Of And: The Calibration Weights Balances

The Calibration of Weights
and Balances
Edwin C. Morris
and
Kitty M. K. Fen
Monograph 4; NML Technology Transfer Series
' Hl rJ ''Ln
D-06840-Q
The Calibration of Weights and Balances
Edwin C. Morris
and
Kitty M. K. Fen
Monograph 4: NML Technology Transfer Series
National Measurement Laboratory CSIRO
BMLSTORGEN
AP 336
This monograph combines and updates two earlier CSIRO publications: The
Calibration of Balances by David B Prowse, first published in 1985, and Balances and
Weighing by Edwin C Morris, first published in 1992.
The monograph is one of a series giving practical advice on the measurement of physical
quantities. The intention is to produce monographs in all flelds of physical metrology.
Other monographs are:
Monograph 1: “Uncertainty in Measurement: the ISO Guide”
Monograph 2: “Statistical Background to the ISO Guide to the Expression of
Uncertainty in Measurement”
Monograph 3: “Traceable Measurements”
NML has produced a three-book set covering the measurement of temperature and
humidity. The set is titled: “Handbook of Temperature Measurement”, ed. R Bentley,
Springer 1998 (ISBN 981^021-12-1) and is available from DA Information Services
(61-3-9210 7777).
National Measurement Laboratory CSIRO, March 2002
Copies of the monographs are available from

National Measurement Laboratory
Bradfield Road (PO Box 218)
Lindfield, NSW, Australia
(phone: 61-2-9413 7761)
© CSIRO Australia, 2002

Printed in Sydney
Cartoons by Christophe Granet: Copyright Hagen Cartoons 2002
ii
False scales are an abomination to the Lord, but a just weight is His delight.
Proverbs 11:1
iii
iv
Contents
1. Introduction 1
1.1. Outline of the book 3
2. Air buoyancy and conventional weighing 5

2.1. Air buoyancy 5
2.2. Conventional weighing 6
2.3. Definition of conventional mass 7
2.4. Buoyancy correction for conventional mass 8
2.5. Measurement of true mass, volume or density 9
3. Mass calibration by direct comparison

3.1. Introduction
3.2. Substitution weighing
3.3. Extended substitution weighing
3.4. Checking mass measurements 13
3.5. Measuring balance sensitivity 14
3.6. Measuring the repeatability of mass comparison 15
3.7. Examples 17
4. Mass calibration using weighing schemes 19

4.1. Introduction 19
4.2. Weighing schemes 20
4.3. Analysis of weighing schemes 22
4.4. Examples of weighing scheme analysis 28
4.5. A simple scheme for a 5,3,2,1,1' decade 29
5. Uncertainty estimation for mass calibration 31

5.1. Introduction 31
5.2. Uncertainty of direct comparison 31
5.3. Uncertainties for weighing scheme 41
6. Calibration of electronic balances 47

6.1. What is an electronic balance? 47
6.2. The scope and purpose of the calibration 47
6.3. Balance calibration 48
6.4. Recording and reporting the results of calibration 53
6.5. Using a calibrated balance 64
6.6. Measuring corrections when only one calibrated weight is available 68
7. Calibration of single-pan, two-knife-edge balances 71

7.1. What is a single-pan two-knife-edge balance? 71
7.2. The scope and purpose of the calibration 71
7.3. Balance calibration 72
7.4. Recording and reporting the results 79
7.5. Using the calibrated balance 91
7.6. Measuring corrections using the comprehensive tolerance test 93
7.7. Measuring corrections using a least-squares calibration 94
8. Uncertainty for balance calibration 95

8.1. Electronic balances 95
8.2. Single-pan two-knife-edge balance 104
Definitions and symbols 115
Appendices:
A. Use of balances 121
A. 1. Environment 121
A. 2. Laboratory practice 124
B. Use and handling of weights 127
B. l. Types of weights 127
B.2. Handling of weights 128
vi
B.3. Care of weights 128
B.4. Estimation of the OIML class to old and/or special weights 129
C. Traceability of Australian mass standards 131
D. Least-squares analysis of weighing schemes 133
E. Least-squares calibration of weights installed in balances 141
E. 1. Numerical example 143
F. Sample report for the calibration of weights 147
G. Statistical tests I49
G.l. Thet-test I49
G.2. An example of the t-test 150
G,3. The F-test 151
G, 4. Examples of the F-test 151
H. Air density measurement 153
H. 1. Air density equation 153
H.2. Measurements of temperature, humidity and pressure 154
I. Mass and density limits for the OIML classification of weights 155
J. Buoyancy uncertainty in mass calibration I57
K. A brief outline of balance construction 161
K. 1. Electronic balance 161
K. 2. Single-pan, two knife-edge balances 162
L. An introduction to uncertainty calculation 165
L. l. Uncertainty calculation 165
L.2, Rounding of results 167
Bibliography 169
vii
viii
Chapter 1
Introduction
In one form or another weighing is widely used in industry and commerce. For these
measurements it is important to know the accuracy of the balances used, together perhaps with
other information such as corrections to be applied to readings. Responding to this need, the
CSIRO published The Calibration of Balances by David B Prowse [1] in 1985 as a guide to
good laboratory practice in this field.
Standard weights with high precision calibrations must be available for the accurate and
traceable calibration of balances as well as for legal metrology. A CSIRO technical
memorandum Balances and Weighing by Edwin C Morris [2] was published in 1992 to cover
the techniques required for the calibration of standard weights. Since these two publications
first appeared some major revisions have become necessary, and the present publication
addresses this.
Although mass metrology has its roots in quite ancient times, the development of
balances proceeded slowly until the 1950s. At that time, two-pan beam balances were still
most commonly used for analytic (moderate precision) and high precision measurement. Since
then they have been progressively replaced, firstly by the single-pan two-knife-edge balance
and more recently by electronic balances. The more accurate electronic balances use
electromagnetic force compensation (EFC), which balances part or all of the gravitational
force on the object being weighed with an electromagnetic force. For lower accuracy
requirements, balances using strain gauge transducers (load cells) are common and have the
advantage of greater robustness.
Over the last 30 years, single-pan two-knife-edge balances have in turn been largely
replaced by the electronic kind (both EFC and strain gauge), which are now approaching total
dominance of laboratory benches. Two-pan balances have become obsolete and, with very
few exceptions, are relegated to dusty storerooms or displayed for purely aesthetic reasons.
Other mechanical scales are still sometimes used for larger masses but are likewise being
replaced.
Over the same period, the quality of weights has also improved, though less
dramatically. There are standards, most notably OIML R111 [3], which specify every relevant
1
2 7. INTRODUCTION
characteristic (density, surface finish, magnetic susceptibility, shape, etc) so that a user can
know what to expect from weights of a given class.
Recently, all branches of metrology have benefited greatly from the widely accepted ISO
Guide to the Estimation of Uncertainty in Measurement [4]. At the same time, quality systems
are receiving increasing attention with the advent of standards such as ISO 17025. These
developments have emphasised the need for proper uncertainty analysis.
This publication takes these developments into account while combining the two earlier
publications referred to above. Most of the earlier material has been included here, a notable
exception being the chapter on two-pan balances from The Calibration of Balances. Once
again certain areas are not specifically mentioned, including the calibration and testing of
weighing devices used for trade, large capacity weighing devices such as platform scales,
weighbridges and some types of load-cell systems. However the techniques presented here
should be applicable in some form to all these areas.
There are two approaches to mass measurement, depending largely on the required
accuracy. In the first approach, measurements of relatively high accuracy, such as those
usually needed for the calibration of weights, are carried out using substitution weighing. This
involves using a balance to measure only the very small mass difference between a reference
weight and the weight to be calibrated. As a result, nothing needs to be known about the
balance apart from its sensitivity and standard deviation (repeatability), both of which may
vary with load. Issues such as linearity, hysteresis, calibration of balance weights and even off-
centre loading are normally irrelevant.
Substitution weighing is not restricted to calibrating weights and may be used whenever
a very accurate result is required. It is always necessary however for the combined mass of the
reference weights to be nearly equal to that of the object being weighed to minimise
dependence on balance characteristics.
With the second approach, one notes the balance’s zero reading z with the pan empty,
and the reading r with the object to be weighed on the pan. If these readings are in mass
units, the mass of the object is then taken to be r — z. This procedure is easier but generally
less accurate than the first approach. As full use is now being made of the balance, it is
necessary that it be calibrated if only to allow an estimate of uncertainty. A correction C,
obtained from the balance calibration report, may be applied so that the measured mass
becomes r-z + C. In this case, an accuracy approaching that of substitution weighing may be
possible.
As the first approach demands relatively little knowledge of the balance and is usually
used for the calibration of weights, it is described in the chapters on mass calibration. Using
the second approach, the mass measurement is relatively straightforward, but the
characteristics of the balance are important. It is therefore discussed in the chapters on balance
calibration.
1.1 OUTLINE OF THE BOOK 3
1.1 Outline of the book

Chapter 2 deals with issues such as air buoyancy and conventional mass, which are important
for all aspects of mass metrology. The remaining chapters deal with mass calibration (chapters
3 to 5) and balance calibration (chapters 6 to 8) respectively. More detailed or more
specialised material is given in appendices.
Chapter 3 is concerned with mass calibration by direct comparison, which is the most
widely used method. Note that the measurements of balance repeatability and sensitivity
described here differ from those required for balance calibration. Chapter 4 describes the use
of weighing schemes for mass calibration. For the same number of weights, the schemes
require more measurements than direct comparison but give a more accurate and more
consistent result. They also allow an easier check on errors in the readings or balance defects.
Chapter 5 covers the estimation of uncertainty in mass calibration.
Chapters 6 and 7 deal with the calibration of electronic balances and single-pan two-
knife-edge balances respectively. Following the pattern of The Calibration of Balances, these
chapters are very largely self-contained, though some cross-references to the other parts of the
book are inevitable. Both chapters give a description of the tests, examples of the tests, a
sample report form based on the examples, and a basic guide to using a calibrated balance.
The calculation of uncertainties associated with balance calibration is summarised briefly in
chapters 6 and 7, and is dealt with in much greater detail in chapter 8.
4
“Sometimes buoyancy can be neglected, but sometimes large errors are introduced
if it is overlooked”
Chapter 2
Air buoyancy and conventional weighing
2.1 Air buoyancy

Balances are normally thought of as instruments that compare masses, but balances of all kinds
compare force rather than mass. Suppose, for instance, that the readings of a balance are
exactly the same for a test weight T as they are for a reference weight R. The common
interpretation is that T and R have exactly the same mass, but strictly speaking we know only
that the net downward force on the balance pan is the same for Tas it is for R. If the masses of
the weights are mr and tt ir , then the equality of these forces means that
+ + (2.1)
where g is the acceleration due to gravity. Here /wyg and mpg the gravitational forces acting
on the two weights, while Xt and Xp are the sums of all other vertical forces acting on them.
Only if these other forces are negligible or equal does equation (2.1) reduce to
zwj. = (2.2)
Most of the other forces (eg stray magnetic forces, or forces resulting from air currents)
can be made negligible by careful laboratory management or balance design. However the
buoyancy force due to immersion in air can be removed only by weighing in a vacuum, and
this is usually quite impractical. Moreover buoyancy is a relatively large force; for light
materials such as wood it exceeds 0.1% of the gravitational force.
Since air buoyancy cannot be eliminated, it must be taken into account somehow, even
in weighings of quite modest accuracy. Assuming that air buoyancy alone contributes to Xt
and Xp, we can calculate these quantities by applying Archimedes' principle, which states that
the buoyancy force on an object is equal to the weight of air displaced by the object:
Xr=-P.yTg = -p.^ and X,=-p^V,g = -p^^

Pr pR
5
6 2. AIR BUOYANCY AND CONVENTIONAL WEIGHING
where Pa is the density of air, pr and Vt are the density and volume of the test weight, and Pr
and Yr are the density and volume of the reference weight. The minus signs indicate that
buoyancy is an upward force. Substituting these equations in equation (2.1), we obtain
i-£^\ = fn
ntj^ R (2.3)
Pt >
Equation (2.3) is the basis of all buoyancy calculations associated with weighing. In
very accurate weighings, such as a comparison of national secondary standards of mass with a
platinum-iridium primary standard, the buoyancy corrections shown in brackets in
equation (2.3) have to be determined as accurately as possible. This means that the densities
of the weights must be measured if they are not already known. The density of the air in the
balance case during the weighing also has to be measured either directly or indirectly. It is
usually not too difficult to measure the densities of the weights with sufficient accuracy, but
measuring the air density can pose severe problems.
2.2 Conventional weighing

In well equipped laboratories, the densities required for buoyancy corrections can be measured
if necessary. In the community at large however, such measurements are generally
impractical. Fortunately, the accuracy required of mass comparisons in commerce, industry,
and even most scientific research is not especially high so that buoyancy can be dealt with in a
way that hardly ever requires the knowledge of densities. This is done by using a weighing
convention that is known by several names, including
• Conventional weighing
• Mass on the 8.0 basis
• Apparent mass
• Weighing in air
• A conventional value of the result of weighing-in-air
For the sake of brevity, the convention is referred to in this book as conventional
weighing, and the result of a conventional weighing is called the conventional mass. When
discussing conventional weighing, mass as normally defined is sometimes called true mass in
order to distinguish it from conventional mass’. The symbols m and M are used here for true
and conventional mass respectively.
Throughout the world, conventional weighing is used for almost all mass measurement.
Even in national institutes of metrology such as NML, it is used most of the time. From the
point of view of most users, conventional weighing simply means that whether one is
calibrating weights, calibrating balances or weighing machines, or weighing some
’ Hence true mass is a measure of an object’s resistance to acceleration, or equivalently, true mass
determines the gravitational force experienced by an object.
2.2 CONVENTIONAL WEIGHING 7
commodity, one makes no allowance for air buoyancy. As will be seen below, this neglect of
buoyancy is made possible by requiring all mass standards used in conventional weighing to
have densities that lie within prescribed limits. (Nevertheless a buoyancy correction may still
be necessary at high altitudes or for mass measurements of unusually high accuracy. See
Section 2.4.)
Conventional mass differs significantly from the corresponding true mass (see Table
2.1), but this doesn't matter, as most people aren't interested in true mass. They want only a
widely accepted convention that will yield consistent weighings without the need for difficult
buoyancy corrections and that gives mass values reasonably close to true mass.
Table 2.1: The difference in parts per million between conventional mass M and true mass m. For an
object with a density of 8000 kg.m’^, conventional mass is equal to true mass.
Density (M-nt)/m
Material (kg.m’3) (ppm)
platinum 21 400 94
lead 11 300 44
brass 8 400 7
aluminium 2 700 -295
water 1 000 -1050
cork 250 _____ ^700 1
2.3 Definition of conventional mass

It follows from equation (2.3) that the relative error E that will result if buoyancy is ignored in
a weighing is given (to a very good approximation) by
J__
(2.4)
< Pt Pr a
Equation (2.4) shows that if the density of the test weight is equal to that of the reference
weight, the buoyancy error is zero. The need for buoyancy corrections in the calibration of
weights could therefore be eliminated by requiring all weights to have a certain standardised
density such as 8000 kg.m'^.
Unfortunately it is not possible to ensure that all weights have a density of exactly
8000 kg.m'\ or any other value. We can however require that the density be close enough to
the standard value that the resulting buoyancy correction can be ignored in most weighings
without significant error. Hence weights satisfying this condition can be treated as having a
density exactly equal to the standard value. It is with these considerations in mind that
conventional mass is defined as follows:
The conventional mass of an object is equal to the true mass of a weight of density
8000 kg.m'^ which at a temperature of20 X: balances the object in air of density 1.2 kg.m~\
Note that the definition applies to any object, not just weights. The specified air density
of 1.2 kg.m'^ was chosen because this value is typical for air-conditioned laboratories near sea
level and the temperature of 20°C takes account of thermal expansion, but for weights the
temperature is not at all critical. Stainless steel and similar alloys have densities close to
8000 kg.m'^, which explains the choice of this density in the definition.
The OIML (International Organisation for Legal Metrology) recommendation covering

mass standards [3] gives tables showing how close to 8000 kg.m'^ the densities of weights
must be (see Appendix I). In effect, if a weight used in conventional weighing has a relative
maximum permissible error of £ parts per million, then it must have a density p in kg.m’^ lying
within the range
8000 8000
(2.5)
£
(if £ > 60, then only the lower limit is relevant). If the density lies in this range, then the
weight can be considered to have a density of exactly 8000 kg.m’^. This in turn makes it
possible for us to verify a hierarchy of mass standards according to the above definition of
conventional mass^.
For an object of density p and true mass zw, it follows from equation (2.3) that the
conventional mass M is given by
(2.6)
0.99985P
2.4 Buoyancy correction for conventional mass

For measurements of relatively high accuracy, one may still need to add a buoyancy correction
to a measured conventional mass, particularly at higher altitudes where the air density is
significantly less than 1.2 kg.m'^. The need arises mostly in the calibration of weights, and
seldom for other mass measurements. Even for weight calibration, a buoyancy correction is
unlikely to be required at altitudes below 300 m. Near sea level, the correction may be
necessary only for weights with densities lying well outside the OIML limits or for very
acciu*ate mass measurements of light materials.
For substitution weighing (in which a test weight or other object is directly compared
with a reference weight R of the same nominal mass, see Chapter 3), the buoyancy correction
is given by
2 Strictly speaking, if the density of a test weight satisfies equation (2.5), then when the weight is
compared with a standard of density 8000 kg.m'^, a variation in the air density of 10% above or below
the value of 1.2 kg.m'^ does not lead to a variation in the result of the comparison exceeding £/4 ppm It
follows that if we simply ignore the buoyancy correction in this comparison (equivalent to taking the
density of the test weight to be 8000 kg.m’^), then the measured conventional mass of the test weight
will be in error by no more than £/4 ppm.
2.4 BVOYANCY CORRECTION FOR CONVENTIONAL MASS 9
SM = -1.2) ----- L M
\P Pr
where pa is the air density at the time of the weighing, p is the density of the test weight (or
other object), pR is the density of the reference weight, and SM is to be added to the measured
conventional mass M of the test weight (or other object) in order to obtain the correct
conventional mass. Notice that the buoyancy error in substitution weighing is zero if the air
density is equal to 1.2 kg.m or if the test and reference weights have the same density.
For mass measurement with a calibrated single-pan two-knife-edge balance for which
the corrections of individual dial readings have been determined, the buoyancy correction is
again given by equation (2.7), where pR is the density of the inbuilt weights of the balance. In
most cases, pR can be taken to be 8000 kg.m'^ if it is not known.
For mass measurement with a calibrated electronic balance, the buoyancy correction is
given by
nx rPa“F2 /7^(cal)-1.2
(SVf = ------------------------- M (2.8)
L p pR \
where Pa is the air density when M is measured, A(cal) is the air density at the time the
balance was calibrated and Pr is the density of the reference weights used to calibrate the
balance. In most cases, Pr can be taken to be 8000 kg.m’^ if it is not known.
If in doubt about the need for a buoyancy correction, substitute likely values of the
densities into equation (2.7) or (2.8) to estimate the likely correction. For both Pa and A(cal),
use the mean value of air density for the altitude of the laboratory, see Table J.l, and if pR is
not known, assume a value of 8000 kg.m'^.
If the estimated correction is significant compared with the desired uncertainty, then an
accurately determined correction will need to be applied. To calculate accurately, p must
be known (by measurement, from reference tables, or as information supplied by a
manufacturer) while both pa and /7a(cal) must be measured (see Appendix H). For the more
accurate substitution weighings Pr should be known, but otherwise it may be taken as
8000 kg.m'. None of these densities needs to be known particularly accurately; an expanded
uncertainty of 1 % would almost always be quite adequate.
2.5 Measurement of true mass for volume or density determination

Sometimes it is necessary to measure the true mass of an object or liquid sample. Since
density is defined as the true (not conventional) mass per unit volume, this mostly occurs in
measurements involving the density of relatively light materials such as water. Two common
instances are:
• Measurement of the density of a liquid or the volume of a container.

