Vous êtes sur la page 1sur 26

Winter 2003 The Standard Page 1

The Standard
Vol. 17, Issue 1 The Newsletter of the Measurement Quality Division, American Society for Quality Winter 2003

Chair’s Column
My goal is to formulate a long-term strategy and
vision for the Measurement Quality Division where its
membership can grow. We still need to provide the
value to our core membership. However, there are
other professions besides metrology that use mea-
surements in their daily lives that can benefit from the
MQD. When considering that, we can include every-
one in some way or another. In a Division leadership
role, my goal shall be to explore ways and means to
attract new members by publicizing our efforts and
activities that can provide value to other individuals.
I look forward to this new role and to cultivate new
relationships in the measurement community.
I have a lot to learn in this new role. I shall start out
by listening to others and interacting with them. Please
feel free to contact me.

Message from the Chair-Elect (2002-2003), Sincerely,


Dilip Shah Dilip Shah (ASQ, CQE)

I am excited to be working with the MQD division in E = mc3 Solutions


a leadership role. I have been an MQD division mem-
197 Great Oaks Trail #130
ber since 1992 and an ASQ member since 1984.
Wadsworth, OH 44281-8215
Allow me to introduce myself:
Tel: 330-328-4400
I have over 25 years experience in electronics,
instrumentation, metrology, measurement and com- Fax: 330-336-3974
puter applications of statistics in the quality assurance E-mail: emc3solu@aol.com
areas. I have been employed in various positions with
Ham Call sign: KC8EEX
Phillips Electronics (UK), Kodak Ltd. (UK), Instruments
Division of Monsanto Corporation, Flexsys America
and Alpha Technologies. I am currently a Principal of E
= mc3 Solutions, a consulting practice that provides
training and other solutions in ISO9000/QS9000,
ISO17025, measurement and computer applications. IN THIS ISSUE
I volunteer my time with the local Akron-Canton (Ohio)
ASQ section as junior past section chairman (2001- Editor’s Column ................................................. 3
2002). I have been a member of the advisory board of CCT Update ........................................................ 4
the University of Akron Engineering and Science Tech- CCT Exam Dates ................................................ 5
nology Division since 1988. In addition to the MQD, I Standards News ................................................. 6
also belong to the Statistics Division of American The Learning Curve ........................................... 7
Society for Quality. I have participated actively in the Upcoming Events ............................................... 7
measurement-related issues through NCSLI (National Phil Stein’s Latest Paper ................................... 9
Conference of Standards Laboratories International) MQD Officers .................................................... 24
and the West Coast-based Measurement Science Regional Councilors ........................................ 25
Conference by presenting papers and tutorials on
measurement-related issues.
Winter 2003 The Standard Page 2

The Standard
Vol. 17, Issue 1 The Newsletter of the Measurement Quality Division, American Society for Quality Winter 2003

Publication Advertising Publication Information


Staff
2003 Rates for a single publication: The STANDARD is published quarterly
by the Measurement Quality Division of
Executive Editor Business card size ..................... $20
ASQ; deadlines are March 15, June 15,
1/4 page ..................................... $35 September 15 and January 15.
Frank Voehl 1/2 page ..................................... $70
Harrington Group Full page .................................. $120 Text information intended for
publication can be sent via electronic
11501 Lake Underhill Road A 15% discount will be applied for multi- mail or through postal mail on 3 1/2"
Orlando, FL 32825 edition ads. diskette in Microsoft Word saved in
Tel: 1-800-ISO 9000 Rich Text Format (RTF). If it is not
Ads must be formatted in MS Word or
Fax: feasible to send text in electronic form,
as a TIF file.
E-mail: FVoehl@aol.com clean printed text can be submitted.
Advertising must be clearly identified as
an ad and should relate to the field of Graphics or illustratios must be sent in
measurement quality. a TIF file format.
Managing Editor
Ads must not imply endorsement by the Photographs of MQD activities are
Mark Schoenlein Measurement Quality Division or ASQ. always appreciated.
P.O. Box 206 Publication of articles, product releases,
Perrysburg, OH 43552 advertisements or technical information
Tel: 419-247-7285 does not imply endorsement by The
Fax: 419-247-8770 Letters to the Editor Measurement Quality Division of ASQ.
E-mail: mark.schoenlein@ While The STANDARD makes every
us.o-i.com The STANDARD welcomes letters effort to ensure the accuracy of articles,
from members and subscribers. Letters the publication disclaims responsibility
should clearly state whether the author for statements of fact or opinion made
is expressing opinion or presenting by the authors or other contributors.
facts with supporting information. Material from The STANDARD may not
Commendation, encouragement, be reproduced without permission of
constructive critique, suggestions, and ASQ. Copyrights in the United States
alternative approaches are accepted. and all other countries are reserved.
If the content is more than 200 words, © 2003 ASQ, MQD. All rights reserved.
we may delete portions to hold that
limit. We reserve the right to edit letters
and papers.
Website Information
The Measurement Quality Division
homepage can be found on the internet
at www.measurementquality.org. Past
editions of the STANDARD in PDF
format are available there.
Winter 2003 The Standard Page 3

Exec Editor’s Column

The Global Race is Still On


by Frank Voehl

As this edition goes to press, the Japanese are a new way. In other words, to achieve efffective
rapidly redefining their organizations; and European levels of innovative practices, an organization must
firms are moving perhaps less rapidly but with increas- improve its creativity in its work groups and individu-
ing momentum to become more competitive by cutting als as well as in its measurement labs. And it must
costs through restructuring, reengineering, and knowl- foster the right kind of organizational culture that will
edge management interventions. Additionally, there encourage creativity and turn it into innovation.
are the burgeoning competitors from Pacific rim na-
The articles in this edition describe the character-
tions, Eastern Europe, and Latin America.
istics that an organization’s culture needs to possess
The key question is: How does a firm obtain a to achieve strategic competitive advantage through
global competitive advantage in the face of such stiff innovation. As Peter Drucker has said over and over
competition? Quality management alone no longer again: Every organization--not just business--needs
serves this purpose, as a successful continuous im- one core competence: innovation. And every orga-
provement program merely keeps you in the race. It nization needs a way to record and appraise its
doesn’t help you leapfrog the competition because innovative performance.
savvy competitors have programs of their own and are
not just standing still. Furthermore, these savvy
competitors are using speed strategies, reengineering,
and flexible manufacturing; and many have begun
using their own metrology programs. So what’s the
next source of competitive advantage going to be?
Innovation!
Innovation is the only sustainable competitive ad-
vantage to any situation. It is what enables an orga-
nization to create its products and services and to
differentiate them from those of its competitors. Inno-
vation is also where ideas come from that enable an
organization to cut its costs. Innovation is the focus of
this edition, especially in terms of how organizations
use innovation to achieve competitive advantage in
the global marketplace of tomorrow.
However, global competition is not the only chal-
lenge facing businesses and their staff as they enter
the Year 2003. Change is occurring at an accelerating
rate, and new technology is being introduced at break-
neck speed. The workforce is becoming more and
more diverse. There is a growing scarcity of highly
skilled workers, and we are smack dab in the middle
of a transformation from an industrial- to a knowledge-
based society. Constituencies are becoming more
demanding, and the entire business environment is
becoming more extremely complex by the moment.
To meet these strategic challenges and take ad-
vantage of opportunities they create, businesses need
to embrace creative problem solving and innovation in
Winter 2003 The Standard Page 4

