
Do show all data, but change the average
and control limit calculations by:
Dropping data of of the beginning
Dropping data of of the end
Dropping individual datum point(s) and
circling them
Split into two or more baselines
Examples follow

-29-
Initial Baseline

In all cases, trends should be investigated

to determine the special cause(s)
A circle or a shift in the baseline is a trend
Generally, short term shifts are left
circled, long term shifts have a new
baseline
Remember the goal is Prediction of
Future Performance

-30-
Three data points removed from the beginning
of the data set, leaving a stable baseline from
the fourth point onwards:

-31-
Three data points removed from the end of the
data.

-32-
Two data points within the data have been
removed, and the remaining data used to
construct the baseline.

-33-
The data have been split into two baselines. A
permanent decrease in the level appears to
have occurred.

Both
baselines
satisfy the
MW Rule

-34-
Real Example
50

45
Average = 21
40 (May 04 - May 05)
35
Average = 13
30 (Sep 05 - Oct 06)
25

20

15

10

0
Jul-04

Jul-05

Jul-06
May-04

May-05

Nov-05
Sep-04

Nov-04

Jan-05

Mar-05

Sep-05

May-06

Sep-06
Jan-06

Mar-06 -35-
1a. Establish Expectations

Stable performance is not necessarily

good
Management needs to determine if the
current stable baseline is acceptable or
unacceptable
May use benchmarks, customer opinion,
risk analysis, or management philosophy
to make this decision

-36-
Monitoring

Update charts on the required time

interval
Check for trends against the trending
rules
Circle any trends, inform owning
management and look for special cause(s)
Do not shift a baseline unless there is a
trend (baseline proven guilty)
The following slides show an example
trends life cycle
-37-
Initial, Stable Baseline

Average = 9.5
Initial Stable Process
(Oct 98 - Oct 00)
20
15 25 points in a
10 stable baseline
5
0

Aug-00
Oct-98

Dec-98

Feb-99

Apr-99

Jun-99

Aug-99

Oct-99

Dec-99

Feb-00

Apr-00

Jun-00

Oct-00

Dec-00

Feb-01

Apr-01

Jun-01

Aug-01
-38-
2. Birth of the Trend

Average = 9.5
Initial Indication of Trend 2 of 3 at
(Oct 98 - Oct 00) 2 standard
20 deviations
15 above average,
10 circled

5
0
Oct-98

Feb-99

Apr-99

Aug-99

Feb-00

Aug-00

Oct-00

Apr-01

Aug-01
Dec-98

Jun-99

Oct-99

Dec-99

Apr-00

Jun-00

Dec-00

Feb-01

Jun-01
-39-
Initial Actions

Report the trend

Search for Special Causes
Compare Pareto chart of detailed data during the
trend to previous stable time interval
Consider initiating corrective actions and an
ImpAct (correction action management) if trend
is adverse
Even if Improving Trend, consider actions to
reinforce the improvement and spread the word
(OPEX / Lessons Learned)

-40-
3. Maturity

Waiting for
Average = 9.5
More Data Gathered 3 changes of
(Oct 98 - Oct 00) Direction
20 > 7 months in a row
above average
15
10
5
0
Feb-99

Apr-00

Feb-01
Oct-98

Dec-98

Apr-99

Jun-99

Aug-99

Oct-99

Dec-99

Feb-00

Jun-00

Aug-00

Oct-00

Dec-00

Apr-01

Jun-01

Aug-01
-41-
New Baseline Established

Average = 9.5
New Baseline Established Average = 13
(Dec 00 - Aug 01)
(Oct 98 - Oct 00)
20
15
10
5
0

Aug-00
Feb-99

Apr-99

Aug-99

Feb-00

Apr-00

Feb-01

Apr-01

Aug-01
Oct-98

Dec-98

Jun-99

Oct-99

Dec-99

Jun-00

Oct-00

Dec-00

Jun-01
Go back to Step 1 and determine if this baseline is acceptable or not.

-42-
Life Cycle of a Trend may be
used as the basis for
Dashboard Colour
The colour is not just setCoding
based upon
level, but also includes trends
- If you are currently at a good level,
but there is a trend in the wrong direction,
you will pick up a Red
- If you are currently at a bad level, but
are making progress, you get positive
reinforcement with a Green.