• Using Archimedes’ principle (hydrostatic weighing) to measure the density of a liquid or a
solid sample.
If Archimedes’ principle is not used, one can simply make conventional mass
measurements (corrected for buoyancy if necessary) and use equation (2.6) to obtain the true
mass from the conventional mass.
When using Archimedes’ principle, the measurements are usually analysed using a
variant of equation (2.3):
(2.9)
Pr J k Pr J
where pi is the density of the liquid in which the test object is immersed when it is balanced
against reference weight(s) of conventional mass Mr . The quantity Mr is the apparent mass of
the immersed object and it may be obtained from substitution weighing or simply read off a
calibrated balance. (In either case, a buoyancy correction may needed.) Therefore the
densities Pr , and pa must be known, as must the true mass mr of the test object, the true mass
niR corresponding to Mr and either pi or pr.
Fortunately, no significant difficulty is raised by this, except for very accurate

(± 0.002%) density or volume measurements. First of all, Pr may be taken as 8000 kg.m’^ and
hence the true mass niR will be equal to the conventional mass Mr . Secondly, Pa can be
calculated from a well known formula (see Appendix H) using air temperature and pressure
measurements of quite modest accuracy (±1 K and ±500 Pa) with the relative humidity taken
as 50%.
Usually Pt will be known or easily calculated with sufficient accuracy, but on rare
occasions it may be necessary to obtain it by iteration. Having measured Mt separately, one
then calculates hit from equation (2.6). Finally, as either pi or Pt is known, the other may be
calculated.
Chapter 3
Mass calibration by direct comparison
3.1 Introduction
This chapter deals with the calibration of weights. For this task, substitution weighing is
strongly recommended, as it is for the very accurate mass measurement of any object. The
principal advantage of substitution weighing is that one need not be concerned with balance
characteristics such as linearity, corrections to readings, masses of in-built weights, hysteresis,
etc. Therefore one need not calibrate the balance for these characteristics nor worry that they
may be drifting with time. The only quantities that must be known are the balance sensitivity
and repeatability of mass comparison, both of which may vary with load. Another important
advantage is a much clearer line of traceability for mass standards.
A requirement of substitution weighing is that the test weight or object must be

compared with a mass standard or group of standards of nearly the same mass. For weights
this means that the test and reference weights should be nominally equal. For other objects,
mass equality to within approximately 0.02% is recommended where possible.
Throughout this chapter and the next, the buoyancy correction (see section 2.4) will not
be mentioned; it will be assumed that the observer makes the correction where necessary. The
observer should also be familiar with the basic rules for handling balances and weights, which
are detailed in Appendices A and B.
Techniques for minimising the effects of a drift in balance readings are mentioned
several times in what follows, but the observer should nevertheless do whatever is possible to
minimise the drift itself. The causes are often linked to temperature, for instance a large
variation in laboratory temperature, using a balance during its warm-up period, or using
weights that are not in thermal equilibrium with the balance. See Appendix A for further
information.
11
12 3. MASS CALIBRATION BY DIRECT COMPARISON
3.2 Substitution weighing

Simple substitution weighing consists of the following three steps and is the minimum that is
required to determine the mass Mt of the test weight in terms of the mass Mr of the reference.
The quantities r;, ti etc. are balance readings to be noted at the indicated point of the
procedure. Some balances need to be arrested except when taking readings.
(i) Place the reference weight R on the pan of the balance, ry;
(ii) Remove R from the pan and place the test weight Ton the pan, tp,
(iii) Remove T from the pan.
Two extra weighings in reversed order are often added, leading to the five steps shown
below. This group of four weighings is called double substitution weighing or just double
weighing. Provided the time intervals between the readings are approximately equal, double
weighing will largely eliminate the effect of drift in the readings. (The weighing sequence T,
R, R, T is equally acceptable.)
(i) Place the reference weight R on the pan of the balance, n;
(ii) Remove R from the pan and place the test weight T on the pan, tf,
(iii) Move T well off the pan and then place it back on the pan, t2,
(iv) Remove T from the pan and place R on the pan,
(v) Remove R from the pan.
Notice that it is not necessary to zero the balance before one commences the weighings,
though one may do so. Indeed, for single-pan two-knife-edge balances it is often necessary to
offset the zero slightly before one starts (to give a small positive reading with the pan empty)
to avoid the pointer running off the scale. Once the reading rj has been taken, the balance
must not be tared or zeroed, nor should any tare weights or built-in weights be moved.
Some balances have a high resolution range (with ten times the normal resolution)
covering perhaps one-fifth the total range of the balance, and sometimes the high resolution
range can, by retaring, be moved anywhere throughout the load range of the balance. In such
cases one may take advantage of the higher resolution by retaring the balance after placing
weight R on the pan in step (i).
The first two weighings give

Mj. = Mj^ -I- S{t^ -Tj) (3.1)
and from the last two,
=^R -''2) (3.2)
where S is the sensitivity reciprocal, defined as the mass of a small weight added to the pan of
the balance divided by the resulting change in the balance reading (see Section 3.5). (For a
well adjusted balance, S should be very nearly equal to 1.) From these equations, the mean
value of Mt obtained from double weighing is
M^ = M + 5(Zi + Zj — ^*1 — ^2)/ 2 (3.3)

3.3 EXTENDED SUBSTITUTION WEIGHING 13
3.3 Extended substitution weighing

Extended substitution weighing consists of repeating a double weighing two or more times to
give the readings
ri ti t2 r2 rs tj t4 ... r„ (3.4)
where n is an even numbe?. The measured mass of T is
n ___
=Mj,+S{t-r} {3.5}
1=1
where {t - r} is the mean value of the measured mass differences (r, - r,). Once again, weights
should be lifted well off the pan between readings and the time intervals between readings
should be approximately equal within each group of four.
Extended weighing reduces the uncertainty of the mass comparison and also allows the
repeatability of mass comparison of the balance (see Section 3.6) to be determined (see
Example 3.7.2}. The value of so obtained can be compared with previous determinations of
s„ (see section 3,4) as a check on the measurement.
3.4 Checking mass measurements

A measurement of Mt using simple substitution or double weighing may be sufficient to give
the mass value with the desired uncertainty, but it is advisable to check measurements to
ensure that no error has been introduced. This is particularly so for simple substitution as a
single measurement can easily contain an undetected operator error. For this reason, simple
substitution is not recommended for direct comparison calibration and should be used only in
weighing schemes (see Chapter 4) that include a measurement check.
The simplest check is to compare the results from each half of a double weighing; in
other words, to compare S{ti — ri} with S{t2 -r^)- These two quantities should differ by no
more than 3.2^^, where Sm is the repeatability of mass comparison for the balance (see
Section 3.6). For the more critical calibrations, some kind of check weighing is recommended.
Two approaches to check weighing are mentioned here:
(a) The mass of a check weight C is measured at regular intervals using the same
balance and reference weight as the measurements to be checked. A record of the measured
mass values Mq is kept, and ideally the record should contain at least 5 previous
measurements. The results are analysed using a t-test (see Appendix G, where the quantity Q
in Section G.l becomes Mq }. An example is given in Section G.2.
In the absence of significant drift, n could be odd.

(b) The repeatability of mass comparison for the balance (at one or more loads) is
determined in the course of mass measurements by carrying out extended substitution
weighing. Alternatively could be measured separately from the measurements to be
checked, as described in Section 3.6.
For the purpose of the check, at least one previous measurement is needed, though a
record of several recent measurements is better. If the latest value obtained for is
significantly greater than the earlier ones, the operator is alerted to a possible problem. To
determine whether an increase in balance repeatability is significant, the F-test is used. See
Appendix G for details. See also the first example in Section G.4.
If a balance is limited by its resolution, the measured value of may be somewhat

unreliable and may even be zero. However this checking procedure can still be used provided
a zero value of is not included in the record of previous measurements. When the present
value of Sj„ is zero, the balance can be taken to have passed the test.
Should check (a) or (b) fail (i.e., the value of t or F is too large), then a potential problem
is indicated and must be investigated. Points to consider are
• The balance. Has it been allowed to warm up properly? Has its performance deteriorated?
• Laboratory conditions. Are there large temperature variations or air-conditioning
problems? Is the relative humidity reasonably steady in the range 40 - 60%RH?
• The weights, including the reference. Are they free of dust etc and at the same
temperature as the balance? For check (a), it is possible that weight C has drifted and there
may be nothing wrong with the measurements being checked.
• The operator. Is the operator new to this work. Has something been overlooked or a
reading taken incorrectly?
None of these checking procedures is perfect. False positive and false negative
indications are both possible.
3.5 Measuring balance sensitivity

Once the balance has been adjusted at the location where it is to be used, the sensitivity
reciprocal 5 of a modem balance usually remains close to unity. Any variation with time or
with load is normally insignificant for substitution weighing. For such balances S need not be
measured very frequently; once per year should be sufficient. Some balances may need to be
checked more often, and it is up to the operator to choose a frequency based on the history of
the instrument. Where there is no history, the sensitivity should be measured on the first three
occasions that the balance is used.
The sensitivity reciprocal can be measured by adding a small weight of mass Mêns to the
balance pan after the last weighing of a sequence. Alternatively, separate weighings may be
carried out in the following way:
3.5 MEASURING BALANCE SENSITIVITY 15
(i) Place weight A of mass on the pan, rf,

(ii) Add the sensitivity weight, r2.
Weight A must not be touched between the weighings, nor should any built-in or tare
weights be moved after reading rj has been taken. The balance may need to be arrested while
placing A on the pan, but not when the sensitivity weight is added. The sensitivity reciprocal
at load M^ is given by
where sens = |r,

I 2
- r,11I.
Ideally should be at least 20 times larger than the largest mass difference AM likely
to be measured on the balance (see Section 5.2.1). (Example: If a 1 kg balance is used to
calibrate Fi and F2 weights, then M,ens should be at least 20 times the maximum permissible
error of 15 mg for a F2 kilogram i.e. M^^ns should be at least 300 mg.) A mass in the range 20
to 200 times AM is probably best. For single-pan two-knife-edge balances however, Mêns
cannot exceed the range of the optical scale.
Since S does not have to be accurately known, variations with load of a few percent are
often permissible for the calibration of standard weights. Modem balances usually show much
less variation than this, so S will normally have to be measured at only one load, such as half
maximum load.
3.6 Measuring the repeatability of mass comparison

The standard deviation of a balance is a measure of its repeatability. Even if it is never used to
check measurements (see Section 3.4) or to check balance performance, the standard deviation
must be measured and recorded as it contributes to the overall uncertainty of mass
measurements. For substitution weighing, the most relevant expression of balance standard
deviation is the repeatability of mass comparison s^, which is the standard deviation of a
single mass comparison S(z,. -rj, see equations (3.1), (3.3) and (3.5). Note that a different
expression of balance standard deviation, called the repeatability of measurement is used in
Chapters 6, 7 and 8.
Normally the repeatability of mass comparison is measured every six months, but the
balance history and usage may lead an operator to choose a period as long as a year or as short
as a week. The need for a measurement is indicated whenever odd or inconsistent results are
obtained.
One can obtain a value of Sm from a set of measurements such as an extended

substitution weighing (see the example in Section 3.7.2} or it can be measured separately from
any mass measurement. The latter course is recommended unless the balance is used very
infrequently (less than once in six months). Whichever approach is used, s^ should be based
on at least nine degrees of freedom unless the accuracy required for mass measurement is very
easily obtained from the balance. For the number of degrees of freedom is the number of
measurements minus one (see ‘Definitions’ at rear of book). Four degrees of freedom are
recommended as a minimum.
To measure separately, place a weight on the pan, note the reading, and then remove
the weight (arresting the balance as required). Do this 20 times to obtain the readings ri, r2,
rs, ... r2o. Calculate the 10 mass differences
Si = S{ri - r2}, S2 = S(r3 - r4}, S3 = S(r5 - r6}, ... Sio = S{rig - r2o}
and take to be the standard deviation of these quantities. Since ten measurements have been
made, this value of will have nine degrees of freedom, and if the time intervals between the
readings are the same, it will not be affected by a moderate drift. The example in Section 3.7.3
below illustrates this procedure.
Note: The standard deviation s of n quantities Sj, S2,... can be calculated from either of the two formulae:
where n is the number of measurements of <J(10 in the above example), and S is the mean value of the
For most modem balances, it would be sufficient to measure the repeatability of mass
comparison at maximum load only. Even if a consistent variation were discovered, it should
seldom be necessary to measure at more than half load and maximum load. (In this case
adopt the half-load value up to half load, and the full-load value beyond that.)
When obtaining from an extended substitution weighing, it is very desirable that drift
be eliminated from the balance readings or twice as many weighings will be required to
achieve nine degrees of freedom.
• In the absence of noticeable drift, complete 10 mass comparisons (n= 10 in equation 3.4).
The standard deviation of the mass differences 5'(Z, - r,) is then a measurement of Sm with nine
degrees of freedom.
• If drift is an unavoidable problem, carry out 20 mass comparisons and combine the
differences in pairs to give 10 quantities such as the second term on the right-hand side of
equation (3.3). In this case, the standard deviation of these quantities multiplied by ^2 is a
measurement of with nine degrees of freedom.
Some balances are limited by their resolution. When a weight is placed on such a
balance several times, the same reading is likely to be obtained repeatedly (in the absence of
drift). Thus most if not all of the S will be zero, and the calculated standard deviation may be
somewhat unreliable and could even be zero. The true repeatability of the balance is not zero
3.7 EXAMPLES 17
of course; the limited resolution just makes it too small to be measured. The value of Sm
obtained from equation (3.7) should nevertheless be recorded as the measured repeatability of
mass comparison for the balance, even if it is zero.
3.7 Examples
3.7.1 Double substitution weighing and sensitivity check
An E2 reference weight R has a mass of 1000.001 1 g. It is used to calibrate a Fj test weight T

and the following balance readings are obtained (after taring at 1000 g):
ri 0.000 9 g
t] -0.000 3 g
t2 -0.000 2 g
rz 0.001 0 g
Since the sensitivity of the balance has not been checked for some time, a sensitivity weight
having a mass of 0.019 955 g is then placed on the balance (without moving the reference after
noting reading rj) to give the reading
0.020 9 g
From equation (3.6), the sensitivity reciprocal is given by

5= =1.0028
0.0209-0.0010
and since the permissible uncertainty for a sensitivity reciprocal is a few percent, S will be
taken as exactly 1.0. From equation (3.3) one then finds
M 7. =1000.0011 + (- 0.0003 - 0.0002 - 0.0009 - 0.0010)/2 = 1000.0011 - 0.0012 = 999.9999 g
3.7.2 Extended substitution weighing and a calculation of repeatability of mass

comparison
A weight T is calibrated against a reference R having a mass of 199.999 94 g. Extended

substitution weighing is used with the following results, in grams:
rj 199.999 89 r3 199.999 87 rs 199.999 90

ti 200.000 25 ti 200.000 25 ts 200.000 27
t2 200.000 27 t4 200.000 22 t6 200.000 21
r2 199.999 92 r4 199.999 89 r6 199.999 90
From equation (3.5), taking S to be 1,

M^ =199.999 94 + ^(0.000 36 + 0.000 35 + 0.000 38 + 0.000 33 + 0.000 37 + 0.000 37)
= 199.999 94 + = 199.999 94 + 0.000 36 = 200.000 30 g

6
18 3. MASS CALIBRA TION BY DIRECT COMPARISON
Since the drift in the readings is very small, one can calculate the repeatability of mass
comparison (with 5 degrees of freedom) from the six mass differences 0.000 36, 0.000 35,
0.000 38, 0.000 33, 0.000 37 and 0.000 37 g. The mean mass difference has already been
found to be 0.000 36 g, so the six quantities (j. to be substituted into equation (3.7a) are
0.0, -0.000 01, 0.000 02, -0.000 03, 0.000 01 and 0.000 01 g. The resulting value of is
0.000 018 g.
3.7.3 Repeatability of mass comparison (Sm)
The repeatability of mass comparison for a 1 kg balance (with S = 1) is measured at full load
by placing the same weight on the pan 20 times. The readings, in grams, obtained after taring
at 1 kg are
0.000 6 0.000 7 0.000 6 0.000 4 0.000 7 0.000 6 0.000 6 0.000 8 0.000 8

0.000 8 0.000 7 0.000 8 0.000 9 0.000 8 0.000 8 0.001 0 0.000 9 0.000 8
0.000 9 0.000 9
There seems to be a slight drift in the results, but provided the readings are taken at equally
spaced intervals, this does not matter when the standard deviation is measured by using the
mass differences Bj, S2, etc shown in Section 3.6. The mass differences are
-0.000 1 0.000 2 0.000 1 -0.000 2 0.000 0 -0.000 1 0.000 1 -0.000 2 0.000 1

0.000 0
The mean is -0.000 01 and the standard deviation calculated from equation (3.7a) is
0.000 14 g.
Chapter 4
Mass calibration using weighing schemes
4.1 Introduction
A weighing scheme consists of several mass comparisons carried out using the techniques of
Chapter 3. Three or more weights are involved, at least one of which must be a reference
while one (or more) may be a check weight. The rest will be test weights. Weighing schemes
are the norm for class Ej calibrations, optional for E2 and Fi calibrations, but are not normally
used for class F2 and below.
In general, schemes have several advantages:

(a) For a set of weights, a more consistent calibration will result.
(b) Some schemes make very little use of reference weights, thereby giving them some
protection from mass changes due to usage. Indeed it is possible to calibrate an entire
set of weights using just one reference.
(c) Checking the measurements is easier. It is normal, for instance, to incorporate a check
weight (see Section 3.4) in the scheme. This may allow the use of single substitution
weighing, at least halving the number of readings per mass comparison.
(d) Some schemes give a value for a quantity such as the balance repeatability of mass
comparison s^, which can be used to check the weighings using a statistical test (see
Section 3.4 and Appendix G).
However there are also some potential disadvantages:

(a) It is sometimes necessary to place several wei^ts on the balance pan. This may be
tricky and the operator may have to be careful to avoid error due to off-centre loading.
(b) More mass comparisons may be necessary (compared with direct comparison) to
calibrate the same number of test weights.
(c) The analysis of a weighing scheme to obtain the test masses and their uncertainties is
generally quite complicated and must be done by computer. However modem
spreadsheet programs can perform this task quickly and easily. Such a program can be
prepared using instructions given later in this chapter.
In most circumstances, one may choose between schemes and direct comparison,
but for two situations weighing schemes must be used:
19
20 4 MASS CALIBRA TION USING WEIGHING SCHEMES
(a) Calibrating the masses of the in-built weights of most single-pan two knife-edge
balances (see Appendix E).
(b) Calibrating a set of weights when one has only one reference weight.
4.2 Weighing schemes

Over the years, a large number of schemes have been proposed by many authors, but only
three are described in detail here. The Closed Cycle is mostly used for calibrating two test
weights (or one test and one check weight) in terms of a single reference, where all three
weights have the same nominal mass. The Simple Decade and the Extended Decade schemes
are for calibrating weights belonging to a weight set.
The mass comparisons that make up the schemes are listed below using equations such
as
A-‘r B — C + D + E + a (4.1)
This equation indicates that weights A and B (as a group) are compared with weights C, D and
E (as a group) to obtain the mass difference a. To measure a, one uses substitution weighing
as described in Sections 3.2 and 3.3, where R now stands for the group C + D + E, while T
stands for the group A + B. Thus equation (3.5) may be rewritten
A+B C+D+E ~ (4.2)
4.2.1 Closed cycle
The closed cycle belongs to a family of weighing schemes designed for weights of nominally
equal mass. The version given in equations (4.3) shows a single reference and two test
weights. However weight T2 could be a check weight.
R = 7/ + a;
7/ = T2 + a2 (4.3)
T2 = R + as
Ideally, 2^ = 07 + 612 + ^2^ would be zero, so the magnitude of this quantity, referred to as
“Sum a”, provides an immediate check on the measurements (see Chapter 5 for further
details).
The simplicity of the closed cycle allows the masses Mn and Mt 2 of the test weights to
be calculated from reasonably simple equations:
4.2 WEIGHING SCHEMES 21
Mti - Mr + {-2a I + a2 + a3)/3 (4.4a)

Mt 2 —Mr —{—ai - a2 +lai)/?) (4.4b)
(cf. equations D.7). Alternatively the spreadsheet described in Section 4.3 could be used to
calculate the test masses. If T2 is a check weight, the mass Mt 2 would be used to check the
outcome of the cycle as described in Section 3.4.
4.2.2 Simple decade and extended decade
In weight sets, the weights are grouped into decades, such as 5,2,2’, 1; 5,2,1,!’; 5,3,2,1; etc,
where 5, for instance, may stand for a 5 kg, 500 g, 50 g, ... or 5 mg weight depending on the
decade in question. The following discussion applies to the 5,2,2’, 1 and 5,2,1,!’ structures as
these are by far the most common.
Note. A prime is used to indicate a second weight having the same nominal mass as another weight of the
decade. Thus a 5,2,2’, 1 decade contains two weights of nominal mass 2.
In decade weighing, the test weights are not compared directly with weights from a
reference set. Instead, a weight from the test set called the head weight acts as the reference
weight. Apart from the head weight and test weights, a decade normally includes a check
weight.
The head weight of a decade is the smallest weight of the next decade up; thus 100 g is
the head weight of the 50 g to 10 g decade. The check weight is an additional weight that is
added to make the decade up to 5,2,2T,1 ’. For the 5,2,2',1 structure, the check weight would
be a 1' weight. The head weight normally belongs to the same set as the weights being
calibrated, while the check weight is from some other set. The mass of the head weight must
be known in advance, as it is effectively the reference weight for the decade.
To see how decade weighing may be used to calibrate a complete set of weights, it is
best to consider an example. Suppose a set of test weights has the 5,2,2’, 1 structure and
extends from 20 kg to 1 mg. Since the 20 kg and 10 kg weights do not form a complete
decade they must be measured by direct comparison or by a closed-cycle scheme. In either
case they are compared with weights from a reference set.
The 10 kg test weight then becomes the head weight for the calibration of the 5 kg to
1 kg decade. The check weight for this decade is any suitable 1 kg weight. After the 5 kg to 1
kg calibration is complete, the 1 kg test weight becomes the head weight for the next decade,
and so on. Notice that, once the 10 kg test weight has been calibrated, the reference set is no
longer needed (unless the check weights are taken from it).
22 4 MASS CALIBRATION USING WEIGHING SCHEMES
The simple and extended decades are given by equations (4.5) and (4,6) respectively.
10 — 5 + 2 + 2’ + 1 +<3;
5—2 + 2’ + !’ +6^2
2=2’ + (4.5)
2’ = 1 + 1’ + a4
1 = 1’ +
10 = 5 + 2 + 2’ + 1 + ai
10 = 5 + 2 + 2’ + 1’ + a2
5 = 2 + 2’ + 1 + as
5 =2 + 2’+!’ + <24 (4.6)
2+1 = 2’+ 1’ + <25
2+1’ = 2’ + 1 + 0(5
2 =1 + 1’ + <27
2’ =1 + 1’ + as
7
where 10 is the head weight and either 2’ or 1’ is the check weight. Mass values for the test
weights are obtained by analysing the available data, which consist of the quantities aj to as, or
aj to as, plus the mass value of the head weight.
For the simple decade, this gives five equations in five unknowns, so mass values for the
test and check weights may be obtained from the solution of simultaneous equations. For the
extended decade, there are eight equations in five unknowns (i.e., three degrees of freedom). In
such cases, the mass values that best fit the data must be determined.
4.3 Analysis of weighing schemes

Weighing schemes are best analysed using a spreadsheet program, and this section gives
instructions for writing such a program. These instructions are for EXCEL, so some formulae
may need modification for other spreadsheet programs.
No attempt is made to justify or explain the mathematics except to the very limited
extent that is necessary for writing the program. The instructions are thus intended for the
reader with a reasonable understanding of spreadsheet programming, but not necessarily any
understanding of matrix algebra or linear regression. For readers wishing to know more, the
mathematical background to what follows is given in Appendix D.
Figure 4.1 shows how the left-hand side of the spreadsheet could be set out. (Readers
may choose their own lay-out, but if it is not the same as that shown in Figures 4.1 and 4.2,
4.3 ANALYSIS OF WEIGHING SCHEMES 23
some of the cell addresses used in formulas given below may need to be modified.) The
closed cycle scheme is used in this illustration in order to fit it conveniently on to a page, but
similar lay-outs are easily devised for decade schemes, see Section 4.3.1 for details.
Cells Cl9, C20, DI9, D20, El9, E20, C23 and C25 contain formulas (given below)
and the indications “0” and “#NUM!” are due to the fact that no input data has been
entered. The entries in cells B25 and C25 are not relevant for closed cycles and should be
omitted. (They are shown in Figure 4.1 because they are required for the extended
decade scheme described later.)
Figure 4.1: The left-hand side of a spreadsheet for analysing closed cycle weighing schemes.
The input data are entered into the shaded cells of Figure 4.1. The mass of the reference
weight R will be entered at E4 while the results of the three mass comparisons will be placed
in DI 1, D12 and DI3. The standard uncertainties and corresponding degrees of freedom to be
entered in E5, E6 and cells El 1 to Fl3 will be discussed in chapter 5. Here we are interested
only in calculating the values of the test masses, not their uncertainties. This can be done by
simply putting all the input uncertainties equal to 0.01 mg and all the degrees of freedom equal
to 10. (Any reasonable values will do.)
The right-hand side of the spreadsheet, shown in Figure 4.2, shows the arrays in which
most of the calculations are carried out. The array labelled X (shown in cells H3 to J6) is
known as the design matrix as it describes the structure of the weighing scheme. The design
matrix for a closed cycle is shown in Figure 4.2, while those for decade schemes are given in
Tables 4.1 and 4.2.
This array should be named X. To do this, highlight cells H3 to J6, move the cursor into
the name box (at the far left of the formula bar) and click, type the letter X, then press [Enter].
The array Va has the same number of rows as columns, namely the number of mass
comparisons plus one. So Va is a 4 x 4"* array for a closed cycle. The quantity in cell L3 is the
square of the standard uncertainty (i.e., the variance) of the reference mass, so enter the
formula =E5^2 in this cell. The other cells down the diagonal of the array contain the
variances of the mass comparisons, so enter =E1 K2 in M4, =E12^2 in N5 and =E 13'^2 in 06.
Enter zero in every other cell of Va. Highlight the array and name it Va, using the procedure
of the previous paragraph.
The first quantity in array a is the mass of the reference weight in grams, so enter =E4 in
cell Hll. The lower cells contain the results of the mass comparisons in grams, so enter
=D11/1000 in H12, =D12/1000 in H13 and=D13/1000 in H14, Name this array a.
Figure 4.2: The right-hand side of a spreadsheet for analysing closed cycle weighing schemes.
Array VM has 3 rows and 3 columns (for a closed cycle), and the formula to be entered
in this array is
=MINVERSE(MMULT(TRANSPOSE(X),MMULT(MINVERSE(Va),X)))
but a special formula entry procedure is required: Firstly highlight the cells where the array is
to be located (cells J11:L13 in this example). Then type the formula and press [Control] +
A yxz array has y rows and z columns