Accreditation Board for Engineering and Technology


CCT Program Update •Registration as a Professional Engineer
•The signatures of two persons—ASQ members,
members of an international affiliate society, or mem-
bers of another recognized professional society—veri-
by Christopher L. Grachanen fying that you are a qualified practitioner of the quality
sciences
ASQ’s Certified Calibration Technician (CCT) pro- CCT exam question development began at the CCT
gram has been proceeding on its carefully laid out plan Item Writing (IW) workshop (ASQ refers to exam ques-
like a finely crafted WWVB synchronized timepiece. To tions as items) conducted at ASQ headquarters in
say a lot has happened with the CCT program since my Milwaukee, Wisc., on October 5-6, 2002. This work-
last update would be an understatement. First and shop attended by 24 dedicated metrology profession-
foremost, the long awaited CCT Body of Knowledge als used the CCT BOK as the basis for developing
(BOK) has been finalized and published. Readers may multiple choice exam items. Exam items are required
remember that the BOK was developed from a series to be referenced to publications available in the public
of workshops, phone interviews, mail surveys and domain. The IW workshop yielded over 300 exam
many long hours of careful evaluation and analysis. questions each with a reference, answer key and
The CCT BOK reflects the participation of hundreds of written explanation for each multiple choice item.
calibration professionals from industry, academia and
government agencies. The BOK is divided into the
following six major topics, each with their own subtop-
ics and associated explanatory subtext:
I. General Metrology (30 Questions)
II. Measurement Systems (25 Questions)
III. Calibration Systems (25 Questions)
IV. Applied Mathematics and Statistics (20 Questions)
V. Quality Systems and Standards (15 Questions)
VI. Uncertainty (10 Questions)
Note: The CCT exam has a maximum time limit of four
hours to complete the 125 questions.
Training / Education / Experience eligibility require-
ments for taking the CCT exam is a minimum five years
of on-the-job experience in one or more of the areas of Item Writing Workshop
the CCT BOK. If you have completed a degree from a ASQ Headquarters – October 5-6, 2002
college, university, or technical school with accredita-
Back Row: Jun Bautista, Jr., Jim Smith, Dave Stone,
tion accepted by ASQ, part of the five-year experience
Steve Ludwig, Sharry Masarek, Michael Berry,
requirement will be waived, as follows (only one of
Graeme Payne (background), Mary Anderson, Jeff
these waivers may be claimed):
Osborne, Joe Persons(background), Steve Niesen
•Diploma from a technical, military, or trade (foreground), Mike Boyd, Bob Bohrer, Jessie Colvin,
school—two years waived Herb O’Neill (background), Dave Nebel, Dave
•Associate degree—two years waived Brown, Raul Padilla
•Bachelor's degree—two years waived Front Row: Dave Wirtz, Chris Grachanen, Hershal
•Master's or doctorate—two years waived Brewer, Dilip Shah, Al Germann, Jim Noonan
ASQ requires proof of professionalism before one
can take the CCT exam. Proof of professionalism may IW workshop questions were the focus of the Item
be demonstrated in one of three ways: Review (IR) workshop attended by 14 metrology pro-
•Membership in ASQ, an international affiliate soci- fessionals on December 13-14, 2002, at ASQ head-
ety of ASQ, or another society that is a member of the quarters. IR workshop participants carefully analyzed
American Association of Engineering Societies or the each item carefully for applicability, readability and
technical correctness / reference publication linkage.
Winter 2003 The Standard Page 5

The fruits of these efforts yield the exam questions that and encourages ASQ in this endeavor. We will assist
will be used by the Exam Review workshop scheduled wherever we can.’ Our ASQ hats are off to NCSLI for
for April 4-5, 2003, at ASQ Headquarters. their vote of confidence and support for the CCT
program.

CCT Exam News

from Mary Martin


ASQ Certification Administrator

The CCT Exam will be offered during the June and


December exam cycle. Important upcoming dates:
Exam Date Application Deadline
June 7, 2003 (pilot) April 4, 2003
Item Review Workshop December 6, 2003 October 3, 2003
ASQ Headquarters – December 13-14, 2002
For more info visit www.asq.org and click on the
Background: Karl Haynes, Dave Nebel, Jack
“Certification” tab or call ASQ at 1-800-248-1946 and
Lucernoni, Wayne Paupst, Karl Wigdal, Rick
request item B1331.
Roberson
Foreground: Jay Bucher, Mary Anderson, Dilip
Shah, Carlis Stuber, Walter Nowocin

As previously mentioned, each CCT exam item


must reference a publication that is available in the
Certified Calibration Technician
public domain. A listing of these references has been (CCT) Test Prep
published on the ASQ website and are listed within Computer Based Training Module... content,
each of the CCT major exam topics. This reference terminology, concepts, graphics, formulas,
publication listing, as well as the CCT BOK, is not real world examples, and practice questions
intended to limit the subject matter or be all-inclusive of all designed to better prepare a technician
what might be covered in an exam but rather is meant for the CCT exam. All in one source!
to clarify the type of content to be included in the exam.
Available now!
Information about the CCT program, its BOK and
reference publication listing, exam prerequisites, exam Electrical, Dimensional, Temperature, Pres-
dates and the latest news updates may be found at: sure, Torque, Flow, Time & Frequency,
Measurement Uncertainty, 17025 Compliance.
http://www.asq.org/cert/types/cct/index.html
Each module comes with complete testing
http://www.measurementquality.org/cct/fs_cct.html
and documentation of training.
A high water mark for the CCT program occurred
See www.wptraining.com for a free online
during the Fall 2002 National Conference of Standards
demo and content outlines
Laboratories International (NCSLI) board of director’s
meeting. At the meeting it was decided to draft a letter WorkPlace Training, Inc.
of support for the CCT program with a strong recom-
mendation for some form of “Hands-On” proficiency Wayzata, MN 55391 USA
testing. Charlie Motzko, president of NCSLI, reiterated 1-800-472-2564
that ‘the board supports the goal for the CCT program
Winter 2003 The Standard Page 6

Standards News Managing Editor’s Note

by Bill McCullough

ANSI/ASC M1 “American National Standard for


Calibration System” is up for review. Bill McCullough
is convening a committee to consider the fate of the
document. Other than clean up and reissue, or
removal, there is a new option. Under a new proce-
dure ANSI allows the creation of technical reports. In
ANSI’s words:
“Such a technical report shall be entirely informa-
tive in nature and shall not contain information imply-
ing that it is a standard. It shall clearly explain its
relationship to aspects of the subject that are, or will
be, dealt with in related American National Stan-
dards. Nothing in these procedures precludes a by Mark Schoenlein
standards developer from developing, approving,
and disseminating its own reports or other publica-
tions.” After some brief dowtime the newsletter is back up
and running. And we are trying to reestablish several
We feel that there is a great deal of good technical
regular columns and features including articles from
information in M1 and elsewhere that would make a
our membership.
good technical document. We are actively looking for
that information from those of you in ASQ and other We are also working with our new incoming chair-
organizations such as NCSL. If you would like to man, Dilip, to rebuild the organization (lots of white
contribute to such an effort, have an opinion, or just space on the org. chart) and programs which were
want to make a suggestion, contact Bill McCullough last scheduled on 9/11 day.
at: bill.mccullough@bently.com or call (775) 215- I would like to encourage anyone who might be
1207. interested in providing assistance to the division to
A brief note about ISO/FDIS 10012:2002. The contact Dilip.
final draft is currently being circulated to the ISO/
TC176 member bodies for ballot from December 19,
2002, until February 19, 2003.