43 -43-
Control Decision Colour Leadership
Chart Result Action
Stable Level is Green Stay the
(common Acceptable Course
cause
Level is Not Yellow Improve
variation)
Acceptable System
Trend Adverse Red Corrective
(special Action
cause
Improving Green Reinforce
variation) Stay the Course

-44-
Use of White

Addition of White to the dashboard can be

advantageous:

Use White (or Yellow) for one month away

from a trend
Use for stable at an okay level, but not
superior
Minimizes some of the push to be All
Green while allowing for opportunities for
improvement

-45-
 July 2006

Indicator SW/GWVZ +
(with link to definition) FH Overall PFP K Basins FFTF WS&D WSCF CP D&D CS&I

LEADING INJURY Y W G W W G G Y
INDICATORS

First Aid Case Rate R R G W Y G G G

ORPS W W G G W W G R

Near Misses W W G G W G G W
No. Safety
Inspections G G G G Y G G G
Safety Inspection
Scores G G G G G G G G

HGET Survey
Safety Related
R W G Click
G to seeG R G R

Employee Concerns G W W First

G Aid W G G W

LAGGING INJURY Y W Y W ChartGat G W W

INDICATORS
PFP
OSHA Case Rate Y G R Y G G G Y

DAFW Case Rate Y W G G G G G W

DART Case Rate R W Y G G G W W

Severity Rate W G Y G G G W G

-46-
 First Aid Case Rate Occurrence Reports
50 6
45 CY = 11.1
40 FY = 9.8 5 Avg = 1.7
Does not (Nov 03 - Jul 04)
35 Average = 7.4
(Sep 05 - Mar 06 4 include
30
25 Nuclear Safety
3
20
15 2
10 1 ORPS Redesign
5 November 2003
0 0
Jan-03

Jan-05

Jan-06
Jul-03

Jan-04

Jul-04

Jul-05

Jul-06

Jan-03

Jul-03

Jul-04

Jul-05

Jan-06

Jul-06
Jan-04

Jan-05
Four months at one standard deviation above average, a significant Stable at an acceptable level.
increasing trend R W

Near Misses
5
ORPS Redesign November 2003
4

0
Avg = 0
Increasing
Trend at PFP,
Jan-03

Jul-03

Jan-04

Jul-04

Jan-05

Jul-05

Jan-06

Jul-06

One new report for July 2006, so there is no longer 12 months since the
Red Circled
last report. Chart set to White. W

-47-
 Fluor Hanford Dashboard
Safety and Health
Emergency Preparedness
Indicator SW/GWVZ +
(with link to definition) FH Overall PFP K Basins FFTF WS&D WSCF CP D&D CS&I

Drills Y W W G Y Not Applicable G G

Percent of Drills
Completed versus
Scheduled W W R G Y G G

Percent of Drills
Rated Satisfactory R G G G Y G G

Personnel Proficiency
Observations Y W W W W G G

Non-Improving Trend, Red = Unacceptable Level or Non-Improving Trend

Lets look
Green = Improving Trend or Superior Performance, White = Acceptable, Yellow = Stable, needs improvement or potential
Reporting Period - July 2006

at
CP D&D

-48-
 Drills Completed vs Scheduled Percent of Drills Satisfactory
900% 100%
800% 90% FY 06 Conducted = 8
700% FY 06 Scheduled to Date = 8 80% FY 06 Satisfactory = 8
600% FY 06 Completed = 8 70% Rate = 100%
500% Rate = 100% 60%
50%
400%
40%
300% 30% Average = 100%
200% Avg = 100%
20% (Oct 04 - Mar 05)
100% 10%
0% 0%
Oct-04

Jan-05

Apr-05

Jul-05

Oct-05

Jan-06

Apr-06

Jul-06

Oct-04

Jan-05

Apr-05

Jul-05

Oct-05

Jan-06

Jul-06
Apr-06
After a very unstable Oct 04 - Apr 05, the drill schedule has stabilized at Stable at 100%
100% G G

Personnel Proficiency
100%
90%
80%
70% CY 06 Observed to Date =
60% 100
50% CY 06 Proficient = 100
40%
Rate = 100%
30%
20%
10%
0%
Jan-05

Apr-05

Jan-06

Apr-06
Oct-04

Jul-05

Oct-05

Jul-06

Stable at 100%
G

-49-
Conclusion

Manage your numbers instead of your

numbers managing you!
Statistical Process Control is a tried and
true method for determining if something
is changing
If nothing is changing, should something
be changing?
We can use Colour Coded Scorecards for
management action and feedback

-50-