[Shift] + [Enter]. (That is, press [Enter] while holding down [Control] and [Shift].) Name this
array VM.
Array V has 3 rows and 4 columns. The formula to be entered is

=MMULT(VM,MMULT(TRANSPOSE(X),(MINVERSE(Va)))
and this is entered using the special procedure of the previous paragraph. Name this array V.
Array M has 3 rows and the formula to be entered (using the special procedure) is
=MMULT(V,a).
Name this array M. Arrays Rs and SSR are not needed for a closed cycle scheme.
The remaining arrays are needed for calculating degrees of freedom using the Welch-
Satterthwaite formula (see Chapter 5). Array Vu has the same number of rows and columns as
V, and special care is required for entering this array as each column requires a different
formula. Each column of Vu is found by multiplying the corresponding column of V by one
of the input standard uncertainties and then raising this product to the fourth power. For the
first column, the multiplying factor is the standard uncertainty of the reference mass; for the
second column, it is the standard uncertainty of the first mass comparison; for the third
column, it is the standard uncertainty of the second mass comparison, and so on.
For the first column of Vu the formula is therefore =(E5*N11:N13)M and it is entered
using the special procedure, i.e. highlight the first column of Vu, type in the formula and press
[Control] + [Shift] + [Enter]. For the remaining columns, formulas are =(E11*011:013)^4,
=(E12*P11:P13)M and =(E13*Q11:Q13)M. Highlight all of Vu and name it Vu.
Array N has 4 rows and contains the reciprocals of the input degrees of freedom with
=1/E6 entered in the first row, and =1/F11, =1/F12 and =1/F13 in the subsequent rows. Do
not use the special procedure for entering these formulas. Name this array N.
Array D has 3 rows and the formula to be entered (using the special procedure) is
=MMULT(Vu,N). Array DF has 3 rows and each row is found by dividing the square of a
diagonal element of VM by the corresponding row of D. Thus the formulas for DF, starting
with the first row, are =J1K2/O26, =K 12^2/027 and =L13^2/028. Do not use the special
procedure for entering these formulas.
It is now possible to write the formulas required on the left-hand side of the spreadsheet.
The first cell of array M (i.e.. Hl 9) gives the mass of the reference and hence should be equal
to cell E4. The lower cells give the masses of the test weights. So enter the formula =H20 in
cell Cl9, and =H21 in C20. The standard uncertainties of the masses of the test weights are
the square-roots of the diagonal elements of array VM, starting with the second. So enter the
formula =SQRT(K12) in cell D19 and =SQRT(L13) in cell D20. The degrees of freedom for
these standard uncertainties are in array DF, starting at the second row. So enter the formula
=Q27 in cell E19 and =Q28 in cell E20. Finally, the formula for cell C23 is =SUM(D11:D13).
Cells Cl9 and C20 should be formatted to show six figures after the decimal point, while
cells DI9 and D20 show four and cells El9 and E20 show none. The other cells in Figure 4.1
may be left unformatted. There is no need to format any cell in Figure 4.2, but the
programmer may like to do so.
4.3.1 Adapting the spreadsheet to particular weighing schemes
The spreadsheet described in Section 4.3 is suitable for the closed cycle scheme, but may be
readily adapted to any scheme with only one reference weight. For the decade schemes given
in this chapter, the following modifications are needed. (This list should be read in
conjunction with the instructions in Section 4.3.)
• The “Reference” becomes the “Head weight”

• The “Mass Comparisons” section is extended to show either five (simple scheme) or
eight (extended scheme) comparisons. The entries in the “Comparison” column would be
taken from equation (4.5) or (4.6).
• The “Results” section is extended to show five test masses (5, 2, 2', 1 and 1’) instead
of T1 and T2.
• The “Sum a” section is not required. The “Decade Std Devn” section is required for
the extended decade scheme only.
• Array X is as given in Table 4.1 or Table 4.2.
• Array Va is 6x6 cells (simple scheme) or 9x9 cells (extended scheme). The top left-
hand cell is the variance of the head weight. As for the closed cycle scheme, the other cells on
the diagonal of Va are the variances of the mass comparisons. All other cells in the array are
zero.
• Array a has six rows (simple scheme) or nine rows (extended scheme). The first row
contains the mass of the head weight in grams while the other rows contain the results of the
mass comparisons in grams.
Table 4.1: Design matrix X for the simple decade scheme

1 0 i 0 i 0 0 i 0
1 -1 i -1 I -1 -1 1 0
0 1 i -1 I -1 0 I -1
0 0 I 1 i -1 0 I 0
0 0 0 1 -1 I -1
0 0 I 0 i 0 1 i -1
Table 4.2: Design matrix X for the extended decade scheme

1 0 i 0 I 0 0 0
1 -1 I -1 i-1 -1 0
1 -11 -1 I-1 0 -1
0 1 -1 i-1 -1 0
0 1 i -1 1-1 0 -1
.....6 0 1 i-1 1 -1
. ... 6 0^1 i-1 -1 1
0 0 1 0 -1 -1
0 0 10 11 -1 -1
• For both schemes, array VM is 6x6 and contains the same formula as given in
Section 4.3.
• Array V is 6x6 (simple scheme) or 6x9 (extended scheme) and contains the same
formula as given in Section 4.3.
• For both schemes, M has six rows and contains the same formula as given in
Section 4.3.
• The array Rs is not required for the simple scheme. For the extended scheme, it has
nine rows and contains the formula
=MMULT(X,M)-a
which is entered using the special procedure described in section 4.3 for array VM. Name this
array Rs.
• The quantity SSR is not required for the simple scheme. For the extended scheme,
SSR is a single cell containing the formula
=MMULT(TRANSPOSE(Rs),Rs)
which is also entered using the special procedure described in section 4.3.
• Array Vu is 6x6 (simple scheme) or 6x9 (extended scheme). For each column, the
array formula to be entered is as described in Section 4.3. However most cell addresses will
inevitably be different to those given for the closed cycle scheme. If, for instance, the standard
uncertainty of the second mass comparison is in cell E12 while the third column of V is in
cells Q14:Q19, then the formula to be entered in the third column of Vu is
=(E12*Q14:Q19)^4.
• Array N has 6 rows (simple scheme) or 9 rows (extended scheme). As in Section 4.3,
the first row contains the reciprocal of the degrees of freedom for the head weight standard
uncertainty, while the following rows contain the reciprocals of the degrees of freedom for
each mass comparison in turn.
• For both schemes, array D has 6 rows and contains the same formula as given in
Section 4.3.
• For both schemes, array DF has 6 rows where each row is found by dividing the
square of a diagonal element of VM by the corresponding row of array D (as in Section 4.3,
though most cell addresses will be different).
• As for the closed cycle, array M contains the masses of the test weights, starting at the
second row. Thus the second row shows the mass of the 5 weight, the third row shows the
mass of the 2 weight, and so on. As in section 4.3, carry these five masses into the second
column of the “Results” section.
• As for the closed cycle, the diagonal elements of array VM are the variances of the test
weights, starting at the second diagonal element. So the second diagonal element is the
variance of the 5 weight, etc. As in section 4.3, carry the square-roots of these five diagonal
elements into the third column of the “Results” section.
• As for the closed cycle, array DF (starting at the second row) contains the degrees of
freedom required in the fourth column of the “Results” section.
• For the extended scheme, the formula for the decade standard deviation is
=SQRT(CA/3)*1000, where CA is the cell address of the quantity SSR (L19 in Figure 4.2, for
example).
4.4 Examples of weighing scheme analysis

The following examples can be used to check spreadsheet programs written to analyse
weighing schemes.
4.4.1 Closed-cycle example
An E2 reference is used to calibrate two Fi test weights. The reference mass is 500.000 45 g
and the results of the mass comparisons are -1.1 mg, 1.4 mg and -0.1 mg. As uncertainties
are not to be calculated, put the standard uncertainties of the reference mass and mass
comparisons all equal to 0.01 mg. Put the degrees of freedom equal to 15 for the reference
mass and equal to 10 for the three mass comparisons.
The spreadsheet should give 500.001 62 g for the mass of T; and 500.000 28 g for the
mass of T2, while Sum a should be 0.2 mg (see Chapter 5 for the significance of Sum a). The
standard uncertainties shown for both test masses should be 0.0129 mg with 31 degrees of
freedom, and although these figures have no significance (as the input uncertainties and
degrees of freedom were chosen arbitrarily) they can be used to check the spreadsheet.
4.4.2 Simple-decade example
The Fl head mass is 100.000 29 g, and the results of the mass comparisons (in mg) are
0.455 -0.125 -0.170 -0.060 0.270
As uncertainties are not to be calculated, put the standard uncertainties of the reference mass
and mass comparisons all equal to 0.01 mg. Put the degrees of freedom equal to 15 for the
reference mass and equal to 10 for the five mass comparisons
The spreadsheet should give 49.999 72 g for the mass of the 50 g weight, 19.999 88 g
for the 20 g weight, 20.000 05 g for the 20' g weight, 10.000 19 g for the 10 g weight, and
9.999 92 g for the 10' g weight. The standard uncertainties shown for the test masses should
be 0.01 mg with 44 degrees of freedom for 50 g, 0.0072 mg with 20 degrees of freedom for
20 g, 0.0057 mg with 33 degrees of freedom for 20' g, and 0.0069 mg with 26 degrees of
freedom for both 10 g and 10' g. Although these standard uncertainties and degrees of
freedom have no significance (as they were chosen arbitrarily) they can be used to check the
spreadsheet.
4.4.3 Extended-decade example
The Fl head mass is 100.000 29 g, and the results of the mass comparisons (in mg) are
0.455 0.750 -0.370 -0.125 0.105 -0.415
-0.195 -0.060
4.3 ANAL YSIS OF WEIGHING SCHEMES ANAL YSIS 29
As uncertainties are not to be calculated, put the standard uncertainties of the reference mass
and mass comparisons all equal to 0.01 mg. Put the degrees of freedom equal to 15 for the
reference mass and equal to 10 for the eight mass comparisons
The spreadsheet should give 49.999 72 g for the mass of the 50 g weight, 19.999 886 g
for the 20 g weight, 20.000 037 g for the 20' g weight, 10.000 175 g for the 10 g weight, and
9.999 914 g for the 10' g weight. The decade standard deviation should be 0.0168 mg (see
Chapter 5 for the significance of this quantity). The standard uncertainties shown for the test
masses should be 0.0071 mg with 44 degrees of freedom for 50 g, 0.0045 mg with 64 degrees
of freedom for 20 g and 20' g, and 0.0043 mg with 62 degrees of freedom for 10 g and 10' g.
Although these standard uncertainties and degrees of freedom have no significance (as they
were chosen arbitrarily) they can be used to check the spreadsheet.
4.5 A Simple scheme for a 5,3,2,1,1’ decade

For a simple 5,3,2,!,!’ decade scheme, the mass comparisons and design matrix are
10 — 5 + 3 + 2 +0;
1 0 0 0 0 0
5 = 3 + 1 + 1’ +<22 1 -1 -1 -1 0 0
3 = 2+1 +02 0 1 -1 0 -1 -1
2 = 1 + 1’ + 04 0 0 1 -1 -1 0
0 0 0 1 -1 -1
1 = 1’ +<25 0 0 0 0 1 -1
A spreadsheet very similar to that described in section 4.3.1 for the simple 5,2,2’,1,1’ decade
scheme can be used for the analysis of this scheme.
Placing several yveights on a balance can be tricky

30
Chapter 5
Uncertainty estimation for mass calibration
5.1 Introduction
This chapter shows how to calculate the uncertainties of masses that have been measured using
the techniques described in Chapters 3 and 4. The reader is referred to Appendix L for a brief
introduction to uncertainty calculation and to other sources [4, 5] for a more complete
treatment. As far as possible, symbols and terminology in this chapter follow references [4, 5].
Uncertainty calculations for direct comparison calibrations are shown first. The
application to weighing schemes is then discussed.
5.2 Uncertainty of direct comparison
5.2.1 The model
For the purpose of calculating uncertainties, we need an equation, called the model, which
expresses the measurand as a function of the parameters used in its determination. For direct
comparison the measurand is Mt , the mass of the test weight being determined, and the
equation that best serves as a model is obtained by combining equations (3.5) and (3.6):
M ---------
(5.1)
sens
where Mr is the mass of the reference weight, Msens is the mass of the small weight that was
used to measure the sensitivity reciprocal of the balance (see Section 3.5), Ar,ens is the change
in the balance reading that occurred during the measurement of sensitivity reciprocal (see
Section 3.5), and (Z — r) is the mean value of the measured mass differences (z, — r,).
Note that equations (3.1), (3.2) and (3.3) are just special cases of equation (3.5) with
n = 1 and n = 2, respectively. Notice also, that Mêns is approximately equal to Ardens because
the sensitivity reciprocal is approximately equal to one.
31
32 5 UNCERTAINTY ESTIMA TION FOR MASS CALIBRA TION
An examination of equation (5.1) shows that the sources of uncertainty that affect the
measurement of Mt are
1. The calibration uncertainties in Mr and Msens
2. The uncertainties due to the instability of Mr and Msens
3. The repeatability of mass comparison for the balance, affecting both /N^sens and (Z - r)
4. The resolution of the balance, affecting Ibsens and (Z - r}
5. The buoyancy uncertainty, since the buoyancy correction of equation (2.7) is not included
in equation (5.1)
6. The rounding of Mt in the final statement of results
These are the uncertainty contributions most likely to occur in practice. A more
comprehensive listing is found in reference [7], though the vast majority on this list are
significant only under unusual circumstances.
Equation (5.1) is also used to obtain the sensitivity coefficients c, needed to calculate the
effect on the measurand Mt of the above uncertainties in the quantities on the right of that
equation:
For Mr Ci=\
For Msens c
' J^sens
(5.2)
For Arsens e
'
-
^sens
For (t-r) c^
Note: These sensitivity coefficients are found by taking the partial derivative of Mt with respect to the relevant
quantity. To find the sensitivity coefficient for for instance, one must evaluate----- —. From equation (5.1),
sens
3My M ----- ft — r)
---- - —~ , since Mas mentioned above.
dJrsens Arsens M sens
For the calibration of standard weights, the sensitivity mass Mens should be at least twenty
times larger than the mass difference (t-r) (see Section 3.5), and hence the sensitivity
coefficients for Mens and Jrêns should be smaller than 0.05. When this condition is met, the
uncertainties in these quantities are insignificant and can be ignored. When calibrating non-
standard weights, such as weights for a pressure balance, however, the c, values may be greater
than 0.05 and consequently uncertainties in Mezw and Arse„s may be significant (see
Section 5.2.3.2)
Each of the above sources of uncertainty will now be examined in turn.

5.2 UNCERTAINTY OF DIRECT COMPARISON 33
5.2.2 Uncertainty contributions for direct comparison
(1) Calibration uncertainties of the reference weights
Calibration reports will normally give the expanded uncertainty U and the value of coverage
factor k used. In this case C/, = U, ki = k and the number of degrees of freedom Vi equivalent
to k can be found from Table L.l. (See Appendix L for an explanation of the quantities
Ui and ki.} Sometimes a report may give a standard uncertainty m and the degrees of freedom
V. Then, U, = u, ki = 1 and v, = v.
It is sometimes necessary to use more than one reference weight. To calibrate a 300 g
weight from a set with the 5, 3, 2, 1 structure for instance, the reference may be a 100 g weight
together with a 200 g weight. The uncertainty of a combination of N reference weights may
be found by simple addition of their standard uncertainties — a simplification based on the
reasonable assumption that they are highly correlated (see Note below). Thus, if each
calibration uncertainty is converted to a standard uncertainty, with that for the f reference
mass being we have for the reference weight combination
^.•=Z“y (5.3)
7=1
Since U, is a standard uncertainty, ki = 1 and one should take v; to be equal to the degrees of
freedom Vj for the largest of the contributions, Uj(max).
As an example, consider the calibration of a 300 g weight requiring reference weights of 100 g
and 200 g. The former has a reported (expanded) uncertainty of 0.15 mg with a coverage factor
of 2.0 and the latter 0.25 mg with a coverage factor of 2.1. Thus, the standard uncertainties for
the individual reference weights are
• For the 100 g weight: m /oo = 0.15/2.0 = 0.075 mg.
• For the 200 g weight: U200 = 0.25/2.1 = 0.119 mg.
and from equation (5.3), the combined reference uncertainty is therefore U, = 0.075 + 0.119
= 0.194 mg. The larger standard uncertainty, that for the 200 g weight, has 19 degrees of
freedom (since the coverage factor is 2.1). Hence, we have U, =0.194 mg, it, = 1 & v; = 19.
Note: The covariances Cy* between the N reference weights are not normally known, but if they are, then a better
estimate of tZ, is obtained from equation (5.4). Once again, put equal to the degrees of freedom Vj for the largest
of the contributions, Uj(max).
N N-\ N
(5.4)
7=1 7=1 *=7+1
Equation (5.4) becomes identical with equation (5.3) as the covariances approach their
maximum values, the condition known as full, or 100%, correlation. In practice, reference
weights will be highly correlated, particularly if they are from the same set. Hence
equation (5.3) is a reasonable approximation.
34 5 UNCERTAINTY ESTIMATION FOR MASS CALIBRATION
(2) Instability of the reference weights
The mass of a weight will invariably change somewhat after calibration. Such changes are
largely unpredictable and will affect the uncertainty of reference mass values. In what follows,
it is assumed that the uncertainties quoted in the reports for reference weights do not include a
contribution to account for these changes. This contribution must therefore be included in the
calculations presented here. If it is clear from the report on the reference weights that quoted
uncertainties do include the effect of mass instability, then the uncertainty contribution
discussed in this subsection should be omitted from the calculations.
The estimation of this uncertainty component is obviously very difficult unless a long
calibration history or other data, such as a history for similar weights, is available (see Note
below). In most cases the calibrator has little option but to use a rule devised by someone who
has studied mass variations in similar weights under similar conditions of use and storage.
Such a rule may, for instance, relate the standard uncertainty due to mass instability to the
OIML maximum permissible error (MPE, see Appendix I).
Experience at NML has shown that this contribution can be surprisingly large even for
properly handled weights from a reputable manufacturer. It is therefore recommended that, in
the absence of other information and under reasonably good laboratory conditions, the
standard uncertainty be taken as 8 % of the MPE for the relevant OIML class, with 4 degrees
of freedom (corresponding to an estimated 35% relative uncertainty for this estimate). For
weights used in relatively poor conditions or by inexperienced personnel, this estimate may
have to be increased by a factor of 2 or more. As tT, is a standard uncertainty, = 1.
If there is doubt about the effective OIML class of reference weights (which can often be
the case for older weights), then the laboratory that calibrated the weight may be able to offer
advice.
When using two or more reference weights in combination, one can take the instability
contributions for the various weights to be uncorrelated (unless a calibration history suggests
otherwise). They can therefore be added in quadrature (equation L.3).
Note: Obtaining an estimate of the instability uncertainty from a calibration history is clearly more satisfactory
than merely adopting a certain percentage of the OIML MPE, but it must be approached with caution.
The history should contain at least five calibrations, preferably carried out at the same laboratory and having similar
uncertainties. Provided the mass values from these calibrations show no consistent trend, one finds their standard
deviation sm and compares it with the mean standard uncertainty ur of the calibrations (obtained by dividing each
reported calibration uncertainty by its coverage factor and averaging the results).
The correlations between the calibration uncertainties complicate the analysis, but it is reasonable to take the
instability standard uncertainty to be equal to if sm is greater than Ur /2, and zero otherwise.
If a trend is evident in the mass values (eg, they tend to rise with time) then the trend should be eliminated from the
data before proceeding as above. One must then decide whether to correct reported mass value for the trend or to
treat the trend as an extra component of uncertainty. The latter course is safer.
For this estimate of C/„ k, = 1 and v,- is equal to one less than the number of calibrations analysed.
(3) Repeatability of mass comparison for the balance
The repeatability of mass comparison (section 3.6) is the repeatability of a mass difference
(î - r,) obtained in substitution weighing. Although the circumstances of balance sensitivity
measurement are somewhat different, the repeatability of can also be taken as
Since (t - r) is the mean of n differences, the standard uncertainty of this quantity will be
These estimates of tT,- are standard uncertainties so ki =1, and Vi is the number of
degrees of freedom of s„, which is usually nine.
(4) Resolution of the balance
For a digital balance, the “true” reading lies within a band equal to the discrimination (least
count) and centred on the observed reading. For example, if the discrimination is 0.000 01 g,
an observed reading of 2.356 73 g indicates a “true” reading lying between 2.356 725 and
2.356 735 g with a rectangular probability distribution. The resolution uncertainty is therefore
± a, where a is half the discrimination (the semi-range).
__ The resolution uncertainty for a difference between two readings, such as ^sens or
~ 5 will be ^2 larger. So we take C/, = ^2 a. Since the reducing factor for the semi-
range of a rectangular distribution is ^3 [4, 5], ki=y[3= 1.73. Moreover, the resolution
uncertainty for a digital balance has infinite degrees of freedom, although, in practice any
value greater than about 100 will do in the calculations that follow (say, = 1000).
The readability of an analog scale is typically ± a tenth of the smallest scale division.
Therefore take a = ± 1/10 of the smallest division.
(5) Buoyancy
In conventional weighing, the buoyancy correction given in equation (2.7) is not normally
made, particularly in laboratories at altitudes within 300 m of sea level. The buoyancy
uncertainty is therefore the uncertainty that results from ignoring the correction, and it is given
(see Appendix J) by
=4.5x1 QÂp^RM mg (5.5)
where Apr is the total width (in kg.m’^) of the OIML density limits (see Appendix I) for the
class of the test weight T, R is given for various altitudes in Table J.l, and M is the nominal
mass of the test weight in grams. Since all the quantities in equation (5.5) are well defined, the
degrees of freedom are very high; the exact figure is not important so one may take it to be
1000. So Ui = Ub and since Ub is a standard uncertainty, A:, = 1.
Note: If the correction given in equation (2.7) is made and a similar correction was made when the reference
weight was calibrated, then provided p^, Pr and pr are known with uncertainties of 1% or less, Ub will be negligible
except in some extreme cases. To check that this is so, Ub may be calculated from equation (5.6) [3]:
(5.6}
M VPt Pr ) Pt Pr Pr
where Upo, u ^t and u ^r arc the uncertainties of pr and Pr respectively and Pai is the air density when the reference
mass was calibrated. The degrees of freedom are not easily calculated and it is recommended that it be equated to
the degrees of freedom for Upa, UpT or UpR depending on which of these makes the largest contribution to uy.
(6) Rounding
Numbers are rounded to remove unwarranted detail in the least significant digits and the
rounding process begins by considering the magnitude of U [5]. It should be done only once,
when the final measurement result is expressed. Briefly, express the uncertainty to only one
significant figure when the most significant digit (the first non-zero digit from the left) is 5 or
more and to two significant figures otherwise, and the values should be rounded up except
where rounding down involves a change of less than a few percent. Finally, the value of the
measurand should be rounded to match that of the uncertainty.
If, for example, the measured mass values are rounded to the nearest 0.001 mg, then the
rounding uncertainty is ± 0.0005 mg with a rectangular probability distribution. Thus we
would put Ui= 0.0005 mg, ki= 1.73, as the distribution is rectangular (see item (4) above),
and = 1000.
5.2.3 Examples in calculating the uncertainties for a direct comparison
The two following examples make use of summary tables (such as Table 5.1) that are easily
adapted to a spreadsheet program. The shaded columns show the location of the input data Ui,
ki, Ci and 14. In the spreadsheet, the m , column contains the formula given by equation (L.2),
the combined uncertainty Uc is calculated from equation (L.3), Veff from equation (L.6), k from
equation (L.8) and U from equation (L.7).
5.2.3.1 Calibration of a 10 kg Fi weight
In this first example, a 10 kg Fi weight is calibrated in terms of an E2 reference using double