Important Upcoming Events

Conference Dates Location Contact Number Website

Quality Expo April 15-17, 2003 Rosemont, IL 888-267-3794 www.quality.reedexpo.com

Dimensional May 12-15, 2003 Nashville, TN Fax 865-574-4261 pritchardew@y12.doe.gov


Workshop

AQC 2003 May 19-21, 2003 Kansas City, MO 800-248-1946 www.asq.org

NCSL August 17-21, 2003 Tampa. FL 303-440-3339 www.ncsli.org


Winter 2003 The Standard Page 7

Since there is nothing more I can report at this time


on my charter subject, and there are no current reader
inquiries to pontificate on, let’s move on to a divergent
The Learning Curve topic. Over the past 30 years I have often been queried
concerning my opinions of various proposed policies
for the certification of metrology practitioners of all
This is the thirty-second in a contiguous series of levels. Such questions I could consider normal, as
essays, ostensibly chartered to be concerned with the such a subject is somewhat allied to metrology educa-
subject of the education of professional metrologists tion. To over simplify my position (and to get your
but very often wandering in divergent directions attention), I am against the certification of metrology
according to the caprice or megrims of this columnist. technicians--I am opposed to the implication in any of
They are invariably in the form of an insouciant open the formats I have seen thus far. Do not get me wrong,
letter to our Boss, the most respectful Editor-in-Chief of I do not oppose the concept. I in no way oppose the
his somewhat aperiodic journal (but he seems to like it). concept of requiring demonstrable certifiable compe-
tence of all individuals engaged in any way in any
echelon of the measurement process. But as of this
time I have not as yet been shown any practical means
of achieving irrefutable proof of competency in all of the
vast myriad of scientific technologies that compose the
measurement sciences that the proposed certifica-
tions would imply.
Back in 1997, the Measurement Science Confer-
ence, during their 26th annual meeting held in Pasa-
dena, Calif., sponsored an important panel session on
Phil Painchaud the topic of certification of metrology technicians. Our
former Editor-in-Chief, DeWayne Sharp, was session
chairman. The panelists included: Ms. Nancy Grayhek,
Dear Boss:
Manager of EIL Instrument Company’s western opera-
As I start my latest iteration of this continuous saga tions (they were a major nationwide independent com-
(May 28, 2002), I find that the last installment (written mercial calibration organization); Mr. Emil Hazarian,
January 10, 2002) has not, as they say in the newspa- Chief Metrologist of the Los Angeles County Depart-
per industry, “hit the streets as yet.” Hence, I have no ment of Weights and Measures (one of the largest in
questions from readers concerning that column to the USA); Mr. Herb O’Neil, head of the calibration
answer or to comment upon. Since our readers have program at Ridgewater College, Hutchinson, Minn.;
not as yet been able to read the portentous announce- Mr. Kevin Ruhl, head of Metrology Operations at TRW
ment made in that issue, let me reiterate: The bacca- Space and Electrons Group, Redondo Beach, Calif.;
laureate program leading to a Bachelor’s Degree in Mr. Phil Stein, then chair of this Measurement Quality
Quality Assurance with a Measurement Science Op- Division; and, of course, myself. Each panelist was
tion at California State University at Dominguez Hills requested to prepare a short (one minute maximum)
has been approved by the university administra- position paper, to be read at the time of that individual’s
tion. It is now undergoing the torturous approval route introduction. (Incidentally, I find that I have copies of all
through the Office of the Chancellor of the State of those position papers. If any of you readers are
University System and of the California Post Second- interested, I’ll be happy to send you copies.)
ary Education Commission. At the time of this writing,
My position paper pretty well describes my feelings
indications are that, despite traditional bureaucratic
on the subject: “I hold serious doubts if metrology
impediments, the necessary approvals will be granted
technicians can ever betruly, effectively, and uni-
in time for the program to commence in the Fall, 2002,
versally certificated with the conditions as they
semester. Furthermore, it appears that not only is the
exist today--for two fundamental reasons:
classroom format going to be approved, but also the
on-line (via Internet) version! As of this writing, the First, there are the political reasons. To be effective
university is planning to schedule classes to start then. and universally recognized, such a certification (or
Have you filed your application for studentship in this licensing if you will), in order to preclude biases to-
pioneering program yet? No? Why not? How do you wards or against certain segments of the population,
expect to become a true professional in the measure- would necessarily have to be operated and controlled
ment sciences? by some public entity such as the state. This would
Winter 2003 The Standard Page 8

bemuch the same as the current licensing for engi- In 1963, the California State Legislature created a
neers, brain surgeons, undertakers, barbers, physiolo- “blue ribbon” advisory committee, the California Pro-
gists, cosmetologists, chiropractors, etc. People in fessional Metrology Committee (CPMC), put the full
general, and the typical free spirits that characterize resources of the state behind it, and chartered it to draft
the metrology technician populace, rebel at the thought legislation to “Certificate (the words “License” and
of governmental control. And to make such a system “Certify” were specifically not used) all persons within
truly fair and unbiased, “grandfathering,” so common in the State that are in any way engaged at any level in the
the licensing procedures of certain other trades and design, manufacture, or service of any devices that
professions, would have to be totally eliminated. All measure or in any way quantify any goods or services
candidates would have to be individually examined as manufactured or sold within the State of California or in
to their own qualifications, and with no consideration of any way have an influence on the health and/or welfare
any duration of time that they may have been working of any person in the State of California.”
at a function, or of the questionable opinions of That group labored for over seven years with the
superiors who are desirous of retaining faithful problem. They, too, foresaw the impediments I men-
drones irrespective of relative comparisons to tioned above. They did devise a complex scheme that
norms established for the certification level. would have circumvented most of them and which
Second is the complexity of the technical aspects included the prohibition of “grandfathering.” They also
involved. There are literally hundreds of technologies had in their draft a clause which would prohibit any
involved in the science of metrology, each with a person who served on the CPMC from ever becoming
myriad of permutations of technical variation, of fine “certificated.” Their draft legislation included three
specialization, and of level. Add to that complexity the categories of individual: Metrologist (Scientist), Me-
astronomical number of measurement devices and trology Engineer (Designer), and Technician/Inspector
their variations of make, model, generation, and so (Repair/Calibrator). Each of these had three levels of
forth, and the magnitude of possible practical permuta- competence, and as starters 33 specific disciplines
tions of specific certificates becomes mind boggling as were named (it was planned to add more as required).
well as impossible to adminster. Without specific The draft never reached implementation due to exter-
certifications, the entire program would be meaning- nal factors beyond the committee’s control, and the
less. Technicians are doers. They perform very project had extended for such a duration that the
specific tasks, generally with designated items of hard- legislators who were sponsoring it had moved on.
ware. Without specific certification for the specific Those who replaced them knew of or cared little about
function on the specific hardware, any such pro- the causes which instigated the project.
gram would be meaningless.” A by-product of their efforts, which remains to this
While I wrote the above monologue five and one- day, is the Measurement Science Conference (MSC).
half years ago, my opinions have not changed much. Ten years ago I was asked to write the early history of
Of the six panelists, five in their position papers-- the MSC. It was impossible to separate the early
although each looking at the problem from a different history of the MSC from the CPMC. As a result, my
perspective--came to virtually the same conclusion. early history of the MSC contains a rather detailed
Over 200 people crowded into a room posted for 90 explanation of the early work of the CPMC. If anyone
maximum for the hour and one-half-plus session. By a is interested, I have a few copies. That should be
show of hands it was determined that well over half of enough for this issue. Don’t forget or procrastinate; get
the attendees were employers, metrology laboratory your applications in to Dean Gordon or Dr. Watson for
managers, or other supervisors of metrology techni- enrollment in the only baccalaureate level program in
cians. During the session, several managers (some measurement science in this nation. Remember, you
executives in large commercial metrology organiza can get your degree on campus, on-line, or, if you can
tions) got up and stated they would not under any get enough students together, they will come to you
circumstances give any credibility to any certifications and present the program at your facility.
issued by any entity outside of their own organizations. You can still reach me at::
When the session was concluded, by another show of
hands it was determined that greater than 90% of the painchaud4@cs.com
attendees believed that certification of metrology tech- or if that one does not work try
nicians would be an exercise in futility; on the other
olepappy@juno.com
hand, the agreement among those who claimed to be
employers of metrology technicians was unanimous.
Winter 2003 The Standard Page 9

Statistical Issues in Measurement


By Philip Stein
An invited paper for the ASQ Statistics Division
July 2002

Measurements and measurement science (metrology) have a long and rich history of a strong relationship with
the statistical sciences. This has been exemplified at the U.S. National Institute of Standards and Technology
(NIST), formerly the National Bureau of Standards (NBS), where this nation keeps its standards and supports
metrology and metrology research. At the risk of annoying many other outstanding NBS and NIST statisticians
by their omission, I note that for many years the Bureau employed John Mandel, Jack Youden, Churchill
Eisenhart, Brian Joiner, Robert Paule, Lynne Hare, Raghu Kacker, Mary Natrella, and Harry Ku.
There’s a reason for this. Metrology is essentially a statistical pursuit. It is the study of measurement error (we’re
now supposed to call it measurement uncertainty). Was there no uncertainty, there would be very little science
in measurement science; only a great deal of good engineering would be required to build and use measuring
equipment.
Statistics, therefore, are necessary to understand measurements. In fact, 30 years ago at NBS, a philosophy
of measurement assurance was proposed by Eisenhart, Cameron, and Pontius. They reasoned that measure-
ment is, in fact, a process. As such, it may be addressed by the entire armamentarium of process improvement
tools developed and made familiar by the quality profession. (Envision, if you will, that measurement is a
manufacturing process whose output is numbers). We can, and do, use Pareto charts, scatterplots, histograms,
fishbone diagrams, control charts, designed experiments – the entire collection of quality tools, for the design,
control, and improvement of measurements. More about this later.

Statistical Thinking in Measurements


Statistical thinking is crucial in making and understanding measurements, and natural variation is as poorly
understood and respected here as in other quality areas. In this paper, I hope to address two rather different topics:
statistical thinking in measurements – as a tutorial for non- statisticians – and statistics in metrology, to give the
technically-oriented quality professional a detailed taste of some of the more interesting and unusual ways that
industrial statistics are used every day in support of measurement science. I’ll also discuss the relatively new
international guide for the expression of uncertainty in measurement, which should be of great interest to
apprentice and journeyman alike – since it will certainly play an increasing role in many quality applications in the
near future.
Although we most often think about measurement, and about statistical applications, in the context of
manufacturing and other engineering pursuits, the principles, and the methods, are equally useful in all other
areas where quality is practiced – for example healthcare, service industries, and business management. For a
good sample of efforts in this area, see the cover story and lead articles in the February, 2001, issue of Quality
Progress, where Tom Pearson and I discussed the use of metrology tools in business.

Statistical Tthinking and Engineering Specification


Specifications are so widely used, and so useful, that we often don’t even notice that we are using them. They
are needed in virtually every circumstance where people need to communicate details about objects, equipment,
or processes. Specifications are almost always stated quantitatively – “the nominal diameter is a shaft is 0.156
inches”– and much of the time these quantities are given without comment or qualification. Even the most cursory
inspection of this concept, though, says that a stated value and unit are of little utility without information both about
context and about variation.
In much of our daily lives, state and local government operate a crucial but invisible infrastructure of weights
and measures standardization. As a result, we can buy produce and gasoline without thinking much about
measurement variation. When dealing with measurements in our professional lives, though, the supporting
structure is our own responsibility to a much greater degree. The classic metrology quote is from Lord Kelvin:
Winter 2003 The Standard Page 10

"When you can measure what you are speaking about, and express it in numbers, you know something about
it; but when you cannot measure it, when you cannot express it in numbers, your knowledge is of a meager and
unsatisfactory kind: it may be the beginning of knowledge, but you have scarcely, in your thoughts, advanced to
the stage of science."
But, in fact, expressing something in numbers is only part of the equation. In order to ‘advance to the stage
of science,’ expression of a quantity requires two other statements: a unit of measure and some description of
the quality of the state of knowledge of the quantity. This information and more are sometimes called ‘metadata’
– data about the data.
For our purposes here, we’re most interested in a statement of quality for a number. Most often, we would like
this to be an expression of spread or possible variation. The VIM (International Vocabulary of Measurement,
1993), a world-wide standard, defines uncertainty as 'Parameter, associated with the result of a measurement,
that characterizes the dispersion of the values that could reasonably be attributed to the measurand.' A
measurand is a particular quantity subject to measurement.
We make the further distinction that a single measurand is expected to have a unique value when measured
in a single measurement act and that the ‘dispersion of values’ observed is due to sources of variation within the
measurement process. In many cases, repeated measurements of a unit will experience variation because the
item being measured also varies. It’s important to separate these variations both in our minds and where possible
in our calculations. We try to focus on, understand, and quantify the uncertainty of the measurement process; and
once this has been accomplished, we can separately characterize the variation of the object or process being
measured.

Quality of a Measurement Result


Although uncertainty has been defined in international standards, most of the time the quality of a
measurement process – confounded with the variation of the item being measured - is described by a tolerance.
The tolerance concept is deeply rooted in engineering and manufacturing practice and history. Virtually every
specification written – for products, processes, or measurements - describes not only a target value but the
permissible variation around that target and does so using a tolerance.
Tolerance statements can be described statistically as what is called a uniform or rectangular distribution. In
this model, if a product or process is in tolerance, the measurement value can be found somewhere inside the
interval bounded by the upper and lower tolerance limits. It absolutely cannot be outside those limits.
The rectangular distribution offers no information about what the measurement value might be. It has an equal
chance of being found anywhere between the limits and no chance of being found outside. The tolerance
statement is an absolute – it leaves no room for ambiguity. The value is somewhere inside the interval – period.
Statisticians and scientists have long known that the rectangular distribution is not usually a good description
of how either manufacturing or measuring processes actually operate. The rectangular is appropriate and useful
in other applications (as George Box said, “all models are wrong – some are useful”). When a process operates
smoothly, without disruption or deliberate changes, the result is usually best described by a different distribution
– the Normal or Gaussian, represented by the familiar bell curve.
In this model, the measurement value is most likely to be found at or near the target value. Values farther away
from the target are less likely. Unlike the rectangular distribution, there is a lot of information about where inside
the interval the actual value is likely to be.
There is thus a small chance that the measured value is very far from the target, and in theory the distribution
stretches infinitely far in both directions. So another way in which the Normal is different from the rectangular is
that the value can be outside the tolerance limits; although it may be improbable, it is possible.
We can say that the rectangular distribution represents the “voice of the customer,” indicating the desire of the
user of a measurement to know absolutely the limits between which a value will lie.
The Normal distribution represents the “voice of the process”, indicating that no such absolute knowledge is
possible or available in real life.
Winter 2003 The Standard Page 11

Following this logic, the best specifications do not use rectangular tolerances at all but rather specify the
‘dispersion of values that could be reasonably attributed to the measurand.’ This could be done simply, even for
the general public, with specifications such as “95% of all trains will arrive within ten minutes of their scheduled
time,” or “at least 50% of the pills in this bottle will have the stated amount of active ingredient.”
Educating the public in these concepts has been almost fruitless. While quality engineers and other quality
professionals continue to stress the understanding of variation, published specifications and internal manufac-
turing documents still speak almost universally in tolerances.

Statements of Uncertainty in Reporting of Results


Is any progress being made in changing this? Yes. In the calibration and metrology business, international
quality system standards such as ISO/IEC 17025 require the understanding and use of uncertainty in reporting
of results, and this same requirement is quickly being deployed to test labs by the accreditation bodies. This effort
is strongly supported by a new world-wide standard (1995) – The Guide to the Expression of Uncertainty in
Measurement - universally known as the GUM.
While accreditation is not required of laboratories under ISO 9000:2000, it is a requirement of QS-9000 (third
edition) for labs that provide outside services to registered organizations. In support of this, many manufacturers
of measurement and calibration equipment are now writing specifications for their products in terms of GUM
uncertainties.
At the same time, International standard ISO 14253-1 for reporting of dimensional measurements has taken
a different approach to tolerances. Although this standard still has a tolerance ‘feel’ to it, defining ‘zones’ for in-
spec and out-of-spec results, it also defines ‘uncertainty ranges’ around the specification limits, where
measurement uncertainty makes a definitive declaration of in- or out- impossible (See Figure 1). There is strong
pressure, especially from Europe, to adopt this standard for physical dimensions, even all the way down to the
machine shop floor. While such deployment is unlikely to happen soon, at least things are moving in the right
direction.