substitution weighing on a 30 kg balance having a repeatability of mass comparison of
7.5 mg (with 9 degrees of freedom) and a discrimination of 5 mg. The laboratory is within 20
metres of sea level and the reference has a history of only 3 calibrations.
We first calculate the various sensitivity coefficients. The difference {t-r) between
reference and test masses is approximately 27 mg, whereas the weight used to measure the
balance sensitivity reciprocal had a mass {Mêns} of 5 g. From equations (5.2), we have
Ci for Mr = 1
Ci lox M,ens =
Ci lox ^r,ens =-21!5^ =-Q
Ci for (Z - r) =1
Next, we evaluate the component uncertainties that result from the six sources listed in
the above section.
1- The calibration uncertainties: In the calibration report, the expanded uncertainty of the
reference mass Mr is given as 5 mg with a coverage factor of 2.1. From Table L.l we see that
this coverage factor implies 19 degrees of freedom. Hence CT, = 5 mg, it, = 2.1 and = 19.
The calibration uncertainty for Mêns is ignored, as the sensitivity coefficient for this quantity is
less than 0.05 (see Section 5.2.1).
2- The instability in M^. As a satisfactory calibration history for the reference does not exist,
the instability standard uncertainty is taken as 8% of the MPE for a 10 kg E2 weight, with 4
degrees of freedom. Hence CT, = 0.08 x 15 mg = 1.2 mg, ki = 1 and = 4. The instability of
Msens is ignored, as the sensitivity coefficient is less than 0.05.
3. Balance repeatability: The balance has a repeatability Sm of 7.5 mg with 9 degrees of

freedom. As n = 2 for double substitution weighing, CT, = 1.51^2 = 5.30 mg, ki = 1 and v; = 9.
The uncertainty in Arsens due to repeatability is ignored, as the sensitivity coefficient is less
than 0.05.
4- Resolution of the balance: The discrimination is 5 mg, so the resolution uncertainty is

±2.5 mg (i.e., the semi-range, a = 2.5 mg). Hence U=Ca = 3.54 mg with ki = 1.73 and
v; = 1000. The corresponding uncertainty in Arsens is ignored, as the sensitivity coefficient is
less than 0.05.
5. Buoyancy: Since the laboratory is near sea level, the buoyancy uncertainty will be
negligible compared with the OIML MPU for an Fi weight. However it may be significant
compared with the uncertainty actually achieved. The OIML density band Apr is 1340 kg.m'^
for an Fl weight and R is 0.02 kg.m'^ (Table J.l). So, from equation (5.5):
Ub = 4.5 X 10'^ X 1340 X 0.02 x 10000 = 1.206 mg
giving Ui = 1.206 mg, ki = 1 and = 1000.
6. Rounding: Both Mt and its expanded uncertainty U will be rounded to the nearest 1 mg,
i.e., to ± 0.5 mg. So, CT, = 0.5 mg, ki = 1.73 and v; = 1000.
The above data are shown in the uncertainty summary table, Table 5.1. The report for
this weight would quote an uncertainty of 0.014 g with a coverage factor of 2.1.
Table 5.1: Summary table for calculating the calibration uncertainty for a 10 kg Fj weight calibrated by direct
comparison using double substitution weighing.
Component Values
Component (/ = 1, 2,... 6) (/, (mg) ki |c,| Ui (mg) Vi U^/Vi
1. Calibration of Mr 5 2.1 1 2.381 19 5.669 1.7E+00
2. Instability in Mr 1.2 1 1 1.200 4 1.440 5.2E-01
3. Balance repeatability on (/T7) 5.30 1 1 5.300 9 28.090 8.8E+01
4. Balance resolution on (^) 3.54 1.73 1 2.046 1000 4.187 1.8E-02
5. Buoyancy 1.206 1 1 1.206 1000 1.454 2.1E-03
6. Rounding of Mt 0.5 1.73 1 0.289 1000 0.084 7.0E-06
Combined std. uncertainty Uc (mg) 6.40

Degrees of freedom 19
Coverage factor k 2.10
Expanded uncertainty U (mg) 13.44
5.2.3.2 Calibration of non-standard weights
In this example we consider the calibration of six brass weights that are to be used on a
pressure balance. Their quality is roughly equivalent to class F2 and they have masses of
approximately 55.065 g, the largest mass being 55.070 g. Four substitution weighings are
made on a 100 g balance having a discrimination of 0.1 mg and a repeatability of mass
comparison of 0.19 mg (with 9 degrees of freedom). The reference weight set has only been
calibrated once.
The reference could be a 50 g + 5 g + 50 mg + 10 mg + 5 mg group or perhaps a 50 g +

5 g + 100 mg group. However to simplify the weighings, just the first two weights are used.
Thus the maximum mass difference {t - r) is 55.070 - 55 = 0.070 g. The sensitivity
reciprocal of the balance was measured with a 100 mg weight with an uncertainty of 0.012 mg
and a coverage factor of 2.0. The sensitivity coefficients, from equations (5.2), are as follows:
Ci for Mr = 1
Ci for Men, =70/100 = 0.7
Ci ioT Nrsens =“ 70/100 =-0.7
Ci for {t - r) =1
Since the sensitivity coefficients for Mêns and Nrêns are greater than 0.05, the uncertainties
involving these quantities cannot be ignored (see Section 5.2.1) as in the last example. Indeed,
it will be seen, in Table 5.2, that the quantity Nrsens is responsible for the largest contribution to
the uncertainty of Mt . The component uncertainties are now evaluated.
1 • The calibration uncertainties: The calibration uncertainties of the 50 g and 5 g weights are
0.09 mg and 0.035 mg respectively with a coverage factor of 2.0, implying 60 degrees of
freedom. As described in Section 5.2.2 (1), the standard uncertainty of this group must be
found by simple addition of the standard uncertainties of the two weights:
Ui =0.09/2 + 0.035/2 = 0.045 + 0.0175 = 0.0625 mg
with ki = 1 (as this Ui is a standard uncertainty) and = 60.
For Msens, Ui =0.012 mg, ki = 2.0 and = 60.
2- The instability in and As calibration histories for these weights do not exist, the
instability standard uncertainty is taken as 8% of the Fj MPE.
For the 50 g weight: U = 0.08 x 0.30 mg = 0.024 mg
For5g: Ui = 0.08x0.15 mg = 0.012 mg
For 100 mg: Ui = 0.08 x 0.05 mg = 0.004 mg
In each case, ki = 1 and v; = 4.
3. Balance repeatability: The balance repeatability is 0.19 mg with 9 degrees of freedom. As

four weighings were made when measuring (F^), U = 0.19/^/4 = 0.095 mg. When
measuring Ardens, only a single weighing was made, so U =0.19 mg. In both cases, ki = 1
and î = 9.
4. Resolution of the balance: The discrimination is 0.1 mg, so the resolution uncertainty is
±0.05 mg (i.e., the semi-range, a = 0.05 mg). Hence, for both the (F^) and the Ardens
measurements, Ui= ^2 a =0.071 mg with it, = 1.73 and k = 1000.
5. Buoyancy: The buoyancy uncertainty is seldom significant for F2 test weights and in this
case it is sure to be negligible, even if the laboratory were located at high altitude, as both the
test and reference weights are made of brass and hence will have very nearly the same density.
6. Rounding: Both Mt and its expanded uncertainty U will be rounded to the nearest
0.01 mg, i.e., to ± 0.005 mg. So, C/, = 0.005 mg, ki = 1.73 and k = 1000.
The above data are shown in the uncertainty summary table, Table 5.2. The relatively
large contributions due to Ar^^ns (0.133 and 0.029 mg) highlight the need to use reference
weights with a combined mass very close to that of any non-standard object when making
accurate mass measurements by substitution weighing. It follows from equations (5.2) that the
smaller the difference between reference and test objects, the smaller {t - r} will be, and
hence the smaller the q values for M,e„s and Ardens- Small c, values in turn reduce the effect of
Msens and Arsens Uncertainties.
The report for the weights would quote an uncertainty of 0.38 mg with a coverage factor
of2.06.
40 J UNCERTAINTY ESTIMATION FOR MASS CALIBRATION
Table 5.2: Summary table for calculating the calibration uncertainty for 55.065 g weights calibrated by direct
comparison using four substitution weighings.
Component Values
Component (/ = 1, 2,... 11) f/,(mg) >1, k.-l (mg) Vi Ui
1. Calibration of Mr 0.0625 1 1 0.063 60 3.9E-()3 2.5E-07
2. Calibration of Mêns 0.012 2 0.7 0.004 60 1.8E-05 5.2E-12
3. Instability in 50 g 0.024 1 1 0.024 4 5.8E-D4 8.3E-08
4. Instability in 5 g 0.012 1 1 0.012 4 1.4E-()4 5.2E-09
5. Instability in Mêns 0.004 1 0.7 0.003 4 7.8E-06 1.5E-11
6. Balance repeatability on (t-r) 0.095 1 1 0.095 9 9.0E-03 9.1E-06
7. Balance repeatability on 0.19 1 0.7 0.133 9 1.8E-02 3.5E-05
8. Balance resolution on (iT7) 0,071 1.73 1 0.041 1000 1.7E-O3 2.8E-09
9. Balance resolution on 0.071 1.73 0.7 0.029 1000 8.3E-04 6.8E-10
10. Buoyancy 0 1 1 0.000 1000 O.OE+00 0.0E4O0
11. Rounding of Mt 0.005 1.73 1 0.003 1000 8.4E-06 7.0E-14
Combined std. uncertainty Uc (mg) 0.184

5.3 UNCERTAINTIES FOR WEIGHING SCHEMES 41
5.3 Uncertainties for weighing schemes

For weighing schemes, the model is more complicated than equation (5.1). However the
model for each individual measurement of mass difference a, is very similar to equation (5.1),
since from equation (4.2):
Although the mass of the reference, Mr , does not appear in this equation, it enters via the
analysis of the weighing scheme. Hence the six uncertainty sources and the sensitivity
coefficients listed in Section 5.2.1 still apply, as does the discussion of Section 5.2.2.
Equations (L.3) and (L.6) are not applicable in the simple forms shown. Instead, Uc and Vgff
must be calculated with the aid of the spreadsheets introduced in Chapter 4, as will be
described below.
To minimise the complexity of the examples given below, it will be assumed that the
sensitivity coefficients for Msens and Arsens are less than 0.05 so that uncertainty contributions
involving these terms can be ignored. This will always be the case for the calibration of
standard weights if Msens is large enough (Section 3.5).
To illustrate the approach to calculating the uncertainties of mass values obtained in

weighing schemes, reference will be made to the spreadsheet for the closed cycle scheme
shown in Figure 4.1. However the following discussion is easily adapted to any weighing
scheme.
Firstly one must combine the uncertainty components involving Mr . Remember that the
head weight is the reference weight for a decade scheme. The resulting combined standard
uncertainty u^R} of the reference mass and the corresponding degrees of freedom Ve^K) are
then entered into the spreadsheet at cells E5 and E6 respectively.
The next step is to calculate the standard uncertainties of the mass differences aj.
Equation (5.7) shows that^r each mass comparison in the scheme, this means combining the
components involving {t - r). The combined standard uncertainty Uc(AM) of the first mass
difference ai and the associated degrees of freedom Vefi(AM) are then entered into the
spreadsheet at cells Ell and Fl 1 respectively. The corresponding quantities for the other mass
differences are entered in the cells below El 1 and Fl 1.
For each measured mass value, the scheme gives a standard uncertainty and
corresponding degrees of freedom (for instance: cells DI9 and El9 for the first test weight
Tl). For this uncertainty, ki= 1 and Ci =1, and it must be combined with the buoyancy
uncertainty (if significant) and the rounding uncertainty to obtain the combined uncertainty of
the measured mass value.
For decade weighing, the head and test weights normally come from the same set and
hence will have very similar densities. Consequently the buoyancy uncertainty will very likely
be negligible for the test weights of a decade scheme, even at high altitudes. It may be
significant for the check weight, but the uncertainty of this weight does not need to calculated.
Examples are now given for a closed cycle and a decade scheme.
5.3.1 Uncertainty estimation for a closed-cycle weighing scheme
This example calculates the uncertainties for the closed cycle weighing scheme shown in
Section 4.4.1, where a 500 g E2 reference is used to calibrate two Fi test weights.
The uncertainty of the reference mass is 0.18 mg with 30 degrees of freedom, so tZ, =
0.18 mg, ki = 2.04 (from Table L.l) and Vj = 30. The instability uncertainty is taken to be
8% of the OIML MPE of 0.75 mg with 4 degrees of freedom, so U, = 0.75 x 0.08 = 0.06 mg,
ki = 1 and = 4. These contributions are combined as shown in the top section of Table 5.3
to give values for u^R} and Ve^R\
The repeatability of mass comparison Sm for the 1 kg balance is 0.11 mg with 9 degrees
of freedom and single substitution weighing is used. Hence Ut = 0.11 mg, ki = 1 and Vi = 9.
The balance discrimination is 0.1 mg, so the semi-range a = 0.05 mg. So for this component,
Ui =5/2 a = 0.071 mg, ki = 1.73 and Vi = 1000 (see Section 5.2.2 (4)). These contributions
are combined as shown in the middle section of Table 5.3 to give values for uÂM} and
VejÂM), which will be the same for all three mass comparisons.
When Uc{R}, Ve^R}, and VgÂM) are entered (as described in Section 5.3) into
the closed cycle spreadsheet, one obtains for both test masses a scheme standard uncertainty of
0.143 mg with 49 degrees of freedom (see Figure 5.1).
The laboratory is within 100 m of sea-level, hence R = 0.02 kg.m’^ in equation (5.5),
while the OIML density range Apr is 1340 kg.m'^. So, for the buoyancy component,
Ui = 4.5 X IO’" X 1340 X 0.02 x 500 = 0.0603 mg, ki = 1 and v; = 1000.
The result will be rounded to the nearest 0.01 mg, ie to ± 0.005 mg. So C/, = 0.005 mg,
ki = 1.73 and k = 1000.
The scheme, buoyancy and rounding uncertainties are combined as shown in the last part
of Table 5.3 to give a combined uncertainty of 0.31 mg with a coverage factor of 2.0. This
analysis applies equally to both test weights.
Table 5.3; The summary table for the calculation of combined expanded uncertainty for two 500 g Fj
weights calibrated by means of a closed cycle scheme. Shaded cells show input data.
Component Values
Component f/, (mg) ki kzl M, (mg) Vi
1. Calibration of Mr 0.18 2.04 1 0.088 30 7.8E-03 2.0E-06
2. Instability in Afj; 0.06 1 1 0.060 4 3.6E-03 3.2E-06
Combined std. uncert. Uc{R) (mg) 0.107

Degrees of freedom v^X^) 25
Component Values
Component Ui (mg) |c,| Ui (mg) Vi u,^
3. Balance repeatability on (737) 0.11 1 0.110 9 1.2E-02 1.6E-05
4. Balance resolution on (737) 0.071 1.73 1 0.041 1000 1.7E-03 2.8E-09
Combined std. uncert. Uc(AA/) (mg) 0.117

Degrees of freedom Vg^zlA/) 12
Component Values
Component CT, (mg) M, (mg) Vi u^ Ui/Vi
From weighing scheme 0.143 1 1 0.143 49 2.0E-02 8.5E-O6
5. Buoyancy 0.0603 1 1 0.060 1000 3.6E-03 1.3E-08
6. Rounding of result 0.005 1.73 1 0.003 1000 8.4E-06 7.0E-14
Combined std. uncert. Uc (mg) 0.155

Coverage factor A: 2.00
As a check on the measurements, the quantity Sum a should be less than =

0.41 mg. In fact it is only 0.20 mg, so we can be reasonably confident in the results of the
scheme. (If Sum a exceeds 2>/3u ,(z 1M), then this is a warning of a potential problem that
should be investigated. See section 3.4. At the very least, the measurements should be
repeated.)
Figure 5.1: The appearance of the Closed Cycle spreadsheet after entering (into the shaded cells) the
data relevant to the example of section 4,4.1.
5.3.2 Uncertainty estimation for a decade weighing scheme
The uncertainties of the five Fi test masses calibrated using the extended decade weighing
scheme of example 4.4.3 are estimated here. The first four mass comparisons were made on a
160 g balance and the last four on a 30 g balance. Double substitution weighing was used
throughout.
The 100 g head weight (also class Fi) is from the same set and was calibrated the
previous day as part of the 1 kg to lOOg decade. The result of that calibration was a mass of
100.000 29 g with a standard uncertainty of 0.055 mg having 42 degrees of freedom. The
head weight instability imcertainty is taken as zero since the calibration was so recent and the
weight was not even moved to another laboratory. There is therefore no need for a section of
the summary table for combining the Mr uncertainties and we can just take uAR} = 0.055 mg
and Ve/7?) = 42.
First four mass comparisons: The balance repeatability Sm is 0.018 mg with 9 degrees of
freedom, so U, = 0.022/ 41 = 0.0156 mg, ki = 1 and = 9. The balance discrimination is
0.01 mg so the semi-range a is 0.005 mg. Hence Ui = V2 a = 0.0071 mg, ki = 1.73 and
Vi = 1000.
Last four mass comparisons: The repeatability is 0.006 mg, again with 9 degrees of
freedom, so U, = 0.006/^2 = 0.00424 mg, A:, = 1, =9. The discrimination is 0.001 mg
and hence t/, = V2 x 0.0005 = 0.00071 mg, ki = 1.73 and v, = 1000.
The calculation of and Kg/zlA/) is shown in Table 5.4.
Table 5.4: The calculation of the combined standard uncertainties of mass comparisons for F, weights
calibrated by means of an extended decade scheme. Shaded cells show input data.
Comparisons 1 - 4
Component Values
Component C/,- (mg) ki Iq I «. (mg) Vi
3. Balance repeatability on (737) 0.0156 ■*1 0.0156 9 2.4E-04 6.6E-09
4. Balance resolution on (737) 0.0071 1.73 1 0.0041 1000 1.7E-O5 2.8E-13
Combined std. uncert. Wc(JA/) (mg) 0.0161

Comparisons 5-8
Component Values
Component CT, (mg) |czl «, (mg) Vi
3. Balance repeatability on (737) 0.00424 1 1 0.0042 9 1.8E-05 3.6E-11
4. Balance resolution on (737) 0.00071 1.73 1 0.0004 1000 1.7E-07 2.8E-17
Combined std. uncert. Uc{AM) (mg) 0.00426

Degrees of freedom Vg^dA/) 9
When the uncertainties and degrees of freedom for the reference mass and the
comparisons are entered into the decade scheme spreadsheet in a manner similar to that
described in Section 5.3, one obtains the standard uncertainties: 0.0287, 0.0116 and
0.0060 mg for the 50 g, 20 g, and 10 g weights, respectively. The degrees of freedom are 49,
51 and 60, respectively.
Since all the weights are from the same set, it is assumed that the densities are very
similar, and hence the buoyancy uncertainty can be ignored. (The 10 g check weight is from
another set, but the uncertainty of its measured mass does not need to be calculated.)
Mass values will be rounded to the nearest 0.01 mg (for the 50 g weight) and to the
nearest 0.001 mg (for the 20 and 10 g weights). So U is equal to 0.005 mg for 50 g and
0.0005 mg for 20 and 10 g, with ki = 1.73 and = 1000. The effect of this rounding
uncertainty is negligible as can be seen from the final combination of uncertainties shown in
Table 5.5 for the 50 g weight. Spreadsheets for the 20 and 10 g weights are similar.
The expanded uncertainties are therefore 0.06, 0.024 and 0.012 mg, for the 50 g, 20 g,
and 10 g weights, respectively.
Table 5.5: The calculation of the combined expanded uncertainty of the 50 g mass value calibrated by
means of an extended decade scheme. Shaded cells show input data.
50 g
Component Values
Component U (mg) A:, Iq I M, (mg) Vi
From weighing scheme 0.0287 1 1 0.029 49 8.2E-04 1.4E-08
5. Buoyancy 0 1 1 0.000 1000 O.OE+00 O.OE+00
6. Rounding of result 0.005 1.73 1 0.003 1000 8.4E-06 7.0E-14
Combined std. uncert. Uc (mg) 0.0288

The mass obtained for the check weight can be used in a consistency check based on the
t-test described in Appendix G. (See the example of Section G.2.) This check is available for
both simple and extended decade schemes.
The decade standard deviation shown on the decade scheme spreadsheet is 0.0171 mg.
As a further check on the measurements, the square of this quantity should be less than four
times the average value of Uc{/YMf. This rule is obtained from an F-test similar to those
discussed in Appendix G. In this case we require 0.017P to be less than
4-(o.O16P +O.OO4262)/2, where the values for Uc{/YM} are taken from Table 5.4. That is, we
require 0.00029 to be less than 0.00055, which it is. If this check fails, the decade should be
repeated. This check is not possible for the simple decade scheme.
Chapter 6
Calibration of electronic balances
6.1 What is an electronic balance?