ISO 14253-1
Uncertainty Range s

Nonconformance Uncertainty Conformance Zone Uncertainty Nonconformance


Zone Range Range Zone

Out of Specification ???? In Specification ???? Out of Specification

Figure 1

How Quality Science and Statistical Science are Used in Measurement Science
We started this paper by explaining how statistics and metrology are inextricably linked by the need to describe
measurements in a way that respects the distributional nature of the results. In addition, quality tools and
statistical tools play an integral part in the daily practice of metrology. Herewith we discuss some examples.
Winter 2003 The Standard Page 12

Measurement Assurance and Quality Tools


Measurement is a process. The quality profession has for the last 80 years or so developed a powerful tool
kit for understanding and improving processes. These tools are based in the statistical understanding of variation
and exploiting analysis of the variation to better describe (then understand, then improve, then control) the target
process.
Envision, if you will, that measurement is a manufacturing process whose output is numbers. Using the familiar
tools of quality assurance, we can carry out programs of measurement assurance. Just as a manufacturing
process is intended to produce consistent product, a measuring process should produce consistent measure-
ments. In order to not confound the measuring process with variation among many measured units, we measure
the same unit repeatedly.
In most cases, the measurement is not destructive of the sample; and we can, therefore, expect to get the same
answer every time. Any observed variation, and there will be variation, can be attributed to the measurement
process. The repeatedly measured sample is known as a check standard. Defining and managing check
standards for destructive tests can be done, but it’s not easy and is beyond the scope of this paper.1
The check standard is repeatedly subject to the same measurement process as the regular measurement
load. This could be, for example, a factory tester for manufactured product, or it could be a calibration workload
in a commercial laboratory. The check standard should, wherever possible, be handled in the same way as other
items being measured. If, for example, there is a cleaning process before the measurement, and if the check
standard will not be damaged or prematurely worn out by repeated cleaning, it should undergo that process just
as product will.
The ideal result of these replicates is for the measurement result to always be the same. When it is not, we
analyze the variation in the same way that we would analyze varying product in manufacturing.
Starting with a histogram, we visualize the data to see if it is distributed roughly as expected. Most of the time,
the measurement data from check standards will be variables and we can expect a Normal result. In some cases
when the quantity measured is an amount of something and when the scale is long and includes or approaches
a zero amount at one end, a lognormal distribution might be expected .2
The spread of the check standard data, usually defined as some multiple of the standard deviation, is a
measure of the capability of the measurement process (by analogy with process capability indices).
To visualize this as a capability, compare the spread of the check standard data to the spread of the product
being measured in normal production. (See Figure 2).

Product distribution

Measurement Distribution
as reflected by check
standard

Product specification
limit
Figure 2
Winter 2003 The Standard Page 13

The measurement distribution as shown by the check standard distribution can be compared to the product
distribution to yield a capability statistic.
Alternatively, you could compare the check standard spread to the specification of the product. After all, your
purpose is to be able to detect the mean and spread of the product and to determine whether it meets the
specifications. If the spread of the measurement process (as shown by the check standard) is large, of the order
of the width of the specification, your measurements won’t be fit for their intended purpose.
A good rule of thumb is that the measurement distribution should be no wider than 20% of the width of the
product distribution. Narrower is better.

1
For further information on how this can be done, see my Measure for Measure column in Quality Progress,
January, 2002.
2
Roughly speaking, Normal is indicated when the measurement error is a constant in the measurement units,
and Lognormal is indicated when the error is a constant percentage.

Control Charts in Measurement Assurance


If the measurement capability is adequate for its intended purpose, the measurement process should be
placed in statistical control. Periodic measurements of check standards may be control charted in exactly the
same fashion that product is monitored. A significant change in the chart, as detected by your favorite
constellation of rules, is likely to be indicating a special cause of variation and should be investigated for root cause
and corrective action taken as indicated.
Ted Doiron of NIST’s Precision Engineering Laboratory was kind enough to supply me with actual data from
various dimensional (length) check standards. Here are three control charts that demonstrate different situations.

Individual X Control Chart of 500 mm Standard over 16 months

UCLx
-750

D-825
e
v
i Mean
a-900
t
i
o
n-975

LCLx

35200 35400 35600


Dayseq

This check standard shows good control.


Winter 2003 The Standard Page 14

Individual X Control Chart of 0.1 inch standard over 1 year


UCLx

Mean
0

LCLx
-1 0 0
D
e
v
i
a-2 0 0
t
i
o
n
-3 0 0
Single point out of control
Special cause

3 5 85 3 5 92 3 60 0 36 0 7
0 5 0 5
Day seq

This check standard shows good control except for a single point where something wrong was discovered
as a result of interpreting the chart.

Individual X Control chart of 42 inch length standard over 8 years

UCLx
0.
D 0
e
v - Mean
i
a0.
t 3-
i
o 0.
n LCLx
6-
0. 330 337 345
9 More than 15/16 point s
00 50 Day 00
seq below the mean

This 42-inch check standard was tracked for almost a decade. Towards the end of that period, the control chart
showed a drift down, indicated above by the last points showing out-of-control for 15 out of 16 points below the
center line (mean). Measurements made with an alternative method confirmed that the standard (not the
measurement method) had in fact shrunk over that time period.
Winter 2003 The Standard Page 15

If the capability was not acceptable when the chart was first installed, but the chart was under control, this
probably indicates a common-cause source of variation in the measurement process. As with any other
improvement effort, measurement assurance tools follow familiar paths such as using designed experiments to
find ways of understanding and reducing measurement variation.
An unexpected benefit of control charting your check standard measurements occurs in the determination of
calibration intervals. Intervals is an important and persistent topic in metrology, since periodic calibration of
instruments is often the largest single expense for a laboratory. Make the interval too short and you could go
broke. Make it too long and you might make a mistake in a measurement result. Worse yet, if you discover that
an instrument went out of calibration while it was being used, most calibration quality system standards require
you to track back through all previous work done by that instrument (back to the last time it was known to be ‘good’)
to make sure that no harm was done. If you have been control charting your check standards, the last time the
measurement process – including the instrument – was known to be good was the most recent point on the chart,
and that was probably only a day or two ago. This is much better than tracing back to the last annual calibration.
Instruments and their processes exhibit two common failure modes – catastrophic and drift. Sudden changes
in a control chart are a strong indicator that something broke. Long-term changes will often track drift of
measurement standards.
If it’s time for a calibration and a control chart of check standards shows continuous, ongoing evidence of
statistical control of the measurement process, why is calibration needed? It probably isn’t. A formal process of
evaluating measurement assurance data can be an enormous money-saver by preventing unnecessary
calibrations. There is a possibility that the reference standard and the check standard are both drifting in the same
direction by about the same amount. When this happens, the control chart won’t show a need for calibration when
the need actually exists. To prevent this unlikely occurrence, calibration is still performed but at a greatly reduced
frequency. For even greater confidence, multiple check standards may be used – the chance of several drifting
together is small indeed.

What If the Uncertainty is Larger Than the Process Requires?


Excessive common cause variation – inadequate capability – can usually be evaluated by conventional design
of experiment and Analysis of Variance (DOE/ANOVA) studies in which appropriate parameters of the
measurement process are deliberately varied and the sources of excess uncertainty found and eliminated. When
designing and carrying out these experiments, be sure to remain aware of the various influence factors that can
increase measurement variation. The measurement and test equipment will contribute, but so will the
environment (temperature, humidity, weather), the operator, the measurement fixture; and even the data
analysis and display software can be error sources. If these are not exercised in an experimental design, they
will never be found and thus can’t be eliminated or reduced.

Statistical Modeling of Measurement Processes


Suppose the problem is to accurately lay out a one-kilometer square of land and to then calculate the actual
area enclosed. Using any tools you might need – compass, tape measure, square, transit or GPS – start at a
marked point, then walk one km in the direction of magnetic North (place a mark), one km East (mark), one km
South (etc.) and again one km West. Upon completion, your first instinct is to expect that you will have returned
to the starting point, but of course measurement errors in length and direction will mean that you won’t get exactly
back to your original mark. The degree to which you have returned is known as closure, and it’s a measure of
the quality of your square-km-generating process.
Winter 2003 The Standard Page 16

At this point there are three fundamental approaches to analysis of the closure you have achieved.

First is the engineering approach. Part of the statement of the original task is a specification of the tolerance
of the process – a limit to the allowed closure error. For example, we could specify that closure anywhere within
a two-meter circle is acceptable. Because you will not have (can not have) perfectly returned to ‘go,’ the square
you have marked will contain a ‘surveyor’s gore’ – an area that belongs in your square according to some
measurements and belongs outside according to other measurements (see Figure 3). If closure has been
achieved to within the stated tolerance, you’ve finished the task. If not, some process rules have to be laid down
that state what to do next.

N
"Surveyor's
Gore"

Circle of permissible
Start error ("Closure")

Finish

Figure 3

Second is the statistical approach. Typically, a second set of measurements would be made as replicates of
the first set. The measurement lingo for this process of collecting more than one datum per cell is ‘overspecifying’
or ‘overdetermining’ the problem. In many cases a more sparse design can be used, where only a fraction of the
cells contain replicates; but because of the sequential nature of the measurement process here, this wouldn’t
make sense.
Winter 2003 The Standard Page 17

In the engineering case above, the cumulative error of all angles and lengths winds up being assigned to the
position of the last point, and all closure error is visualized as arising from the last point being in the wrong place.
Once replicate data are available, another model such as a least-squares fit can be used to fairly distribute the
errors among all individual (angle and distance) measurements. This results in a more accurate picture of the
actual locations of all boundaries of the area, and any standard measure of goodness-of-fit for the model can be
used in place of the closure circle. Although a linear model (polynomial fit of degree one) is often adequate in these
applications, other physical situations have been modeled with polynomials of degree as high as eight. It’s difficult
to imagine a physical phenomenon that can’t be very accurately represented by an equation with nine adjustable
coefficients.
Third is the metrological approach. A professional metrologist has a background and experience in the physics,
chemistry, or other science appropriate to the measurement at hand and may also have a degree of mathematical
sophistication. A metrologist looking at this problem will proclaim “but the earth isn’t flat” and will suggest replacing
a general-purpose polynomial equation with a spherical model that better (but still not perfectly) reflects the
physical realities. This will still result in a fitting procedure, testing of the residuals, and an evaluation of the
accuracy of the results; but the model equation is likely to be much simpler for any given accuracy.
It is not my intent to denigrate either the engineering or the statistical approach. All three methods have
advantages and require different skills from the practitioner, and the choice of an approach represents a tradeoff
between simplicity of application and accuracy of the result. A closure error of two meters might be far better than
is needed for the practical application of this area measurement, at which point the extra data taking and model
making called for by the other methods would be a waste of time and money.

A Statistical Approach To Inspection and Disposition


Another area in which the engineering and the statistical approaches diverge is around complex inspection
processes. Often, these inspections are done by automatic test equipment (ATE). Many measurements are
carried out, often simultaneously, and the results analyzed from an ‘inspection’ point of view. Invariably, this means
univariate tolerance-type thinking in which each answer is separately and independently compared to a set of
limits. The unit under test fails inspection if any one measurement exceeds specifications.
This is certainly an easily understood operation and is intuitively correct; but as so often happens in
straightforward engineering, our intuition is not sophisticated enough. Very often, multiple tests are correlated.
This means that a particular offset or bias in the measured value, due either to product deviation or measurement
error, is given more than one vote in the election to reject the part.

Suppose we’re measuring the shape of a curved part such as the faceplate (panel) of a color television picture
tube. The panel is oriented face up in a measuring jig; and an X-Y grid array of mechanical probes is applied, each
looking for the position of the glass in the Z direction. A conventional definition of part failure would be to reject
if any point is out of tolerance.
Winter 2003 The Standard Page 18

True statistical independence of all of the points, which would support the conventional definition, does not
exist. There is a lot of autocorrelation due to the manufacturing process because the curvature of the panel is
not free to change independently at every point; but, rather, if there is a deviation it will be some sort of gradual
slope, rise, slough, or lump extending over several X-Y points. The measurement process often contributes
autocorrelation as well: for example, if the panel does not sit down perfectly in its jig but perhaps tilts a bit. All
of the points along one side or near one corner will be distorted in a similar way.
Now if the problem is large enough to send several points out of tolerance, rejection is proper; and it’s not
obvious why there’s a problem. Suppose, though, that a perfect measurement would show all points within
tolerance even though, due to one of the aforementioned distortions, some points are not precisely at the values
they should occupy. To this, add a small, acceptable amount of random variability due to ordinary measurement
noise. Now, the probability of failure of one point is large and is magnified by the fact that many points were brought
close to their limits by the correlated variation; and so the chance that one point will randomly be pushed over
the line is unfairly increased.
What’s the ‘right’ way to approach the analysis of these measurement data? It depends on the voice of the
customer. Some degree of distortion is inevitable in a picture tube faceplate, and different forms of distortion are
more – or less – visible and annoying to the viewer. The point-by-point inspection wrongly increases producer’s
risk because it controls distortions that don’t correlate well with viewer dissatisfaction.
One approach to elimination of the increased risk is to adopt a disposition strategy that does not depend on
every single point being correct but rather evaluates the geometry as a whole. The same grid of points is
measured in the same way, but the answers are compared to the nominal curvature in a two-dimensional fitting
process. The residuals from this process are tested using standard goodness-of-fit criteria, and a rejection
threshold is placed on the overall difference between the ideal and actual curvature.
Point-by-point tolerances are still computed, and rejection due to them is still possible in order to catch single
large-value outliers that ruin the picture but don’t have enough leverage to ruin the fit. The tolerance for those
outliers can be much greater than the tolerance was when the points were used for the detailed inspection. Other
analytical methods of outlier detection can also be used at that point.
This same way of thinking has been successfully applied to flatness of surface plates used for dimensional
inspection and to frequency response of bandpass amplifiers (in the latter case gain measurements are made
at many frequencies across the passband and the result compared to an ideal shape. The individual values are
highly correlated so they shouldn’t be allowed independent votes).
A good general approach when the errors are those of geometry, e.g. the picture tube or surface plate, is to
express the errors in terms of a distorted coordinate system – one that can include rotation, translation, and
improper scaling.
Winter 2003 The Standard Page 19

Calculation of Measurement Uncertainty


The ISO Guide to the Expression of Uncertainty in Measurement (GUM), and many references based on
it, set forward a method for enumerating and quantifying various influences on a measurement that contribute
to variation of the reported value. The method is often referred to as an uncertainty budget, in which the influence
quantities are listed and estimated either by statistical (type A) or other (Type B) means. Type B generally refers
to scientific, engineering, or metrological judgment of experts. These estimates are then combined and
expanded to yield a single number, the Expanded Uncertainty.
While each measurement process has its own budget, there are certain common factors that should appear
in all budgets, although they may be estimated and declared as negligible where appropriate. These factors are
inherited calibration uncertainty, resolution, and repeatability or reproducibility.
The GUM is a strong and thorough document and, as with many standards, is seen as mathematically intense,
very compressed, and unapproachable. Many countries, the U.S. included, have reprinted local versions of this
standard and added appendices with examples intended to elucidate; and this has been some help.
The mathematical approach is quite straightforward. It sees the uncertainty as a RSS (root-sum-square, or
‘square root of the sum of the squares’) combination of standard deviations of the individual measurement
influence variations. Corrections are then made for any correlation among the influences; and a ‘sensitivity
factor,’ or weighting factor, is included to facilitate expressing all of the quantities in the same units. The
combined uncertainty is then multiplied by a ‘coverage factor’ based on the total degrees of freedom of the
components. The result is stated as an expanded uncertainty that represents an approximate confidence
interval. The width of the interval is stated and is conventionally cited as 95%. The figure below shows a sample
budget with some real and some fictitious data. Note especially the extensive comments, without which the
budget raises more questions than it answers.