An electronic balance is essentially one that uses an electric sensor, the output of which is
processed electronically to generate a reading shown on a digital display. However many
balances that do not fit this simple description may also be calibrated by the procedures
described here. In particular,
• A balance may have tare weights, but preferably not more than two or three. This allows
the electronic range (i.e., the range of the display for a given selection of tare weights) to
be less than the full range of the balance.
• Instead of being displayed directly, the reading may be transmitted to a computer, data
logger or printer. This does not matter provided the present reading can be observed when
required.
• Some balances have two or more ranges with different characteristics. These ranges may
have to be calibrated separately as described below.
• The balance may be top-loading or have a suspended pan.
See Appendix K for more details of the construction of electronic balances.
6.2 The scope and purpose of the calibration

The main purpose of a balance calibration is an estimation of the accuracy of measurements
made on the balance. However balance calibrations have three other important functions:
(a) A table of corrections to the balance readings is provided.

(b) The user is given information on limitations of the balance, principally the effects of off-
centre loading and hysteresis. This information can act as a diagnostic indicating that
repairs may be required.
(c) The balance report contains information that allows the user to carry out effective checks
on the balance between calibrations (see Section 6.5.2 for details of user checks).
47
48 6 CALIBRATION OF ELECTRONIC BALANCES
The balance is adjusted, using the sensitivity adjustment (see Section 6.3.1) present in
almost all electronic balances, before calibration and before user checks (and at any other
time deemed necessary). In this way, sensitivity drift is eliminated, which is essential if the
user checks are to detect other changes in the balance performance.
6.3 Balance calibration

The following general considerations should be noted:
1. Unlike most other measuring instruments, balances must be calibrated in the position
where they are to be used. They cannot be sent to a calibrating laboratory for calibration.
If it becomes necessary to move a balance after calibration, even to another position in the
same room, the user checks of Section 6.5.2 should be carried out to see whether a
recalibration is necessary. However this rule may be relaxed for low accuracy instruments.
2. Some balances have a high-resolution range (with ten times the normal resolution)
covering perhaps one-fifth the total range of the balance. Sometimes the high-resolution range
is permanently restricted to the first 20% (say) of the load range, while sometimes it can, by
taring, be moved anywhere throughout the load range. The moveability of a high-resolution
range is useful for substitution weighing (see Chapter 3), but is of limited value when using a
calibrated balance in the normal way.
Nevertheless, one can take some advantage of a movable high-resolution range when
calibrating a balance. While the repeatability and the corrections to the balance readings must
be measured without retaring, off-centre loading errors and hysteresis may be measured after
retaring in order to make use of the higher resolution.
3. The last digit of a reading will sometimes flick continually between two successive
numbers. When this happens the reading used should be the mean of the two digits.
4. Where calibrated weights are called for (Sections 6.3.3 and 6.6), the density of the weights
should be in the vicinity of 8000 kg.m'^. The accuracy of the balance determines how close
the density must be to this value. This is usually not a problem provided the weights are made
of stainless steel, brass or bronze, but to be quite sure that the density of the weight is
acceptable, use weights conforming to a standard, such as the OIML classification [3].
5. A client may request that a balance be calibrated over a restricted range. This is
acceptable provided the range restriction is made clear in the report.
6. Some balances have a zero-tracking facility whereby the balance automatically tares out
small departures from a zero reading. This facility should be turned off during balance
calibration and user checking.
6.3 BALANCE CALIBRATION 49
6.3.1 Adjustment
Almost all electronic balances are adjustable as it is necessary to set the scale to the local
gravity (the local value of g). Balance manuals often refer to this procedure as “calibration”,
but, as it is really just a sensitivity adjustment, the term “adjustment” is used here. This avoids
contusion with the far more comprehensive calibration of balances described in this chapter.
The adjustment may be automatic, semi-automatic or manual. It may be activated

whenever the balance is switched on, or only when triggered by the user. It involves a weight
usually referred to in balance manuals as the calibration weight, which may be a built-in or a
designated separate weight. The manual should give instructions for carrying out the
adjustment.
Before adjusting the balance, make sure that it has been switched on for the warm-up
period recommended by the manufacturer, or at least 30 minutes if this period is not known. It
is also recommended that the balance be “exercised” by placing a weight, of mass
approximately equal to full load, on the pan once or twice.
Adjust the balance following the manufacturer’s instructions. In the absence of

instructions: zero the balance with the pan empty; place a weight of known mass (preferably
close to the maximum capacity) on the pan; adjust the balance until it displays the known mass
value; then remove the weight and check that the balance zero has not changed significantly
(by not more than 1 or 2 counts in most cases). If the balance zero has changed, then repeat
the adjustment.
Between successive calibrations, the validity of the most recent calibration depends on
the mass stability of the calibration weight. If a separate weight is used, it must therefore be
handled with great care. It should remain with the balance user (not the calibrator), be used for
this purpose only, and be stored separately from other weights in a suitably lined dust-tight
container.
Many adjustment procedures assume that the mass of the calibration weight is close to
its nominal value. For instance, a balance with a 10 g calibration weight may automatically
display 10.0000 g after adjustment. If it is found that the corrections to the balance reading
(see section 6.3.3) and hence the Limit of Performance (see section 6.4.5) are unexpectedly
large, it is possible that a poor calibration weight is to blame. For built-in calibration weights,
this problem should arise rarely, unless the balance is used in an adverse environment.
6.3.2 Repeatability of measurement
The repeatability of measurement indicates how consistently the mass of an object can be
measured by a balance. It is expressed as the standard deviation Sr obtained from a series of
repeated mass measurements. Since electronic balances do not weigh at constant load, Sr may
vary significantly over the range of the balance, and consequently it should be measured at
more than one load.
A simple mass measurement on a calibrated balance consists of two readings — a zero

reading z and a reading r with the object whose mass is to be measured on the pan. (If drift is
a problem, a second zero reading is recommended, see section 6.5.1.) To measure 5^, one
therefore calculates the standard deviation of n mass measurements, the such measurement
being
(6.1)
Thus if the average value of A/, is M,
(6.2)
The recommended minimum number of measurements (i.e., the minimum value of n in

equation 6.2) is 10, in which case Sr has nine degrees of freedom. This procedure should be
carried out near half maximum and maximum load. Standard weights are not required; any
clean and reasonably smooth piece of turned brass or stainless steel is adequate.
For multi-range balances,

(a) if the balance’s measuring system changes from range to range, the repeatability should be
measured near the mid-point and near the top of each range
(b) if the balance’s measuring system does not change, a measurement near the top of each
range is sufficient.
As a result of the limited resolution of many electronic balances, the A/, values obtained
by the above procedure may all be the same, or very nearly so. In these cases, the measured Sr
will be less than the true repeatability of the balance and may even be zero. The measured
value of Sr should nevertheless be recorded in the balance report.
6.3.3 Corrections to balance reading
Even though a balance has been adjusted to read correctly at a certain load (usually close to the
maximum), it may still be necessary to apply corrections to the readings, for two reasons:
(a) The mass of the calibration weight may differ significantly from its nominal value
(b) The balance reading as a function of load may be significantly non-linear.
The corrections to balance reading should be measured at sufficient equally-spaced

points over the range to ensure that one may safely interpolate between points. This usually
means a minimum of 10 points. The following procedure assumes that the balance calibrator
has a set of calibrated weights (see section 6.6 if this is not the case). As far as practicable, the
calibrated masses should have uncertainties less than the resolution of the balance, in order to
minimise the uncertainty of the measured corrections.
The corrections to balance reading are determined by successively placing calibrated

masses on the pan and observing the readings. The steps are:
(i) read the zero;
(ii) place weight(s) of known mass M on the pan, where M is approximately one-tenth the
range of the balance. Note the reading;
(iii) remove the weight(s) momentarily, then replace them on the pan and note the new
reading. This reading is averaged with the one in (ii) to give ry;
(iv) remove the weight(s) and read the zero. This reading is averaged with the reading in
(i) to give the zero value, zf,
(v) place weight(s) of mass 2M, on the pan and note the reading;
(vi) remove the weight(s) momentarily then replace them on the pan and note the new
reading. This reading is averaged with the one in (v) to give rf,
(vii) remove the weight(s) and read the zero. This reading is averaged with the reading in
(iv) to give the zero value, Z2,
(viii) repeat with weights of mass 5M, 4M, etc, until the capacity of the balance is reached.
If readings r, and z, correspond to weight(s) of total mass M on the pan, then the
correction to balance reading for this load is given by
C. -(r.-zj (6.3)
The correction is the amount that should be added to a measurement (eg, equation 6.1) to
give the correct mass. It should be small and not change significantly with load (within the
manufacturer’s tolerance). The negative of the correction (— C,) is known as the departure
from nominal value.
For balances with more than two ranges, the correction to balance reading may be
measured at fewer than 10 points per range, provided there are at least 20 points covering the
full balance range. If the only change between ranges is the manual loading or unloading of
tare weights, then measure the correction for at least 10 points over the full balance range, and
for at least 4 points on each range.
6.3.4 Effect of off-centre loading
When the centre of mass of the object being weighed is off-centre on the pan, off-centre
loading error (also called shift or comer-load error) may occur. It is difficult to correct
accurately for this because the effect is not always linear with respect to either load or position.
This test is designed to enable the user to decide how accurately objects must be positioned on
the pan for this effect to be negligible. If the measured errors are large compared with the
manufacturer’s specification, the balance is probably in need of repair.
The error is most easily measured by placing a weight on the centre of the pan and then
lifting and placing it successively to front, rear, left and right positions on the pan, noting the
reading each time. Standard weights are not required; any clean and reasonably smooth piece
of turned brass or stainless steel is adequate, provided it has roughly the same shape as a
standard weight.
The test should be carried out at the load recommended by the manufacturer or, if this is
not known, at a load between one-third and half the maximum capacity of the balance. Most
balance manufacturers recommend a one-third load or quote performance figures at this load.
Placing a much larger weight near the edge of the pan could possibly damage the linkages
between pan and sensor, and will not necessarily give a larger reading.
A single weight should be used for this test, and this restricts the choice to values of 1, 2,
5 and perhaps 3, unless special weights or objects are available. For example, a balance of
1200 g capacity should be tested with a 500 g weight, because 200 g is too small.
When moving the weight to the front, back, left or right, lift and place it so that the outer
edge of the weight is approximately aligned with the edge of the pan, without tilting the
weight. If this is not possible (if, for example, the pan curves upwards near the edge) move the
weight as far from the pan centre as possible without tilting it.
The measurement of the off-centre loading error will be subject to the uncertainty arising
from balance non-repeatability. If the repeatability of measurement Sr is similar to or larger
than the off-centre loading error, then the latter will be difficult to measure accurately but its
exact value will be of small importance anyway.
For any kind of dual or multi-range balance, this test should be carried out in the same
way as for single range balances. Thus it should be done at the load recommended by the
manufacturer or at a load between one-third and half the maximum capacity of the balance. If
the balance has a high-resolution range then this range should, if possible, be moved to include
the load at which this test is carried out. This is normally done by retaring the balance.
6.3.5 Hysteresis
One should test for hysteresis when a balance is calibrated for the first time or after a major
repair. After that, it normally need be measured only a few times during the life of the
balance. A properly adjusted balance in good condition should show no more hysteresis than 1
count. If the balance shows more hysteresis than this, then the user should consult the
manufacturer.
Testing at one point, about mid-range, is adequate. Proceed as follows:
(i) place weight(s) of mass M, equal to approximately half the balance range, on the pan and
note the reading,/?;;
(ii) add extra weight(s) of mass M' to bring the balance reading close to full scale;
(iii) remove the weight(s) and read the balance with the weight(s) M still on the pan, qi\
(iv) remove all weights from the pan, and then replace weights M and M'.
(v) remove the weight(s) M' and read the balance with the weight(s) M still on the pan, q2\
(vi) remove all weights from the pan, replace weight(s) M and note the reading, p2’,
(vii) remove all weights from the pan.
Repeat these steps to obtain a second set of readings ps, q3, q4 and P4. The hysteresis is then
given by:
Hysteresis = + P3 + A) fai + ^2 + ^3 + ^4)
Note that if the readings are approximately equally spaced in time, the effects of any
drift in the balance readings will be largely eliminated. If the balance has a high-resolution
range then it should be used, if possible, when carrying out this test (by retaring the balance at
load M if necessary).
In the absence of drift, a simpler approach to measuring hysteresis is possible [1]:
(i) zero the balance, zy;

(ii) place a weight M, equal to half the balance range, on the pan, r;/
(iii) add extra weight M' to bring the balance reading close to full scale;
(iv) remove the extra weight M' and read the balance with M still on the pan, r2/
(v) remove M and read the zero, Z2.
Carry out the above procedure three times and calculate the average differences: (r; - r2}
and (zy- Z2). Take the hysteresis to be (ry- rl) provided it is numerically at least twice as large
as the drift (zy-Z2).
For dual or multi-range balances where the measuring system changes from range to
range, the hysteresis should be measured for each range. For a given range, mass M now
corresponds to the mid-point of the range, while mass M + M' is now close to the top of the
range.
6.4 Recording and reporting the results of an electronic balance

calibration
Examples of the tests described in Section 6.3 are given below, along with a sample report
form. All examples are for a single-range balance with a maximum capacity of 1200 g and a
discrimination of 0.01 g. The procedures are easily extended to multi-range balances.
A class F2 weight set is used to determine the corrections to balance reading. A

calibration report gives the masses and corresponding uncertainties of these weights to be
those shown in Table 6.1, with a coverage factor of 2.04 for the uncertainties. The calculations
could be made using either a calculator or a spreadsheet program. However a properly
checked and protected spreadsheet is less susceptible to error.
Table 6.1: The masses and corresponding uncertainties of the weights used in the example of an
electronic balance calibration. The uncertainty coverage factor is 2.04. (200' indicates a second weight
having a nominal mass of 200 g.)
Calibration
Nominal mass (g) Mass (g) Uncertainty (g)
100 100.000 1 0.000 5
200 199.997 7 0.000 9
200' 200.002 2 0.000 9
500 500.003 9 0.002 2
1000 1000.005 0 0.003 0
6.4.1 Repeatability of measurement (example)
The repeatability would normally be measured at full and half maximum capacity (1200 g and
600 g). However it is more conveniently measured at 1000 g and 500 g and this is acceptable
as the load difference is only 17%. (A 1500 g balance could not be tested at 1000 g and 500 g,
as 1000 g is closer to half load than full load. Test such a balance at 1500 g and 700 g, perhaps
using specially made weights.) In the table of measurements below, M is nominally 1000 g
and all data are in grams.
Measurement
Number Pan Load Reading Mi = ri-Zi
1 0 0.00 (Z;)
M 1000.03 {ri) 1000.03
2 0 0.00
M 1000.03 1000.03
3 0 0.00
M 1000.03 1000.03
4 0 0.00
M 1000.04 1000.04
5 0 0.00
M 1000.03 1000.03
6 0 0.00
M 1000.03 1000.03
7 0 0.00
M 1000.03 1000.03
8 0 0.00
M 1000.03 1000.03
9 0 -0.01
M 1000.03 1000.04
10 0 0.00
M 1000.03 1000.03
6.4 RECORDING & REPORTING THE RESULT.
Using equations (6.1) and (6.2), the repeatability of measurement Sr is calculated to be

0.0042 g, and the maximum difference between successive measurements is 0.01 g. A similar
set of measurements are made at a nominal load of 500 g, and for the purpose of these
examples it is assumed that the repeatability obtained at that load is 0.0000 g.
6.4.2 Corrections to balance reading (Example)
Difference Correction
1 Pan load Reading Means n - Zi
0 0.00
100.000 {Ml} 100.00
100.000 100.00 100.00 (r;) 100.00 0.000
0 0.00 0.00 (Z;)
199.998 200.01
199.998 200.01 200.01 200.01 -0.012
0 0.00 0.00
299.998 300.01
299.998 300.01 300.01 300.01 -0.012
0 0.00 0.00
400.000 400.02
400.000 400.02 400.02 400.02 -0.020
0 0.00 0.00
500.004 500.03
500.004 500.03 500.03 500.03 -0.026
0 0.00 0.00
600.004 600.03
600.004 600.04 600.035 600.03 -0.026
0 0.01 0.005
700.002 700.05
700.002 700.04 700.045 700.035 -0.033
0 0.01 0.01
800.002 800.05
800.002 800.05 800.05 800.04 -0.038
0 0.01 0.01
900.004 900.05
900.004 900.05 900.05 900.04 -0.036
0 0.01 0.01
1000.005 1000.04
1000.005 1000.04 1000.04 1000.03 -0.025
0 0.01 0.01
1100.005 1100.03
1100.005 1100.03 1100.03 1100.02 -0.015
0 0.01 0.01
1200.007 1200.02
1200.007 1200.02 1200.02 1200.015 -0.008
0 0.00 0005
All ^ta in the table are in grams. The uncertainties of these corrections are calculated in
Section 6.4.5.2 and Chapter 8.
56 6 CALIBRA TION OF ELECTRONIC BALANCES
6.4.3 Effect of off-centre loading (example)
As a single weight should be used for this test, its nominal mass must be 500 g as 200 g is too
small. However a suitable piece of metal of approximate mass 400 g could be used (see
section 6.3.4). The readings are in grams.
Back: 501.17
Left: 501.18 Centre: 501.18 Right: 501.19
Front: 501.20
Maximum difference = 0.03 g
6.4.4 Hysteresis (example)
Two weights of mass M and M' are used, where both masses are nominally 500 g. The
headings 1 and 2 refer to the first and second set of readings, and all data are in grams. The
readings shown in brackets need not be recorded.
Pan load 1 2
NI {pi, pL} 500.03 500.03
M + M' (1000.04) (1000.05)
(^;, ql} 500.03 500.03
Zero (0.02) (0.02)
(1000.04) (1000.04)
q^ 500.03 500.02
Zero (0.02) (0.02)
A/ (P2, p^ 500.03 500.03
The hysteresis is therefore
(500.03 + 500.03 + 500.03 + 500.03) - (500.03 + 500.03 + 500.03 + 500.02) 0.01

0.0025
4 4
6.4 RECORDING & REPORTING THE RESULT...... 57
6.4.5 Uncertainties associated with the calibration and use of an electronic

balance
A full treatment of the uncertainties associated with balance calibrations is given in Chapter 8.
However a summary of these uncertainty calculations, for the sample calibration detailed
above, is given here in order to provide the uncertainties that will be quoted in the sample
report of section 6.4.6.
A calibration report for an electronic balance should quote the following:

(a) The uncertainties of the corrections.
(b) The Limit of Performance.
(c) The uncertainty of weighing (if requested by the user).
The uncertainties of the corrections are the uncertainties with which the corrections to
balance readings have been determined. The uncertainty of weighing is the uncertainty of a
mass value that is obtained after applying corrections to balance readings, as in equation (6.7).
In general, it will vary with the load on the balance.
The Limit of Performance, F, of a balance is a quantity that has been in use for many
years and is not an uncertainty in the normally accepted sense. Instead, it gives a very
conservative upper limit to the measurement error that may be expected when no corrections
are made to balance readings. Thus if r and z are the balance readings, the mass of the object
lies in the range (r — z) ± F. The Limit of Performance is an upper and lower bound; in many
applications the balance may weigh much more accurately than is indicated by this figure.
The uncertainties of the corrections and the uncertainties of weighing are expanded
uncertainties that must be calculated by combining their component uncertainties as described
below, but the Limit of Performance F is simply calculated from the equation:
= 2.26«^ + ) (6.5)
where:
Umax is the maximum value of the repeatability of measurement of the balance or one-
third of the discrimination of the balance, whichever is greater.
C„ax is the magnitude of the maximum correction to balance reading C
U(Cmax) is the expanded uncertainty associated with
A client may request that the Limit of Performance be determined not for the full range
of the balance, but for a restricted range corresponding to the usual weighing range (which
may, for instance, be only up to 70% of full range). This could result in a better value for F,
and is acceptable provided the restriction is made clear in the report. For multi-range balances,
F must be determined for every range.
6.4.5.1 Calculation of the Limit of Performance
In section 6.4.1, the maximum value for the repeatability of measurement was found to be
0.0042 g at a load of 1000 g. As this is greater than one-third of the discrimination, uâx is
taken to be 0.0042 g.
The table in section 6.4.2 shows the maximum correction to the balance reading to be -
0.038 g (at a load of 800 g), while the uncertainty of this correction is found, in section 6.4.5.2
below, to be 0.011 g. Hence the Limit of Performance of the balance is
F = 2.26x0.0042 + 0.038 + 0.011 = 0.059 g
6.4.5.2 Calculation of the uncertainty of corrections and uncertainty of weighing
The estimation of these uncertainties is not difficult, but requires much more extensive
calculations than those given above for the Limit of Performance. A brief summary of the
calculations, including the spreadsheet tabulations for a load of 800 g, is given here.
To identify the component uncertainties that must be combined to calculate the

expanded uncertainty of the quantity of interest, the measurand, one should look at the
equation expressing the measurand in terms of the input quantities. For the corrections, for
instance, this is equation (6.3) which shows that we should be looking for any factor that has a
bearing on the uncertainties of M, r or z. These components are listed in Table 6.2 for the
uncertainty of the correction, and in Table 6.3 for the uncertainty of weighing.
For each component uncertainty shown in the tables, Ui is an estimate of its effect in the
form of an expanded uncertainty (for the calibrating mass value, for instance), a standard
uncertainty (eg, the calibrating mass instability) or the semi-range of a rectangular distribution
(eg, the balance resolution). To obtain the corresponding standard uncertainty one must
divide tT, by the reducing factor which is the coverage factor for an expanded uncertainty, or
^3 (1.73) for a rectangular distribution semi-range. In general, one also needs to multiply Ui
by the sensitivity coefficient but in these tables all the c, are equal to one.
For each component uncertainty, an estimate of the degrees of freedom v; is also

required. The combined standard uncertainty is then obtained by combining the components
in quadrature (equation L.3), while the corresponding degrees of freedom are obtained
using the Welch-Satterthwaite formula (equation L.6). From one can obtain (from
equation L.8) the coverage factor k used to obtain the expanded uncertainties U.
These calculations are carried out in the spreadsheets shown in Tables 6.2 and 6.3, while
Table 6.4 gives the uncertainties of correction and uncertainties of weighing calculated in this
way across the full range of the balance.
6.4 RECORDING AND REPORTING THE RESULTS 59
Table 6.2: Spreadsheet for calculating the uncertainty of correction for an electronic balance at a load
of 800 g.
Load: 800g
Component Values
Component (/ = 1,2,... 8) D, (mg) ki Ci Ui (mg) Vi Ui^ uNvi
1. Mass value for 800 g 3.6 2.04 1 1.765 30 3.114 0.323
1 2. Instability in 100 g
0.12 1 1 0.120 4 0.014 5.2E-05
3. Instability in 200 g 0.24 1 1 0.240 4 0.058 8.3E-04
4. Instability in 500 g 0.60 1 1 0.600 4 0.360 0.032
5. Repeatability in r- z 2.97 1 1 2.970 9 8.821 8.645
6. Resolution of r 5 1.73 1 2.890 1000 8.353 0.070
7. Resolution of z 5 1.73 1 2.890 1000 8.353 0.070
8. Rounding of the final result 0.5 1.73 1 0.289 1000 0.084 7.0E-06
1
pombined standard uncertainty Uc (mg) 5.40
Degrees of freedom Veir 93
[Expanded uncertainty U (mg)__________ 10.73
Table 6.3: Spreadsheet for calculating the uncertainty of weighing for an electronic balance at a load
of 800 g
Load: 800 g
Component Values
Component (/ = 1, 2,... 5) U(mg) ki Ci Ui (mg) Vi ul ui*/vi
1. Uncertainty in C 10.73 1.99 1 5.392 93 29.073 9.089
3. Resolution of r 5 1.73 1 2.890 1000 8.353 0.070
4. Resolution of z 5 1.73 1 2.890 1000 8.353 0.070
Combined standard uncertainty Uc (mg) 7.97

Degrees of freedom Ven 92
Table 6.4: Expanded uncertainties for the corrections and weighings for the electronic balance
Uncertainty of Uncertainty of Uncertainty of Uncertainty of

Load correction weighing Load correction weighing
(g) (mg) (mg) (?) (mg) (mg)
100 8 12 700 11 16
200 8 12 • 800 11 16
300 8 12 900 11 16
400 9 12 1000 11 16
500 9 12 1100 11 16
600 11 16 1200 11 16 1
6.4.6 Sample report for an electronic balance
A sample report for an electronic balance appears on the following pages.