Uncertainty Budget for Gage Blocks - 3 Inch


August 17, 2002

Influence Magnitude
Type Distribution Divisor Quotient Square Comments
Quantity (microinch)

Uncertainty of
Reference 1.5 B Expanded 2 0.75 0.57 From NIST
Standard Certificate

Variation of Standard
Measurement 0.78 A Normal 1 0.78 0.61 Deviation from
Process control chart

Resolution of
Mahr Federal
Measurement 0.1 B Rectangle 1.73 0.057 0.0033
Process Gauge

From
Reproducibility 0.55 A Normal 1 0.55 0.3 experimental
data

Total 1.22 1.46


Uncertainty

Expanded 2.5 k = 2 rounded


Uncertainty up
Winter 2003 The Standard Page 20

1. The uncertainty of the reference standard was determined by a calibration traceable to National Standards.
2. The total variation of the gage block measurement process is determined using Measurement Assurance
techniques (check standards and control charts, ref ISO 10012-2), as calculated by commercial software.
3. The check standard for this process is a three-inch gage block.
4. The control chart for this process was in control at the time the unit under test was measured. This provides
evidence that the variation is Normally distributed, thus the standard deviation is the correct statistic to use.

Measurement Model
The following influence quantities are included in the variation measured by the control chart:
1. Temperature difference between the reference standard and the UUT at a nominal 20C. (including operator
body heat, block soak time).
2. Variation in temperature measurement equipment.
3. Variation in the digital readout due to vibration, electrical noise.
4. Random positioning errors and other variation due to operator technique.

The following influence quantities may not have been captured by the control chart but were considered of
negligible magnitude:
1. Difference in thermal expansion coefficient between the reference standard and the UUT, combined with
difference in temperature of the standard and UUT from 20C.
2. Difference in elastic deformation of block surfaces under pressure of the measuring probe.
3. Deformation of blocks under under their own weight.
4. Error in wringing film thickness (only single blocks are measured here).

The following influence quantities have not been evaluated; but, since the control chart is okay, there is
evidence that they do not have a significant effect:
1. Software bugs.
2. Differences in surface finish of the reference and unknown blocks.
3. Long-term drift of the check standard hidden by periodic resetting of the control limits (was not done here).

The following influence quantities have not been evaluated:


1. Drift and wear of the reference standard between calibrations.
2. Gain calibration of the electronic indicator in the gage block comparator.
3. Bias of temperature measurement equipment.
(end of budget)

It is clearly understood by the authors of the standard and by most practitioners that the results are estimates
and are usually pretty rough estimates at that. The assumption of normality for the separate variations is probably
okay but certainly not perfect. In addition, some influence factors are estimated by engineering methods rather
than by measurement – but the expansion coefficients used are that of the nominal material, not of the actual
composition of the device, etc.
In addition, some data about the magnitude of the variation may by available only in some alternate form such
as a specification with a tolerance. The uncertainty of a standard or instrument used as part of a measurement
setup is often obtained by reading its calibration certificate. If the certificate conforms to GUM requirements, the
expanded uncertainty and coverage factor (or confidence interval) will be stated and can be used directly. Other
information, such as manufacturer’s manual specs, nonconformant certificates, and engineering estimates must
Winter 2003 The Standard Page 21

be included with an assumption as to their statistical distribution and, if not Gaussian, transformed so that a
meaningful standard deviation can be calculated. While any transformation is allowed, several are specified in
the GUM and have become part of the routine practice of making budgets.

The Rectangular Distribution


Common practice is to assume that tolerances contain no information about where within the stated interval
the mean might lie at any moment. To include this kind of information in an uncertainty budget, transform the
tolerance interval to an estimate of the equivalent standard deviation by dividing the half-interval by the square
root of 3.

Other Distributions and Their Uses


The GUM discusses the use of other probability distributions in addition to the rectangular (discussed above)
and the Normal, or Gaussian. Specifically, the triangular and the U-shaped distributions are mentioned. Do these
have any practical application?
In many cases, estimation of uncertainty is just that – estimation. This is typical of influence quantities of type
B, whose magnitude is determined by other than statistical means. The incorrect choice of a distribution is unlikely
to make a big difference in the result. Nonetheless, our intent when making these estimates is to enhance our
understanding of the underlying science, and the thinking process used to choose and correctly apply a
distributional assumption supports this understanding.
Still, what use are these forms? Consider the triangular distribution. One appropriate application is when
calculating the temperature effect on calibration of gage blocks (dimensional standards). If two blocks – an
unknown and a reference standard – have been soaking together in the same thermal environment, for example,
on the same metal plate, the most likely temperature difference between them is zero. The likelihood of the blocks
being different decreases with increasing temperature difference and is symmetric – either block could be warmer.
The limit is equal to the environmental temperature swing: one block could be as warm as the room gets, and
at the same time the other block could be as cool as the room gets, although this situation is extremely unlikely.
The triangular distribution represents this situation exactly. If you set a = warmest minus coolest, the possible
range is 2a (one block warmest all the way to the other block warmest). As with the rectangular, this should be
transformed to a standard deviation so that it can be combined with other contributions to the u. The standard
deviation of a triangular is = a/root 6.

In other situations concerning environmental conditions, other factors prevail. Suppose I wish to calculate the
influence of temperature on a single dimensional measurement, such as when using an end (rod) standard. The
standard is being used in a controlled temperature room, and at first blush you could assume that the rod is the
same as the time-average room temperature. This might be a good approximation, but it likely is not. All but the
most sophisticated environmental controls simply turn heaters or coolers on and off according to the demands
of a thermostat. Some systems don’t even run the air circulation continuously but cycle that as well.
A graph of the room conditions versus time will reveal, in these cases, two common conditions: coolong on,
with the room air temperature close to the minimum within a range, and cooling off, with the room air temperature
close to the maximum. The temperature actually spends very little time near the center of the range. This is best
represented by the U-shaped distribution. If the overall temperature range is 2a (=a above and a below the elusive
middle), then, again transforming to a standard deviation, the uncertainty contribution = a/root2.

Correlation Among Influence Quantities


Suppose we are measuring the volume of a rectangular solid using a caliper. Three measurements are made,
one each in the X, Y, and Z direction. While each measurement will have some random component of variability,
a bias in the caliper will produce the same error in X, in Y, and in Z. These errors are perfectly correlated since
they arise from the same source. The total uncertainty of each axis measurement will be partly correlated. The
GUM treats this problem by explicitly including the correlation structure in the calculations, but these relationships
are rarely as well known as they are in this example.
Winter 2003 The Standard Page 22

Simplified Calculations
Because uncertainty is only an estimate, most practitioners will either ignore correlations or, where it obviously
exists, assume perfect correlation and calculate accordingly (add the standard deviations of that portion rather
than including it in the RSS).
Other ways are also used to simplify GUM calculations. It is commonplace to reduce all statements of variation
to the same units before entering them in the budget, therefore setting the sensitivity weights equal to one. If we
are budgeting for a dimensional measurement, we calculate the dimensional effects of temperature as a side
calculation and put the result into the budget as a length variation. This makes the budget itself simpler and easier
to understand.
Finally, many practitioners simply skip trying to figure out the effective degrees of freedom for the overall budget
and just choose a coverage factor (multiplier) of two – stating that this represents approximately a 95% confidence
interval around the stated value.
Taken together, along with software available, these simplified approaches have made uncertainty budgeting
a tractable, if still annoying, chore.