Bradfield Road, West Lindfield NSW 2070, Australia
Phone: (612) 9413 7000 Fax: (612) 9413 7631
REPORT ON
An Electronic Balance
Report No. RN 12345

Maker Mettler Model P 1200 Serial No. 13254
Capacity 1200 g Resolution (least digit) 0.01 g Type Top-loading
Client CSIRO-NML, Bradfield
j Road, West Lindfield NSW 2070
Examined at Room C20E, National Measurement Laboratory
Examined by E Morris
Calibration Date: 28 Nov 2001 Date of Issue of report: 4 Dec 2001
Temperature during test 20.4°C
Before calibration, the balance was adjusted according to the manufacturer’s instructions.
Repeatability of Measurement
1 Load Repeatability of Maximum difference between 1

(g) measurement (g) successive measurements (g)
500 0.0000 0.00
1 1000 0.0042 0.01
Corrections to balance reading
Load Correction Uncertainty (±)

(g) (g) (g)
100 0.000 0.008
200 -0.012 0.008
300 -0.012 0.008
400 -0.020 0.009
500 -0.026 0.009
600 -0.026 0.011
700 -0.033 0.011
800 -6.038 0.011
900 -0.036 0.011
1000 -0.025 0.011
1100 -0.015 0.011
1200 -0.008 0.011
Off-centre Loading
A weight of mass 500 g was moved to various positions on the pan. Except for the Centre
position, the weight was placed as close as possible to the edge of the pan without passing
beyond the edge. The balance readings (in grams) at each position, relative to the centre
reading, are given in the table below
Centre Front Back Left Right Maximum 1

difference |
0.00 +0.02 -0.01 0.00 +0.01 0.03 1
Hysteresis
Load Hysteresis
(g) (g)
500 < 0.01
Uncertainty of weighing of the balance
Load Uncertainty (±) Load Uncertainty (±)

(8) (g) (g) (g)
100 0.012 700 0.016
200 0.012 800 0.016
300 0.012 900 0.016
400 0.012 1000 0.016
500 0.012 1100 0.016
600 0.016 1200 0.016
Limit of Performance for the balance: 0.059 g
Notes
1. The balance has been tested according to procedures detailed in The calibration of weights
and balances by E C Morris and K Fen, National Measurement Laboratory, Lindfield,
NSW, Australia
2. When the sign of the correction is positive (+) the amount should be added to the balance
reading to give the correct mass value and when negative (-) subtracted from it.
6.4 RECORDING AND REPORTING THE RESULTS......... 63
3. Any corrections for air buoyancy should be calculated by assuming that the object being
weighed is balanced against hypothetical weights of density 8000 kg.m’^ in air of measured
density.
4. The Limit of Performance is the upper limit of the measurement error that may be expected
when no corrections are made to the balance readings.
5. The uncertainty of weighing is the uncertainty of a single measurement of mass that is

obtained after applying the corrections given in this report to the balance readings.
6. Each uncertainty has been calculated using the ISO Guide to the Expression of Uncertainty
in Measurement and is expressed by the stated interval, which is estimated to contain the
measurand with 95% probability. The coverage factor for the interval is 2.0.
Dr N Bignell
For Dr B D Inglis
Director, National Measurement Laboratory
6.5 Using a calibrated balance
6.5.1 The normal procedure for measuring mass with a calibrated electronic
balance
There are various ways of using a calibrated balance, but in essence they all reduce to the
following:
• Allow the balance to warm up for the period recommended by the manufacturer, or at least
30 minutes if the recommended warm-up period is not known. Exercising the balance by
by placing a weight, of mass approximately equal to full load, on the pan is recommended.
• The object or material to be weighed should be placed close to the balance until
approximate temperature equality with the balance is attained (see Table B2 for the times
required for weights). However close thermal equality is not very important for low
precision (>30 ppm) weighing.
• With the balance pan empty, note the zero reading, zj. (The balance may be fared before
this reading.)
• With the object to be weighed on the pan, note the reading, rj.
• One may make a second reading after lifting the object off the pan and replacing it, r2
• Remove the object from the pan. If the balance readings are drifting, note the zero reading
again,
• Take the zero reading z to be zj unless a second zero reading is made, in which case
calculate the average zero reading: z = (z; + zD/l.
• Take the reading r to be r; unless a second reading is made, in which case calculate the
average reading: r = (r; + rl)!!.
• If the balance Limit of Performance F is less than the maximum allowable uncertainty for
the measiu-ement, then the measured mass may be simply taken as
Af = r - z (6.6)
• Alternatively, the measured mass is obtained from equation (6.7) where Cm is the correction
to balance reading at load M. In this case, the uncertainty of the measurement may be
quoted as ± U, where U is the uncertainty of weighing of the balance at load M.
M = r-z + C,, (6.7)
The mass value obtained from a calibrated balance is the conventional mass. The value
of the true mass is needed only in relatively rare circumstances (see Section 2.5) and it can be
obtained from the measured conventional mass using equation (2.6). The difference between
true and conventional mass is explained in Chapter 2.
6.5 USING A CALIBRATED BALANCE 65
6.5.2 User adjustment and tests on balances
6.5.2.1 Adjustment
A balance should be adjusted at least monthly when in regular use (see Section 6.3.1). This
ensures that the corrections to balance reading will be as close as possible to those obtained
when the balance was calibrated.
6.5.2.2 Repeatability test
The repeatability of measurement Sr should be determined every six months (see Section 6.3.2)
to ensure that the balance has not developed a defect that will affect mass measurements. A
single measurement near the top of the usual weighing range is sufficient for single-range
balances. For multi-range balances, one determination in each range is recommended.
The values obtained should be recorded so that the latest result .s>(latest) can be
compared with the previous result ^^(previous). Ideally, this test should always be carried out
by the same operator.
Assuming that the repeatability is always measured with nine degrees of freedom, then
^Xlatest) should not exceed 1.785,(previous). Furthermore, .s'/lalest) should not be greater than
twice the repeatability given in the most recent calibration report for the balance. If these
requirements are not met, then there is a fairly strong indication that the performance of the
balance has deteriorated. The user should consult the calibrating laboratory to see if
calibration, servicing or some other action is required.
Note that ^y measurement of Sr yielding a value of zero should not be used as the value
of 5^revious), in such cases use an earlier value, if available. If no earlier non-zero value of
Sr is available, the manufacturer’s specification could be used.
A more sophisticated version of this test is possible, in which the previously measured
values of the repeatability are pooled. See Appendix G for details, and note that it may not be
advisable to include measurements made before the most recent balance calibration in the
pool. This version of the test gives a better indication of any problem arising with the balance.
6.5.2.3 A check on the correction to reading
Using a suitable calibrated weight of mass M, the correction to reading at load M should be
measured following the procedure of section 6.3.3 and using equation (6.3). Ideally M will be
at least one-third the full balance range and its uncertainty Um will be small compared with the
uncertainty of correction. This check should be carried out monthly at a single point, though
multi-range balances should be checked at one point in each range.
The difference AC between the correction given in the latest calibration report and the
value obtained in this test should be numerically less than the combined uncertainty of its
measurement. Using an approximation that is good enough for the purposes of this check, it is
required that
|zlC| < Û^+2s^^+Ul, (6.8)
where U is the uncertainty of the correction at load M. If this condition is not satisfied, then
there is a fairly strong indication that the balance needs recalibration.
6.5.3 Air buoyancy and weighing
All objects exposed to the atmosphere are subject to buoyancy, an upwards force equal to the
gravitational attraction acting on the mass of air displaced by the object. Thus the buoyancy
varies with the volume of an object and hence, for objects of the same mass, inversely as their
density. Since the corrections to balance reading given in a report were obtained using
standard weights with a density of approximately 8000 kg.m they will be in error when
weighing materials of different density, because of the different buoyancy. This buoyancy
error will be small for dense materials (eg, brass, steel, lead) and greatest for materials with a
density of 1000 kg.m*^ or less (eg, water, oil, wood, cork). Errors are also introduced by
variations in air density. Fortunately the accuracy required when weighing light materials is
usually low enough to allow these errors to be ignored.
If an object of mass M having a density p is weighed in air of density Pa on an

electronic balance that was calibrated at a time when the air density was A(cal), then the
correction that must be added to the measured mass to account for the buoyancy error is
\ p^-1.2 pfcal)-1.2l
SM= —---------- ------ M (6.9)
p 8000 J
(Generally the density of air need only be determined with an uncertainty of 1%, which is
achieved by measuring the air temperature, pressure and humidity with instruments of modest
accuracy. See Appendix H for further details.)
For a laboratory with reasonable temperature control, the air density typically varies over
a total range of ± 0.04 kg.m’^ from the average value, due largely to changes in atmospheric
pressure. The magnitude of the buoyancy correction is best seen by considering the examples
6.5 USING A CALIBRA TED BALANCE 61
6.5. The shnded rows show worst-case scenarios, while the unshaded rows show
situations that could easily occur in practice.
Table 6.5; Examples of the buoyancy correction for an electronic balance, calculated from
equation (6.9). All symbols have the same meaning as in that equation. The shaded rows show worst-
case scenarios.
1 AJtitude Material p p.(eal) a^f/M

pa
(m) weighed (kg-m’) (kg.m'^) (ppm) 1
0 brass 8400 1.16 1.24 9.8
0 brass 8400 1.184 1.216 3.9
0 water 1000 1.16 ^1.24 45
0 water 1000 1.184 1.216 18
700 brass 8400 1.14 1.06 -9.2
700 brass 8400 1.116 1.084 -3.3
700 water 1000 1,14 1.06 -133
1 700 water 1000 1.116 1.084 -106
Notice that for a dense material such as brass, the correction is relatively small and
varies only moderately with altitude. For water on the other hand, the correction is much
larger and can increase rapidly with altitude. When weighing water on a calibrated electronic
balance at an altitude of 700 m, the buoyancy correction will usually be in the order of 0.01%.
Lighter materials such as wood will require even greater corrections.
If it is necessary to weigh light materials with a mass uncertainty in the order of 0.05%
or less, estimate the likely size buoyancy correction as follows: For the altitude of the
laboratory, find the mean value of a ; Table J.l may be used for this purpose. Then substitute
this value into equation (6.9) for both Pa and A(cal), while putting p equal to the density of
the material being weighed. For instance. Table J.l indicates that the mean laboratory air
density at 700 m is 1.099 kg.m'\ Substituting this value into the equation for pa and A(cal),
and putting p equal to 1000 kg.m^, one obtains 0.009% for the estimate of the buoyancy
correction when weighing water.
If the correction calculated in the previous paragraph represents a serious error if

ignored, then corrections determined from equation (6.9) will have to be applied.
Consequently the air density at the time of the balance calibration (as well as at the time of
mass measurement) should be known; preferably it will be stated on the balance calibration
report.
68 6. CALIBRATION OF ELECTRONIC BALANCES
6.6 Measuring corrections to balance reading when only one

calibrated weight is available
Using this procedure, it is necessary to have only one calibrated weight. It is necessary
however to have several uncalibrated weights, which are referred to below as tare weights.
The calibrated weight should have a mass M that is approximately one-tenth the range of the
balance and the imcertainty associated with M should be less than the resolution of the balance.
The procedure is as follows:

(i) read the zero;
(ii) place the calibrated weight on the pan and note the reading rf,
(iii) remove the weight and read the zero — this reading is averaged with the reading in (i) to
give the zero value, zf,
(iv) place tare weights on the pan until a reading close to r; is displayed. This effectively
becomes a ‘zero’ reading, like that in (i);
(v) add the calibrated weight to the pan and note the reading rf,
(vi) remove the calibrated weight and read the ‘zero’— this reading is averaged with the
reading in (iv) to give the zero value, 22;
(vii) place additional tare weights on the pan until a reading close to r2 is displayed.
(viii) add the calibrated weight to the pan and note the reading
(ix) remove the calibrated weight and read the ‘zero’— this reading is averaged with the
reading in (vii) to give the zero value, zj;
(x) continue in this way until the capacity of the balance is reached.
The additional balance correction for each step is given by
c,. = M - (r. - z,.) (6.10)
and the total correction for the load iM is found by adding the first i additional corrections:
C,- = Cj + Cj + ... + Cj (6.11)
Multi-range balances should be treated in the manner described at the end of Section 6.3.3.
An example of this test as far as the fourth correction is given in the table below. The
calibrated weight has a mass M of 100.001 g, and all data in the table are in grams. As in the
example given in section 6.4.2, two readings are taken with the calibrated mass on the pan,
while the quantities Mi, M2, etc refer to the tare weights.
Uncertainties are calculated in Section 8.1.1.3.

6.6 MEASURING CORRECTIONS TO BALANCE READING WHEN 69
Additional
Pan load Reading Means Difference Correction Cumulative
rt-Zi M- (ri-Zi) Correction
0 0.00
100.001 100.00
100.001 100.00 100.00 (r;) 100.00 0.001 0.001
0 0.00 0.00 (2;)
Mj 100.02
A/;+100.001 200.03
M;+l 00.001 200.03 200.03 100.01 -0.010 -0.009
Mi 100.02 100.02
200.05
M2+IOO.OOI 300.05
M2+100.001 300.06 300.055 100.005 -0.004 -0.013
M2 200.05 200.05
Ms 300.01
M5+100.001 400.01
M5+IOO.OOI 400.01 400.01 100.00 0.001 -0.012
Ms 300.01 300.01
NR6CA/
70 6. CALIBRA TION OF ELECTRONIC BALANCES
Chapter 7
Calibration of single-pan, two-knife-edge

balances
7.1 What is a single-pan two-knife-edge balance?

A single-pan two-knife-edge balance is one that compares the mass of the object on the
balance pan with the mass of built-in weights. The dials move the built-in weights off the pan
support to restore balance. An optical scale displays the small mass difference between the
object and these weights. The mass of the object being weighed can be read directly from the
dial readings and the optical scale. Sometimes the optical scale is replaced with an electronic
readout. Although there may be a digital readout, it is not an electronic balance as described in
Chapter 6. The optical scale or the electronic display is normally less than 1% of the full range
of the balance.
See Appendix K for more details of the construction of single-pan two-knife-edge balances.
7.2 The scope and purpose of the calibration

The main purpose of a balance calibration is an estimation of the accuracy of measurements
made on the balance. However balance calibrations have three other important functions:
(a) A table of corrections to the balance readings is provided.

(b) The user should have information on the limitations of the balance, principally the effects
of off-centre loading and hysteresis. This information sometimes acts as a diagnostic
indicating that repairs may be required.
(c) The balance report should contain information that allows the user to carry out effective
checks on the balance between calibrations. The repeatability of reading is important in
this regard.
71
72 7. CALIBRATION OF SINGLE-PAN, TWO KNIFE EDGE BALANCE
7.3 Balance calibration

Because this type of balance usually has built-in weights and compares gravitational forces,
moving from one place to another should have little effect on the balance reading.
Nevertheless, it should be calibrated in the position where it is to be used to account for
environmental changes. It should not be sent to a calibrating laboratory for calibration. If a
balance is moved after calibration, the user checks of Section 7.5.2 should be carried out to see
whether a recalibration is necessary.
Where calibrated weights are used in the calibration, the density of the weights should
be in the vicinity of 8000 kg.m'^. The accuracy of the balance determines how close the
density must be to this value. This is usually not a problem provided the weights are made of
stainless steel, brass or bronze, but to be quite sure that the weight density is acceptable, use
weights conforming to a standard, such as the OIML classification [3].
7.3.1 Adjustment
Check and adjust the balance following the manufacturer’s instructions. This process is usually
manual and should be done monthly if the balance is in regular use. It is done by zeroing the
balance and placing a weight equal to the full optical scale value on the pan and adjusting the
optical scale so that it displays its full range value. The adjustment can be done at any load and
a zero load is chosen for convenience.
7.3.2 Repeatability of measurement
Repeatability is a measure of how well a balance will weigh. The repeatability is expressed in
terms of the standard deviation obtained from a series of repeated readings together with the
maximum difference between two successive readings. For a good balance these should not
exceed three times the resolution. However the readings must be obtained in a way that
realistically reflects how a balance is used in practice. These balances weigh at approximately
constant load so there is no need to perform repeatability tests with weights over the full range
of the balance. The repeatability should be measured with weights near half the capacity of the
balance and at both ends of the optical scale. This is to account for the change in effective
shape of the knife edges.
The most realistic way to assess the repeatability is to remove and replace the weight on
the pan for each load tested. This causes a greater disturbance to the balance than merely
releasing and arresting it. The standard deviation Sr should be calculated from n measurements
of BMi, the difference between the zero optical scale reading of the balance (z,) and the reading
with a weight on the pan (r,), i.e.
SMi = ri-Zi. (7.1)
Thus if the average value of SMi is SM ,

7.5 BALANCE CALIBRA TION 73
i=l ’
n -1
(7.2)
The recommended minimum number of measurements (i.e., the minimum value of n in

equation 7.2) is 10, in which case Sr has nine degrees of freedom.
For a single-pan, two-knife-edge balance, it is sometimes the case that Sr is less than the
resolution of the optical scale. As a result, the SMi values obtained by the above procedure
may be all the same and consequently Sr would be zero. The measured value of Sr should
nevertheless be recorded in the balance report.
7.3.3 Corrections to balance reading
Corrections may need to be applied to the balance readings, which are the combination of the
dial readings and the optical scale readings. Even though a single-pan two-knife-edge balance
has been adjusted to read correctly close to the full optical scale, it may still be necessary to
apply corrections to the readings, for two reasons:
(a) The optical scale reading as function of load may be significantly non-linear.
(b) The mass of the built-in weights may differ significantly from their nominal values. (The
corrections to the dial readings are actually the departure of the mass of the built-in
weights from their nominal values.)
When corrections are not needed, one should know whether the balance readings con-
form to the tolerances claimed by the manufacturer, or whether the balance is accurate enough
for its intended use.
7.3.3.1 Optical scale
After the initial set up of the balance as in 7.3.1, the optical scale range should be corrected at
its full range. The purpose of calibration is to measure the correction that needs to be applied
to the balance reading. Readings at 5 points on the optical scale are adequate. A set of
calibrated weights with an uncertainty less than the resolution of the balance is required for the
calibration.
The corrections should be determined at a number of points, say 5, over the range of the
optical scale by successively placing small calibrated weights on the pan and observing the
optical scale readings. The calibrated weights should have nominal mass values equal to
multiples of one fifth of the full optical range. The steps are:
(i) with all dials set to zero, release the balance and note the first zero reading;
(ii) place the weight(s) of mass Mi (where Mi is approximately one fifth of the full optical
range for the first step) on the pan and note the reading
(iii) remove the weight(s) momentarily, then replace them on the pan and note the new
reading. This reading is averaged with the one in (ii) to give, rf,
(iv) remove the weight(s) and re-read the zero — this reading is averaged with the reading in
(i) to give the zero reading for the calculation, zy;
(v) repeat steps (ii) and (iv) to cover the whole optical scale.
If reading r, and z, correspond to weight(s) of total mass A/, on the pan, then the
correction to the optical scale reading for this load is given by
C. = M. -(r. -z,.) (7.3)
The correction is the mass value that should be added to the optical scale reading to give
the correct mass when a reading is made at that point of the optical scale.
7.33.2 Dial readings
All single-pan, two-knife-edge balances have weights installed in the balance. Some balances
have three or even four decades of weights and a full calibration can be very time consuming.
Removing the built-in weights from the balance for calibration is generally not recommended
as it will risk scratching the weights during replacement.
There are a number of possible methods of performing a calibration on these built-in

weights in situ. These will be described in the following sections.
(i) Simple tolerance test — check the conformance to manufacturer's tolerance, simpler than
comprehensive tolerance test. In use, no correction is applied during mass measurement.
(ii) Comprehensive tolerance test — check the conformance to manufacturer's tolerance,

using the method proposed in ASTM standard E319 [8]. In use, no correction is applied
during mass measurement.
(iii) Direct calibration — easy and straightforward method. Corrections are applied to the
balance readings during mass measurement.
(iv) Least-squares calibration — most accurate method. Corrections are applied to the balance
readings during mass measurement.
In the following sections, the simple tolerance test and the direct calibration method will
be described. For the user who requires more accurate mass measurement, corrections should
be applied to the balance readings and least-squares calibration should be used for balance
calibration. Comprehensive tolerance test and least-squares calibration will be discussed at the
end of this chapter and in Appendix E respectively.
The calibration requires that the combination of weights used for a particular load must
have an uncertainty not greater than the standard deviation of the balance.
Note that when calibrating a balance with built-in weights it is very important to
distinguish between the value indicated by the balance (or the value of the built-in weights),
and the reading obtained when a standard weight is placed on the pan. An example will help
make this clear.
Consider a dial setting of 20 g. During calibration, a 20 g standard is placed on the pan

and the balance reads 20.01 g. If a standard of mass 19.99 g were placed on the pan the
balance would read 20.00 g. Therefore the true value associated with the 20 g dial setting is
19.99 g, not 20.01 g. In other words, the balance has a correction of -0.01 g at 20 g. The
actual value of the built-in weight (or weights) is also 19.99 g. This is apparent because
placing 19.99 g on the pan and lifting the ‘20 g’ built-in weight off the pan leaves the reading
unchanged.
In reporting the result of the calibration of the built-in weights, it is important to give the
mass value for the dial setting, or the corresponding correction. Here,
The correction = value of the standard - balance reading
= 20.00-20.01 =0.01 g.
In either case the user has merely to add up the true values for the dial settings or the
corrections used to obtain the correct mass of the object being weighed. In the above example
the reading of 20.01 g should not be given in the Report as this will almost certainly result in
the wrong value being used in calculating the mass.
When more than one dial setting is used, which is the usual case in practice, the correc-
tions or mass values for each dial reading are added together to give the total correction or
value, respectively.
7.3.3.2.1 Simple tolerance test
In general, there are very few balance users who actually apply corrections for the dial
readings. Most want to know whether the corrections to die dial readings conform to the
tolerances laid down by the manufacturer, or whether the balance is sufficiently accurate for
their purpose without applying corrections.
The quickest way to check the tolerances is to set the dial reading so that all the built-in
weights from one decade are lifted from the beam (eg. position 9 or 90 etc.), place standards of
equivalent value on the pan, and read the difference. In some combinations not all the weights
from the same decade are lifted off at position 9 and a second dial setting is required to ensure
that this happens. It is possible (although unlikely) to have two large errors that cancel. This
method only gives one correction for each decade of the built-in weights. The maximum
correction from the measurements of all decades is reported. Therefore it is used for checking
the conformance of the balance rather than applying corrections. If corrections are applied, the
methods outlined in 7.3.3.2.2 or Appendix E should be used.
7.3.3.2.2 Calibration of each dial setting by direct calibration
This involves measuring the correction for each dial setting by using appropriate weights from
a calibrated set. The method is described in the following steps:
(i) with all dials set to zero and no load on the pan, release the balance and note the
reading, zf,
(ii) set the “1” on the dial to be tested, place a calibrated weight of equivalent value on the
pan, release the balance and note the reading, r;;
(iii) arrest the balance, remove the weight and replace it, release the balance and note the
reading rj;
(iv) return the dial to zero, remove the weight from the pan and read the zero, Z2, this
becomes the first zero reading for the next dial setting, R, to be calibrated;
(v) repeat steps (ii) to (iv) for all the dial settings of the balance.
The correction C to the dial setting is calculated by means of the following formula:
c = ^ -[('■; + )/2 - (z; + z J/2] - R (7.4)

where M is the value of the calibrating weight.
This formula yields the amount that must be added to the balance reading to give the
correct value whenever this dial setting is used. The value of C may be positive or negative.
Usually the mass value of the calibrated weight is close to that of the dial setting and it is not
necessary to apply any corrections. This method does not provide any safeguards against
misreading the balance or arithmetical errors, but it has the advantage of being relatively easy
and straightforward to use,
7.3.4 Effect of off-centre loading
When the centre of mass of the object being weighed is off-centre on the pan, shift or comer-
load error may occur. It is difficult to produce figures that can be used to correct the balance
readings because the effect is not always linear with respect to either load or position. This test
is designed to enable the user to decide how accurately objects must be positioned on the pan
for this effect to be negligible. If the measured errors are large compared with the
manufacturer's specification, the balance is probably in need of repair.
This effect is most easily measured by placing a weight on the centre of the pan and then
lifting and placing it successively to the front, rear, left and right positions on the pan, noting
the reading each time. Standard weights are not required; any clean and reasonably smooth
piece of brass or stainless steel is adequate provided it has roughly the same shape as a
standard weight.
The test should be earned out at the load recommended by the manufacturer or, if this is
not known, at a load between one-third to half the maximum capacity of the balance. Most
balance manufacturers recommend one-third or one-half the maximum load of the balance, or
they quote performance figures at this load. Because this effect is often non-linear.
measurement at a larger load will not necessarily give a larger reading. Also placing a large
weight near the edge of the pan could possibly do some damage to the mechanism.
A single weight should be used for this test, and this restricts the choice to values of 1,
2, 5 and perhaps 3 (unless special weights that can be stacked are available). For example, a
balance of 1200 g capacity should be tested with a 500 g weight, because 200 g is too small.
When moving the weight to the front, back, left or right, lift and place it so that the outer
edge of the weight is approximately aligned with the edge of the pan, without tilting the
weight. If this is not possible (if, for example, the pan curves upwards near the edge) move the
weight as far from the pan centre as possible without tilting it.
The measurement of the off-centre loading error will be subject to the uncertainty
arising from balance non-repeatability. If the former is similar to or smaller than the latter, the
off-centre loading error will be difficult to measure accurately but its exact value will be of
little importance anyway. For this reason, the off-centre loading errors should be taken to be
zero unless the standard deviation of the five measurements (centre, front, rear, left and right)
exceeds the repeatability of measurement Sr by 50%.
7.3.5 Hysteresis
One should test for hysteresis when a balance is calibrated for the first time or after a major
repair. After that, it normally need be measured only a few times during the life of the
balance. In a mechanical system hysteresis is almost always caused by fnction. A single-pan
balance with a suspended pan usually shows no detectable hysteresis. Balances with the
loading pan on top of the balance have more components that can be affected by fnction, but a
properly adjusted balance in good condition should show no more hysteresis than one tenth of
a scale division. If a balance shows more hysteresis, the user should consult the manufacturer.
Testing at one point, about mid-scale (optical scale), is adequate. For convenience, the
dial setting is set to zero. The test can proceed as follows:
(i) place weight(s) of mass M, equal to approximately half the optical range, on the pan and
note the reading, p/;
(ii) add extra weight(s) of mass M' to bring the balance reading close to full optical scale;
(iii) remove the weight(s) M' and read the balance with the weight(s) M still on the pan, qf,
(iv) remove all weights from the pan, and then replace weights M and M'.
(v) remove the weight(s) M' and read the balance with the weight(s) M still on the pan, q2,
(vi) remove all weights from the pan, replace weight(s) M and note the reading, p2,
(vii) remove all weights from the pan.
Repeat these steps to obtain a second set of readings ps, qs, q4 and p4. The hysteresis is
then given by:
78 7. CALIBRA TION OF SINGLE-PAN, TWO KNIFE EDGE BALANCE
Hysteresis = + P2 + P3 + P4 Hg, + + gs + ^4)
Note that if the readings are approximately equally spaced in time, the effects of any
drift in the balance readings will be largely eliminated.
In the absence of drift, a simpler approach to measuring hysteresis is possible [1]:
(i) zero the balance, z/;

(ii) place a weight M, equal to half the optical scale, on the pan, r;/
(iii) add extra weight M' to bring the balance reading close to full optical scale;
(iv) remove the extra weight M' and read the balance with M still on the pan, rj;
(v) remove M and read the zero, zj.
Carry out the above procedure three times and calculate the average differences: (r; - rl)
and (zy - Z2). Take the hysteresis to be (r; - rl} provided it is numerically at least twice as large
as the drift (zy - zj). If the drift is excessive, the measurement should be repeated.
7.3.6 Effect of tare
This is occasionally necessary such as when there is a suspected change in sensitivity. Most
single-pan balances have a taring facility, which enables the balance to be zeroed with an
applied load, usually up to the full optical scale of the balance. The tare device often consists
of a spring, which can be tensioned by the turning of a knob. The sensitivity may change with
load if the tare spring is not correctly positioned.
Any errors in the tare system can be determined by conducting the following test:
(i) place a weight M2 with mass value less than or close to the full optical scale on the pan
and use the tare to zero the optical scale, zy.
(ii) Remove weight M2, place the weight My, which is used to test the full-scale value in
section 7.3.1, on the pan and take the reading ry.
(iii) Remove the weight My and reload it on the pan, take the reading
(iv) Remove the weight My and take the zero reading of the optical scale zj.
The result (ry + r2- zi - z2)/2 is then compared with that given in 7.3.1. If the values
differ by more than twice the balance repeatability (section 7.3.2), then the tare mechanism
should be adjusted.
It is difficult to apply corrections for any error in the tare because the tare indicator is
usually not graduated. If it is necessary to do precision weighing on a balance with such an
error, then the correction to the optical scale should be measured for each tare setting used.
7.4 Recording and reporting the results

To illustrate the recording and reporting of results, we use an example. In this example, a 1kg
single-pan two-knife-edge balance with in-built weights and arrest system is used. It has an
optical scale range of 100 mg with 10 mg divisions. The optical scale has a vernier scale that
can be resolved to Img. A class Fi weight set is used to determine the corrections to the
balance reading.
7.4.1 Repeatability of measurement - (example of 7.3.2)
Repeatabilities are measured at half maximum load (500 g) at both ends of the optical scale.
The procedures are as follows:
i) Release the balance with dial reading set to zero and no load on the pan, take the optical
scale reading, zj.
ii) Set the dial readings to 500 g (take off 500 g of in-built weights from the beam), put the
load M = 500 g on the pan, release the balance and take the optical reading, r;.
iii) Remove the load M and set the dial readings back to zero. This puts the 500 g of in-built
weights back on the beam.
iv) Repeat procedures i) to iii) ten times. Then calculate the differences SMt = r, - z, and the
standard deviation of with 9 degrees of freedom. This is the standard deviation at the
zero point of the optical scale.
v) Repeat i) to iv) with dial readings set to 499.9 g. This gives the standard deviation at full
range of the optical scale.
The same weight (500 g) can be used in the tests at both zero & full optical scale. If the optical
reading falls outside of range of the optical scale (0-100 mg in this case) when the dial setting
is set to 500 g or 499.9 g, the zero adjustment / tare knob can be adjusted so that the optical
readings fall within the range of the optical scale. If the zero adjustment knob has been turned,
procedure 7.3.1 should be checked again after the repeatability test.
Dial Readings : 500 g Dial Readings : 499.9 g

Optical Scale Difference Optical Scale Difference I
Pan Reading SMi = ri—Zi Reading î = ri-Zi
Number Load (mg) (mg) (mg)
1 0 zi 0 zj 0
M ri 1 1 ri 99 til 99
2 0 0 1
M -1 -1 99 98
3 0 1 1
M 2 1 98 91
4 0 1 2
M 1 0 98 96
5 0 1 1
M 1 1 100 99
6 0 2 1
M 2 0 99 98
7 0 2 2
M 3 1 100 98
8 0 2 1
M 1 -1 101 100
9 0 1 3
M 3 2 101 98
10 Q 2 3
M 3 1 100 91
standard deviation at zero standard deviation at full range
end point of optical scale point of optical scale
Sr = 0.91 mg Sr = 1.2 mg
Maximum difference Maximum difference between
between successive readings successive readings I
d =3 mg 1 <SV/'rf=3mg 1
7.4.2 Corrections to balance reading - (example of 7.3.3)
7.4.2.1 Corrections to the optical scale - 7.3.3.1
Optical Scale Optical Scale I

Pan Load Reading Means Difference correction (g) =
(g) (g) (g) (g) Mass - Difference |
p 0.000
Ml (0.020 05) 0.020 Zi 0.000
0.020 ri 0.020 0.020 Ci +0.00005
0 0.000
M2 (0.040 03) 0.040 0.0005
M2 0.041 0.0405 0.0400 +0.00003
0 0.001
M3 (0.059 92) 0.060 0.001
M3 0.061 0.0605 0.0595 +0.00042
0 0.001
M4 (0.079 97) 0.080 0.0015
M4 0.082 0.081 0.0795 +0.00047
0 0.002
Ms (0.100 02) 0.102 0.003
Ms 0.103 0.1025 0.0995 +0.00052
0 0.004
Maximum correction = 0.000 52 g
7.4.2.2 Dial setting - 7.3.3.2
(a) Simple tolerance test 7.3.3.2.1
Pan load Dial Setting Optical Scale Value of Dial Correction

M (g) R (g) Reading (g) M- r + z {g) M- r + z -R (g)
0 0 zi 0.000 0
0.8998 0.9 ri +0.001
0.8998 0.9 rj +0.001 0.8993 -0.0007
0 0 Z2 +0.001
9.00013 9 +0.002
9.00013 9 +0.002 8.99963 +0.00037
0 0 +0.002
90.00054 90 +0.003
90.00054 90 +0.002 90.00004 +0.00004
0 0 +0.002
900.0028 900 +0.004
900.0028 900 +0.005 900.0008 +0.0008
0 0 +0.003 ______
(b) Correction to each dial setting 7.3.3.2.2

Optical Scale Optical Value of
Dial reading at zero Scale Difference Value of Dial Correction
Setting load z reading at r-z (g) Standard M-pF-z)
^(g) (g) load r (g) A/(g) (g) (g)
0 0.000
0.1 0.002
0.1 0.001 0.001 0.10002 0.09902 -0.0001
0 0.001
0.2 0.003
0.2 0.002 +0.0015 0.20008 0.19858 -0.0014
0 0.001
0 0.002
100 0.001
100 0.001 -0.0005 100.0001 100.0006 0.0001
0 0.001
0 0.001
800 0.002
800 0.003 +0.001 800.0009 799.9999 -0.0001
0 0.002
900 H
0.004
900 0.003 +0.0015 900.0012 899.9997 -0.0003
0 0.002
7.4.3 Effect of off-centre loading - (example of section 7.3.4)
Weight on pan: 500 g, Dial settings: 500 g
The readings are in grams

Back =0.001
Left = 0.002 Centre = 0.000 Right = 0.000
Front = 0.000
Maximum difference = 0.002 g
7.4.4 Hysteresis - (example of section 7.3.5)
Dial settings: 0 g
Two weights of mass M and M' are used, where both masses are nominally 50 mg. The
headings 1 and 2 refer to the first and second set of readings, and all data are in grams. The
readings shown in brackets need not be recorded.
Pan load 1 2 1
{pi, ps} 0.051 0.051
M + M' (0.100) (0.100)
{qi, qs} 0.051 0.051
Zero (0.000) (0.000)
M + M' (0.101) (0.100)
{q2, ql) 0.052 0.051
Zero (0.000) (0.001)
1 (P2, p A 0.051 0.050
The hysteresis is therefore 0.0005 g
7.4.5 Effect of tare - (example of section 7.3.6)
• Full value of tare =100 mg

• The optical scale is zeroed at no load and it reads 0.1 g when the lOOmg weight Mi (which
was used in 7.3.1) is put on the pan.
• A weight M2 of 100 mg is placed on the pan and the tare adjustment is used to re-zero the
balance. The weight Mi is put on the balance and readings r are taken
zi :0.000
ry: 0.100
r2. 0.098
Z2. -0.001
( ry + ^2- ^7 - ^2) /2 = 0.0995 g
7.4.6 Uncertainties associated with the calibration and use of a single-pan,

two-knife-edge balance
A full treatment of the uncertainties associated with balance calibrations is given in Chapter 8.
However a summary of these uncertainty calculations, for the sample calibration detailed
above, is given here in order to provide the uncertainties that will be quoted in the sample
report of section 7.4.7.
A calibration report for an electronic balance should quote the following:

(a) The uncertainties of the corrections.
(b) The Limit of Performance.
(c) The uncertainty of weighing (if requested by the user).
The uncertainties of the corrections are the uncertainties with which the corrections to
balance readings have been determined. The uncertainty of weighing is the uncertainty of a
mass value that is obtained after applying corrections to balance readings as in equation (7.8).
In general, it will vary with the load on the balance.
The Limit of Performance, F, of a balance is a quantity that has been in use for many
years and is not an uncertainty in the normally accepted sense. Instead, it gives a very
conservative upper limit to the measurement error, which may be expected when no
corrections are made to balance readings. Thus if r and z are the balance readings, the mass of
the object lies in the range (r - z) ± F. The Limit of Performance is an upper and lower bound;
in many applications the balance may weigh much more accurately than is indicated by this
figure.
The uncertainties of the corrections and the uncertainties of weighing are expanded
uncertainties, which must be calculated by combining their component uncertainties as
described below, but the Limit of Performance F is simply calculated from the equation:
= 2.26m^ + C^î (max) + tZ (C^.^, (max)) + C^^ (max) + C(C^^ (max)) (7.6)
where
Umax is the maximum value of the repeatability of measurement of the balance or
one-third of the resolution of the balance, whichever is greater.
Qw/(max) is the magnitude of the maximum correction to the dial reading, whether
obtained in a complete calibration of dial setting or in a tolerance test (see
equation 7.3)
^QiaXmax)) is the corresponding expanded uncertainty of Q/Xmax)
Cop(max) is the magnitude of the maximum correction to the optical scale
C/(Cop(max)) is the corresponding expanded uncertainty of Cpp(max)
7.4.6.1 Calculation of the Limit of Performance
In section 7.4.1, the repeatability of measurement Sr was found to be 0.00097 g and 0.0012 g at
the low end and high end of the optical scale respectively. The maximum value is 0.0012 g at
the high end of the optical scale. As this is greater than one-third of the resolution, from
equation 7.6, Umax is taken to be 0.0012 g.
The table in section 7.4.2.1 and Table 8.15 show the maximum correction of the
optical scale to be 0.00052g and its expanded uncertainty 0.002 g. The maximum correction of
built-in weights is 0.0008 g using the Simple tolerance test (7.4.22 a) and its expanded
uncertainty is 0.0022 mg as shown in Table 8.21. Hence from equation (7.6) , the Limit of
Performance of the balance is
F = 2.26 X 0.0012 + 0.00052 + 0.002 + 0.0008 + 0.0022 = 0.0082 g
7.4.6.2 Calculation of the uncertainty of corrections and uncertainty of weighing
The uncertainties of the optical scale corrections should be estimated for all kinds of balance
calibrations. If the corrections to each dial setting are measured, then the corresponding
uncertainty should be calculated, but for simplicity, it is recommended that only the
uncertainty for the 9 setting of each decade be calculated. For a tolerance test it is only
necessary to calculate the uncertainty of the maximum correction measured. Details of the
estimation are found in Chapter 8. These calculations are carried out in spreadsheets shown in
Tables 7.1 and 7.3 for loads at 80 mg and 900 g. Table 7.5 gives the uncertainty of weighing at
load of 1000 g.
For each component uncertainty shown in the tables below, Ui is an estimate of its effect
in the form of an expanded uncertainty (for the calibrating mass value, for instance), a standard
uncertainty (eg, the calibrating mass instability) or the semi-range of a rectangular distribution
(eg, the balance resolution). To obtain the corresponding standard uncertainty one must
divide U by the reducing factor ki, which is the coverage factor for an expanded uncertainty, or
a /3 (1.73) for a rectangular distribution semi-range. In general, one also needs to multiply Ui by
the sensitivity coefficient q , but in these tables all the c, are equal to one.
For each component uncertainty, an estimate of the number of degrees of freedom v, is

also required. The combined standard uncertainties are then obtained by combining the
components in quadrature (equation L.3), while the corresponding degrees of freedom are
obtained using the Welch-Satterthwaite formula (equation L.6). From Vgff one can then obtain
(from equation L.8) the coverage factor k used to obtain the expanded uncertainty.
Correction uncertainty for optical scale
The uncertainty components of the corrections can be found from section 8.2.1.3 and the
spreadsheet calculations for the example are shown below. The summary tables give the
uncertainty of correction and the uncertainty of weighing calculated in this way across the full
range of the balance.
Table 7.1: Spreadsheet for calculating the uncertainty of correction for the optical scale at a load of
80 mg.
Load: 80 mg
Component Values
Component (/ = 1, 2,... 8) Ui (mg) ki Ui (mg) Vi u7 Ui*/Vi
1. Mass value for 80 mg 0.031 2.04 1 0.015 30 2.3E-04 1.8E-09

2. Mass instability for 10 mg 0.002 1 1 0.002 4 4.0E-06 4.0E-12
6. Resolution of r 0.5 1.73 1 0.289 1000 0.084 0.000
1. Resolution of z 0.5 1.73 1 0.289 1000 0.084 0.000

Degrees of freedom Vetr 14
All the uncertainties of correction obtained in this way across the full optical range are shown
in Table 7.2. In view of these uncertainties, the corrections will be quoted in the report to the
nearest 0.1 mg.
Table 7.2: Expanded uncertainties for the corrections of the optical scale
Uncertainty of correction
Load of
(mg) optical scale (mg)
20 1.7
40 1.7
60 2.0
80 2.0
100 2.0
Dial setting uncertainty
The calculation of uncertainties for the corrections obtained from both the simple tolerance test
(7.4.2.2 a) or calibration of each dial settings (7.4.2.2 b) will be similar. Detailed calculation is
shown in section 8.2. An example at load 900g is shown below.
Table 7.3: Spreadsheet for calculating the uncertainty of correction for the dial setting at a load of
900 g.
Load: 900 g
Component Values
Component (/ = 1,2,... 8) Ui (mg) ki k/l Ui (mg) Vi
Ui^
1. Mass value for 900 g 1.35 2.04 1 0.662 30 0.438 0.006

2. Mass instability for 200 g 0.08 1 1 0.080 4 ; 0.006 1.0E-05
3. Mass instability for 200* g 0.08 1 1 0.080 4 0.006 1.0E-05
4. Mass instability for 500 g 0.2 1 1 0.200 4 0.040 4.0E-04
6. Resolution of r 0.5 1.73 1 0.289 1000 0.084 0.000
7. Resolution of z 0.5 1.73 1 0.289 1000 0.084 0.000

Degrees of freedom Ve/r 41
(a) For the simple tolerance test, the maximum correction is 0.8 mg at a load of 900g,
therefore the uncertainty of the maximum correction is 2.2 mg as shown in the above
table.
(b) For calibration of all the dial settings, each uncertainty should be calculated. However,
for simplicity, it is recommended to calculate the uncertainty only for the 9 setting of
each decade as it normally has the largest uncertainty within the decade. The uncertainty
at the 9 setting can then be used for the uncertainties of that decade.
Table 7.4; Expanded uncertainties for the corrections of the optical scale
Uncertainty of correction of
Dial setting (g) dial settings (mg)
0.1-0.9 1.7
1-9 1.7
10-90 1.7
100-900 2.2 1
Uncertainty of weighing (required only if requested by customer)
The uncertainty components for the estimation of uncertainty of weighing can be found in
section 8.2.2. The spreadsheet calculation for the example is shown below:
Table 7.5: Spreadsheet for calculating the uncertainty of weighing at a load of 1000 g.
Load: 1000g
Component Values
Component (/ = 1, 2,... 9) Ui (mg) ki led u, (mg) Vi
U/2
uNvi
1. Optical scale Correction 2.03 2.16 1 0.940 14 0.883 0.056
2. Dial setting correction(0.9g) 1.69 2.12 1 0.797 17 0.635 0.024
3. Dial setting correction (9 g) 1,70 2.12 1 0.802 17 0.643 0.024
6. Repeatability in r-z 1.2 1 1 1.200 9 1.440 0.230
7. Resolution of r 0.5 1.73 1 0.289 1000 0.084 0.000
8. Resolution of z 0.5 1.73 1 0.289 1000 0.084 0.000
Combined standard uncertainty Uc(mg) 2.36