Final Thoughts
It’s an understatement to say here that statistics and metrology are inextricably intertwined. We have given
just a few examples where taking an approach that respects the statistical nature of measurement variation results
in powerful technical improvement and often in money saved as well.
Yet, statistical thinking about measurement results and measurement data is far too rare. As a laboratory
assessor, I have visited more than 80 calibration labs – some of them more than once. Only two of them were
using control charts, and only one of those was doing it correctly (‘correctly’ not some esoteric concept here – any
quality engineer could do this correctly using only the tools from the CQE body of knowledge).
Instrument users still set calibration intervals based more on guesswork and tradition despite the abundance
of statistical tools to help do it better.
Despite some concerted efforts, tolerance thinking and specifying still rules our engineering attitudes, with the
predictable outcomes of increased risk for both producer and consumer.
Yet, despite these depressing facts, inroads are being made. Wider acceptance and use of the GUM and
formal requirements of quality system standards are gradually forcing the world to change for the better. If we
do our part by educating ourselves and those around us, it can only speed the change.

References:
• Eisenhart, C. Realistic Evaluation of the Precision and Accuracy of Instrument Calibration Systems. J. Res
Nat’l Bur. Stand 67C 161-187 (1963)
• Belanger, B., and Croarkin, C. Measurement Assurance Programs NBS Special Publication 676 - 1984
• Guide to the Expression of Uncertainty in Measurement - 1995. Published by a joint committee of ISO, IEC,
and others (known as the GUM)
• VIM, International Vocabulary of Basic and General Terms in Metrology - 1993. Published by a joint committee
if ISO, IEC, and others
• Taylor, B.N., and Kuyatt, C.E. Guidelines for Evaluating and Expressing the Uncertainty of NIST Measurement
Results NIST Technical Note 1297 - 1994
• General Requirements for the Competence of Testing and Calibration Laboratories INTERNATIONAL
STANDARD ISO/IEC 17025 - 1999
• Geometrical Product Specification - Inspection by Measurement of Workpieces and Measuring Instruments:
INTERNATIONAL STANDARD ISO 14253-1
• Stein, P.G. Measure for Measure - Bimonthly column in Quality Progress, ASQ beginning September, 1999.
(See also lead articles September, 1999, and February, 2001)
Winter 2003 The Standard Page 23

Note on permission to use cartoons. These are public domain freeware and permission is granted. See http:/
/www.strange-matter.com/faq.html if you need details or more cartoons.
Winter 2003 The Standard Page 24

Measurement Quality Division Officers


Chair
Duane Allen .................. U.S. Navy, P.O. Box 5000, Code MS11, Corona, CA 92878-5000 909-273-4783 V
e-mail: allendk@corona.navy.mil 909-273-4599 F
Past Chairman
Mark Schoenlein ........... Owens Illinois Plastics Group, One SeaGate 29L-PP, Toledo, OH 43666 419-247-7285 V
e-mail: mark.schoenlein@owens-ill.com 419-247-8770 F
Chair Elect
Dilip Shah ...................... E=MC3 Solutions, 197 Great Oaks Trail #130, Wadsworth, OH 44281-8215 330-328-4400 V
e-mail: emc3solu@aol.com 330-336-3974 F
Vice Chair for Regions
Samuel Windsor ............ Filltronic Comtek, 31901 Comtek Lane, Salisbury, MD 21804 410-341-7751 V
e-mail: swindsor@fc-us.com 410-341-0330 F
Treasurer
Frank Voehl ................... Harrington Group, 11501 Lake Underhill Road, Orlando, FL 32825 1-800-476-9000 V
e-mail: FVoehl@aol.com
Secretary
Joseph Filipowicz .......... Alliance Compressors, 715 Oakland Drive, Natchitoches, LA 71457 318-356-4570 V
e-mail: joeflip@earthlink.net 318-356-4570 F
Auditing
Karl F. Speitel ................ 14 Kalleston Drive, Pittsford, NY 14534 716-385-1838 V
e-mail: Karlfspeitel@aol.com
Certification
Christopher L. Grachanen ..... Manager, Standards Engineering
P.O. Box 692000 MS070110, Houston, TX 77269-2000 281-518-8486 V
Compaq Computer Corporation, Corporate Metrology 281-518-7275 F
e-mail: chris.grachanen@compaq.com
Programs
Open

Membership
Open

Education
Open

Publications
Open

Newsletter Editor
Frank Voehl ................... See Treasurer

Historian
Open

Standards Committee Representative


Bill McCullough ............. Metrology Manager, Bently-Nevada Corp. 775-215-1207 V
1631 Bentley Parkway South, Minden, NV 89706
e-mail: bill.mccullough@bently.com
NCSL Representative
Christopher L. Grachanen ..... See Certification
Simmons Scholarship
Norm Belecki ................. 7413 Mill Run Dr., Deerwood, MD 20855-1156 301-869-4520 V
e-mail: n.belecki@ieee.org
Website Manager
Christopher L. Grachanen See Certification

Please notify the editor of any changes.


Winter 2003 The Standard Page 25

REGIONAL COUNCILORS
Region 1 Region 9 Region 14
Joseph Califano, Hemagen Diagnos- Dr. Henrik S. Nielson, HN Metrology Chuck Carter, C.L. Carter, Jr. & Asso-
tics, Inc., 40 Bear Hill Road, Waltham, Consulting, Inc., 5230 Nob Lane, India- ciates, Inc. 1211 Glen Cove Drive,
MA 02154 • (417) 890-3766, FAX (617) napolis, IN 46226 • (317) 849-9577, E- Richardson, TX 75080 • (972) 234-
890-3748 mail: hsnielson@worldnet.att.net 3296, FAX (972) 234-3296, E-mail:
Region 2 Region 10 asqccarter@aol.com
Karl F. Speitel, 14 Kalleston Drive, Mark Schoenlein, Owens Illinois Plas- Region 15
Pittsford, NY 14534 • (716) 385-1838 tics Group, One SeaGate 29L-PP, To- Bryan Miller, Champion International,
Region 3 ledo, OH 43666 • (419) 247-7285, FAX Inc., P.O. Box 189, Courtland, AL 35816
Eduardo M. Heidelberg, Carter (419) 247-8770, E-mail: mark. • (205) 637-6735, FAX (205) 637-5202
Wallace, 61 Kendall Dr., Parlin, NJ schoenlein@owens-ill.com Region 25
08859 • (609) 655-6521, FAX (609) Region 11 Open
655-6736 Raymond Perham, Michelin Tire Corp.,
Region 4 Rt 4 Antioch Church, P.O. Box 2846,
Alex Lau, Imperial Oil, 111 St. Clair Greenville, SC 29605 • (864) 458-1425,
Ave W, Toronto, Ont, Canada M5W- FAX (864) 458-1807, E-mail: Please notify the editor of any
1K3 • (416) 968-4654, FAX (416) 968- ray.perham@us.michelin.com, or
errors or changes so that this list
5560, E-mail: alex.lau@esso.com home E-mail: r.perham007@aol.com
can be updated.
Region 5 Region 12
Open Donald Ermer, University of Wiscon-
Region 6 sin Madison, 240 Mechanical Engineer-
Open ing Bldg., 1513 University Avenue,
Region 7 Madison, WI 53706-1572 • (608) 262-
Rolf B.F. Schumacher, Coast Quality 2557
Metrology Systems, Inc., 35 Vista Del Region 13
Ponto, San Clemente, CA 92672-3122 Thomas A. Myers, Bellevue Univer-
• (949) 492-6321, FAX (949) 492-6321 sity, PMP, CQM, 1000 Galvin Rd. S.,
Region 8 Bellevue, NE 68123 • 1-800-756-7920
Open ext. 3714, FAX (402) 293-2035, E-
mail: tmyers@scholars.bell

REGIONAL MAP
The Journal of the Measurement Quality Division
American Society for Quality

The Standard Non-Profit


Organization
U.S. Postage
PAID
Milwaukee, WI
Permit No. 5419
American Society for Quality
Measurement Quality Division

Winter 2003

American Society for Quality


Measurement Quality Division
600 N. Plankinton Ave
P.O. Box 3005
Milwaukee, WI 53201-3005
TIME VALUED MATERIAL

Vous aimerez peut-être aussi