Degrees of freedom Ven 79
Table 7.6: Uncertainty of weighing summary table
Uncertainty of weighing
Load (mg)
(g)
0.1 -0.9 3.6
1-9.9 3.9
10-99.9 4.2
100-1000 42
Sample report for a single-pan two-knife-edge balance

Bradfield Road, West Lindfield NSW 2070, Australia
Phone: (612) 9413 7000 Fax: (612) 9413 7631
REPORT ON
Single-Pan Two-Knife-Edge Balance
Report No. RN 12345
Maker Mettler Model B5 Cl000 Serial No. 654321
Capacity 1000 g Optical Scale Range 100 mg Optical Scale Division 10 mg
Reading to 1 mg by means of optical readout
Type (top-loading or suspended pan) Suspended pan
Client NML, Bradfield Road, West Lindfield NSW 2070
Examined at Room C20E, National Measurement Laboratory
Examined by K Fen
Calibration Date 28 Nov 2001 Date of Issue of report 28 February 2002
Temperature during test 20.5‘’C
Before calibration, the balance was adjusted according to the manufacturer’s instructions.
Repeatability of Measurement
Standard Maximum difference

Optical Scale Deviation of between successive
Load Dial Readings Reading reading readings
(g) (g) (mg) (g) (g)
500 500 0 0.00097 0.003
499.9 100 0.0012 0.003
Corrections to balance reading
(i) Corrections to Optical Scale reading
Optical scale
reading
(g) 0.020 0.040 0.060 0.080 0.100
Correction
(g) +0.0001 0.0000 +0.0004 +0.0005 +0.0005
Uncertainty(±)
(g) 0.0017 0.0017 0.0020 0.0020 0.0020
(ii) Corrections to dial reading [Only one of paras (a) or (b) to be presented]
(a) The corrections of the individual dial readings have not been determined but a simple test
of dialling 9 for each decade yielded a maximum correction of 0.0008 g for all dials, with
an uncertainty of 0.002 g.
(b) The corrections of the individual dial readings have not been determined but a
comprehensive test [8], which ensured that all the built-in masses were tested, yielded a
maximum correction of 0.00 Ig for all dials, with an uncertainty of 0.002 g.
1 Dial Reading Correction Uncertainty(±)

_________ (g) (g)
0.1 -0.0001
0.2 -0.0014 0.0017
0.9
1
2 0.0017
9
10
20 0.0017
90
100
0.0022
800 -0.0 001
900 -0.0 003
The corrections to the dial readings have been determined on the basis of weighing made in air
of density 1.2 kg/m^ against weights of density 8000 kg/m^.
Off-centre loading
A weight of mass 500 g was moved to various positions on the pan. Except for the Centre
position, the weight was placed as close as possible to the edge of the pan without passing
beyond the edge. The balance readings (in grams) at each position, relative to the centre
reading, are given in the table below
maximum
Centre Front Back Left Right difference
0.000 0.000 +0.001 +0.002 0.000 0.002 —
Hysteresis
The hysteresis in the optical scale was found to be 0.0005 g.

Effect of tare
Tare Load Optical Scale Reading with

(g) 0.1 OOg (g)
0 0.100 g
0.1 OOg 0.0995g
Limit of Performance for the balance: 0.0082 g
Uncertainty of weighing of the balance:
Load Uncertainty (±)

(g) (mg)
0.1-0.9 3.6
1-9.9 3.9
10-99.9 4.2
100- 1000 4.7
Notes
1. The balance has been tested according to procedures detailed in The calibration of weights
and balances by E C Morris and K Fen, National Measurement Laboratory, Lindfield,
NSW, Australia
2. When the sign of the correction is positive (+) the amount should be added to the balance
reading to give the correct mass value and when negative (-) subtracted from it.
3. The corrections were calculated on the basis that the set of weights used for the calibration
have density of 8000 kg.m^ and weighed on the balance in air of density 1.2 kg.m’^.
Buoyancy corrections may be necessary at altitudes above 300m.
4. The Limit of Performance is the upper limit of the measurement error that may be expected
when no corrections are made to the balance readings.
measurand with 95% probability. The coverage factor of the correction uncertainty is 2.15
for the optical scale, 2.1 for the dial setting below 500g and 2.0 for the dial setting above
500 g. The coverage factor for the uncertainty of weighing is 2.0 for the entire range.
Dr N Bignell
for Dr. B D Inglis
7.5 USING THE CALIBRATED BALANCE 91
7.5 Using the calibrated balance

7.5.1 The normal procedure for measuring mass with a calibrated single-pan,
There are various ways of using a calibrated balance, but in essence they all reduce to the
following;
• Allow the balance to warm up until the readings are steady.

• The object or material to be weighed should be placed close to the balance until
approximate temperature equality with the balance is attained (see Table B2 for the times
required for weights). However thermal equality is not very important for low-precision
(>30 ppm) weighing.
• With the balance pan empty, turn the dial setting to zero, release the balance and note the
zero reading, z;; arrest the balance. (The balance may be tared before this reading.)
• With the object to be weighed on the pan, turn the dial reading to a load for which the
optical scale reading is within its range; release the balance and note the reading, r.
• Remove the object from the pan, turn the dial setting to zero, release the balance and note
the zero reading again, Z2.
• Calculate the average zero reading: z = {z, + zl}!'!.
• If the balance Limit of Performance F is less than the maximum allowable uncertainty for
the measurement, then the measured mass may be simply taken as
M=r-z+R (7.7)
where R is the dial reading
• Alternatively, a corrected measured mass is obtained from equation (7.8) where Cop is the
correction to the optical scale and Cm is the correction to the dial reading at load M. In this
case, the uncertainty of the measurement may be quoted as ± U, where U is the
imcertainty of weighing of the balance.
Mo= k -z + Cop + Cm + R (7.8)
(The mass value obtained from a calibrated balance is the conventional mass M. The value of
the true mass is needed only in relatively rare circumstances and it can be obtained from the
measured conventional mass using equation (2.6). The difference between true and
conventional mass is explained in Chapter 2.)
7.5.2 User balance adjustment and test
7.5.2.1 Adjustment
A balance should be adjusted at least monthly when in regular use (see section 7.3.1). This is
to ensure that the corrections to balance reading will be as close as possible to those obtained
when the balance was calibrated.
7.5.2.2 Repeatability test
The repeatability of measurement Sr should be determined every six months (see section 7.3.2)
to ensure that the balance has not developed a defect that will affect mass measurements. The
values obtained should be recorded so that the latest result ^^latest) can be compared with the
previous result j^previous). This test should always be carried out by the same operator.
Assuming that the repeatability is always measured with nine degrees of freedom, then
5r(latest) should not exceed 1.785Xprevious). Furthermore, ^/latest) should not be greater
than twice the repeatability given in the most recent calibration report for the balance. If these
requirements are not met, there is a fairly strong indication that the performance of the balance
has deteriorated. The user should consult the calibrating laboratory to see if calibration,
servicing or some other action is required.
Note that any measurement of Sr yielding a value of zero should not be used as the value
of 5r(previous); in such cases use an earlier value, if available. If no earlier non-zero value of
Sr is available, the manufacturer’s specification could be used.
A more sophisticated version of this test is possible, in which the previously measured
values of the repeatability are pooled. See Appendix G for details, and note that it may not be
advisable to include measurements made before the most recent balance calibration in the
pool. This version of the test gives a more sensitive indication of any problem arising with the
balance.
7.5.3 Air buoyancy and weighing
Since the corrections to balance reading given in a report were obtained using standard weights
with a density of approximately 8000 kg.m'^, they will be in error when weighing materials of
different density, because of the different buoyancy. (See Chapter 2 for detailed discussion of
air buoyancy).
For mass measurement with a calibrated single-pan beam balance for which the
corrections of individual dial readings have been determined, the buoyancy correction is given
by equation (7.9), where Pr is the density of the inbuilt weights of the balance. In most cases,
pR can be taken to be 8000 kg.m'^ if it is not known.
7.5 USING THE CALIBRATED BALANCE 93
SM = (p^ -1.2)1-—V
(JS)
\P Pr j
If it is necessary to weigh light (low density) materials with a mass uncertainty in the
order of 0.05% or less, estimate the likely size buoyancy correction as follows: For the
altitude of the laboratory, find the mean value of pa. Table J.l in Appendix J may be used for
this purpose. Then substitute this value into equation (7.9), while putting p equal to the
density of the material being weighed and pR equal to 8000 kg.m’^ For instance. Table J.l
indicates that the mean laboratory air density at 700 m is 1.099 kg.m'\ Substituting this value
into the equation for pa and putting p equal to 1000 kg.m'^ one obtains 0.009% for the
estimate of the buoyancy correction when weighing water.
If the correction calculated in the previous paragraph represents a serious error if

ignored, then corrections determined from equation (7.9) must to be applied.
7.6 Measuring corrections to dial readings using the

comprehensive tolerance test
The comprehensive tolerance test is basically the method proposed in ASTM E319 [8]. Two
combinations of the built-in weights are compared with one calibrated weight. One dial setting
has the same nominal value as the test weight; the other will be smaller by one step on the dial
for the smallest decade. In this manner suitable test loads can be selected so that all the built-in
weights are checked, even if the combination is not known. This method will not give the
values (or corrections) for all the dial readings, unless they are all measured. The method
virtually is a slightly abbreviated version of that given in section 7.3.3.2.2. Unless the
combination is known it is possible to have errors cancelling in built-in weights, but this is
extremely unlikely.
Example: Use the same balance as described in 7.4.
Full scale correction Cop = +0.000 52 g and Cop = 0 at low end of the optical scale
It is noted that the balance has an optical full-scale value of 100 mg. Because the full optical
scale is used in some readings, a correction must be applied to the optical scale reading when it
is used.
As seen from Table 7.7, sufficient dial settings are used to ensure that all the built-in
weights are used. The maximum correction obtained from all the measurements is then
reported. This scheme has the advantage that only two weighings per decade have to be used
and overcomes some of the disadvantages of the simple tolerance test described in section
7.3.3.2.I. The uncertainties of the corrections can be calculated using a method similar to
7.4.6.2.
94 7. CALIBRATION OF SINGLE-PAN. TWO KNIFE EDGE BALANCE
Table 7.7; Determination of corrections of built-in weights using the comprehensive tolerance test
Pan load Dial Setting Optical Scale Value of Dial Correction
M R Reading M ~r-Cop + z M-r - Cop+ z-R
(g) (g) (g) (g) (g)
0 0 Z1 0.000 0
0.1 0.0 ri 0.099 0.0005 +0.0005
0.1 0.1 r2 -0.001 0.0990 -0.0010
0 0 Z2 -0.002
0.29997 0.2 0.099 0.1985 -0.0016
0.29997 0.3 -0.002 0.3000 0.0000
0 0 -0.002
0.49997 0.4 0.098 0.3995 -0.0005
0.49997 0.5 -0.002 0.4990 -0.0010
0 0 -0.002
3.00001 2.9 0.100 2.8975 -0.0025
3.00001 3.0 -0.002 3.0000 0.0000
0 0 -0.002
5.00004 4.9 0.098 4.8995 -0.0005
5.00004 5.0 -0.002 4.9990 -0.0010
0 0 -0.003
etc.
7.7 Measuring corrections to dial readings using a

least-squares calibration
Each decade in the balance generally has four weights to make up the nine, or more,
combinations for the dial settings. To make use of this information in the least-squares
analysis, the appropriate combination for the balance under test must be obtained either by
observation or from the manufacturer. There are over 50 possible combinations used in
different balances by different manufacturers. The least-squares method requires the
observations obtained in section 7.3.3.2.2. However the calculations are fairly involved and for
this reason Appendix E is devoted to explaining them (see [9, 10]).
The least-squares method does have a number of advantages over all other methods:
(i) it is the most accurate method;
(ii) a self-consistent set of values is obtained;
(iii) the standard deviation of the correction for each dial setting can be obtained if desired;
(iv) arithmetical errors and misreadings can be detected from an examination of the residuals.
If the masses of the built-in weights in the balance are required to be known to this
accuracy then consideration should be given to measuring by substitution with a calibrated set
of weights. This should not involve any extra time, because to make use of the accuracy of the
in-built weights a number of repeated readings must be made.
8.2 SINGLE-PAN TWO-KNIFE-EDGE BALANCES 95
Chapter 8
Balance calibration uncertainties

Summary
This chapter describes how to estimate the uncertainty of a balance calibration. The reader is
referred to Appendix L in this book or other sources [4, 5] for a more complete description of
uncertainty estimation. The uncertainty is estimated for a 95% confidence interval.
The uncertainty of the corrections and the uncertainty of weighing are discussed and
sample calculations for a balance are shown. A figure called the Limit of Performance that is
used to describe how accurately a balance may weigh is also discussed.
In a balance calibration report, the uncertainties of the corrections to the balance

readings should be given and the Limit of Performance should also be stated. The uncertainty
of weighing is used when corrections are applied and will be stated in the report only when the
user or client requests it.
This chapter is divided into two sections. Section 8.1 discusses the correction
uncertainty, uncertainty of weighing and the Limit of Performance of Electronic Balances.
Sample uncertainty calculations for the example described in Chapter 6 are also given in this
section. Section 8.2 refers to Single-pan two-knife-edge balances. Sample uncertainty
calculations for correction uncertainty, uncertainty of weighing and Limit of Performance for
the example described in Chapter 7 are given in this section.
8.1 Electronic balances

8.1.1 Uncertainty of correction
8.1.1.1 The measurement (using a set of calibrated weights)
The corrections to the balance reading at various loads are obtained by using combinations of
weights from a calibrated set. The measurement is expressed by the following model:
C = M - {r - z) (8.1)
where C is the balance correction, the measurand, at a particular load, M is the total mass value
of the calibrating weight(s) placed on the pan, (r — z) is the average of n measured values of
the difference (r-z), r is the reading indicated on the balance when the set of calibrating
weights is on the pan and z is the zero reading of the balance with no load on the pan.
The uncertainty in the correction C is determined by considering all uncertainty

contributions to all the values used in its calculation, i.e., in Af, r and z of equation (8.1).
The first step is the calculation of sensitivity coefficients, for M and (r - z), needed
to convert their uncertainties into components of uncertainty for C. In each case the magnitude
of Ci is 1 (see Appendix L). The next step is to list all components and to assign values of
uncertainty, reducing factors and numbers of degrees of freedom. The uncertainty components
are illustrated in the following example.
8.1.1.2 An example in calculating the correction uncertainty
This example applies to the calibration of an electronic balance with a range of 1200 g and a
resolution of 0.0 Ig (described in section 6.4). Corrections were measured at 12 points across
the full range of the balance. Double weighing was used in the measurement. Sample
calculations are given below for loads of lOOg and 800g.
(1) Calibration uncertainties for the calibrating weight(s)
The expanded uncertainty for each of the calibrating masses in the chosen set, as
obtained from the calibration report, is shown in Table 8.1. The report also gives 2.04 (A,),
equivalent to = 30, for the coverage factor used in each. (The quantities A, and Vi are
explained in Appendix L)
Table 8.1: Calibration uncertainties for the set of calibrating weights
Nominal mass value (g) Uncertainty (mg)

100 0.50
200 0.90
500 2.2
1000 3.0
When used to calibrate the balance, the weights are used on their own or in combination
to make up the desired load. Since the mass uncertainties for weights from the same set are
highly correlated, the uncertainties of weights used in combination should be added (assuming
for convenience, full correlation). For example, a load of 800 g requires the three weights, of
mass 100 g, 200 g and 500 g, and so the expanded uncertainty is 0.5 + 0.9 + 2.2 = 3.6 mg. The
uncertainties for other combinations are given in Table 8.2. The number of degrees of freedom
for a combination is equal to that for the weight with the largest contributing uncertainty.
Table 8.2; Uncertainty (C/,) of mass values of the combined calibrating weights chosen to determine
the correction C. The calibration uncertainties of the individual weights are from Table 8.1. The
prime ' is used to indicate a second 200 g weight.
1 Load(g) Weights used t/,- (mg) Load (g) Weights used U(mg) 1
100 100 g 0.5 700 500 g + 200 g 3.1
200 200 g 0.9 800 500g + 200g+ lOOg 3.6
300 100 g + 200 g 1.4 900 500 g + 200 g + 200'g 4.0
400 200 g +200'g 1.8 1000 1000 g 3.0
500 500 g 2.2 1100 1000 g+ 100 g 3.5
600 500 g + 100 g 2.7 1200 1000 g +200 g 3.9
(2) Instability of the calibrating weight(s)
it is assumed that the uncertainties quoted in the reports of reference weights do not include
this component, which must therefore be included in these calculations. If it is clear from the
report on the reference weights that quoted uncertainties do include the effect of mass
instability, then the component discussed in this subsection should be omitted from the
calculations.
As the calibrating weights did not have an extensive history, the standard uncertainty
due to the instability of the calibrating weights was taken to be 8% of the maximum
permissible error (MPE) for the relevant OIML class (see section 5.2.2) with the number of
degrees of freedom equal to 4 (v; = 4). In this case, it is an F2 set so the instability uncertainties
are as shown in Table 8.3. These values of uncertainty are standard uncertainties, so for each,
k{ = 1.
Table 8.3: Instability of the calibrating weights, as standard uncertainties

Nominal mass value (g) Uncertainty (mg)
100 0.12
200 0.24
500 0.60
1000 L20
Unlike the calibration uncertainties, the instability uncertainties are taken to be

uncorrelated. Therefore, when a combination of weights is used, their instability uncertainties
are combined in quadrature (equation L.3). This is done in the spreadsheets shown in Tables
8.4 and 8.5.
(3) Repeatability
The repeatability of measurement is the standard deviation of the difference (r - z), as

discussed in section 6.3.2. For the balance used in this example, it was found (in section 6.4.1)
to be 0.0 mg at 500 g and 4.2 mg at 1000 g with 9 degrees of freedom (v; = 9).
Since the repeatability of measurement refers to a single measured difference (r - z), the
standard deviation of the average of two values (double weighing), as used to determine C, is
smaller. So, for the following calculation, the effect of balance repeatability on C will be
taken as 0.0 mg for all loads up to and including 500 g and 2.97 mg for loads greater than 500
g. Being a standard deviation, = 1.
(4) Resolution of balance
For a digital balance, the "true" reading will lie within a band equal to the least count centred
on the observed reading. In this example, the least count is 0.01 g , so an observed reading of
235.73 g indicates a “true” reading lying between 235.725 and 235.735 g with a rectangular
probability distribution. We take the resolution uncertainty as ± half the least count (± 5 mg)
and k{ = 1.73, since the reducing factor for the semi-range of a rectangular distribution is V3
(4) , i.e., 1.73. The resolution uncertainty for a digital balance has infinite degrees of freedom,
although, in practice any value greater than about 100 will do in the calculations that follow
(say, Vi = 1000).
The resolution of the balance will affect both r and z, so for each put CT, = 5 mg and =
1.73. Also, assuming as the worst case that resolution errors are systematic, their effects will
not be reduced by double weighing (by ^/2).
For dual or multi-range balances, if the zero reading has one more figure (higher
resolution), than the balance reading with load e.g. the least count is 0.001 g at zero reading,
the resolution uncertainty for z becomes CT, = 0.5 mg and = 1.73.
(5) Rounding of the result
The final measurement result (values of C and CT) will be rounded to the nearest 1 mg, i.e.,
with a rounding uncertainty of ±0.5 mg. Thus, we put CT, = 0.5 mg, k, = 1.73, being
rectangular, and Vi= 1000.
(6) Other sources of uncertainty
The effect of off-centre loading will usually be negligible because during a calibration all loads
would be placed centrally on the pan and its effect for such ‘good-practice’ would already be
included in the repeatability of measurement. Also, the effect of hysteresis (section 6.3.5)
would also be negligible when the balance is properly adjusted.
The calculation
The uncertainty components and the calculations are tabulated in Table 8.4 for 100 g
and in Table 8.5 for 800 g.
The formulae used in these spreadsheets are:

• Equation (L.2) to calculate the values of w,

• Equation (L.3) to calculate the combined standard uncertainty Uc
• Equation (L.6) to calculate the degrees of freedom Vgff
• Equation (L.8) to calculate coverage factor k
• Equation (L.7) to calculate the expanded uncertainty U
of 100g
Load: 100g
Component Values
Component (/ = 1,2,... 6) Ui (mg) ki Ui (mg) Vi
Ui^ /Vi
1. Mass value for 100 g 0.5 2.04 1 0.245 30 0.060 1.2E-04

2. Mass instability in 100 g 0.12 1 1 0.120 4 0.014 5.2E-05
3. Repeatability in r- z 0 1 1 0.000 9 0.000 0.000
4. Resolution of r 5 1.73 1 2.890 1000 8.353 0.070
5. Resolution of z 5 1.73 1 2.890 1000 8.353 0.070

of800 g
Load: 800 g
Component Values
Component (/ = 1, 2,... 8) Ui (mg) ki |g/l ^/(mg) Vi
Ui^ U^/Vi
1. Mass value for 800 g 3.6 2.04 1 1.765 30 3.114 0.323

4. Instability in 500 g 0.60 1 1 0.600 4 0.360 0.032
6. Resolution of r 5 1.73 1 2.890 1000 8.353 0.070
1. Resolution of z 5 1.73 1 2.890 1000 8.353 0.070

[Expanded uncertainty U (mg) 10.73
8.1.1.3 Uncertainty of correction when using only one calibrated weight
The calibration method described in Section 6.6 uses only one calibrated weight and
consequently the model of equation (8.1) does not apply. The correct model is obtained by
combining equations (6.10) and (6.11) to give the general equation for the correction at stepy:
Cj =j M-(ri-zi)- ... - (rj - Zj) (8.2)
where Cj (the measurand) is the balance correction at step y, rj is the reading indicated on the
balance at a load jM, M is the mass value of the only calibrating weight placed on the pan and
z is the zero reading of the balance reading with no load on the pan.
The sensitivity coefficient c, for converting uncertainties in M to components of

uncertainty for Cj isy, and that for r or z is 1 (ignoring its sign).
The sources of uncertainty are the same as those in the calibration of balances using a
calibrated set of weights (section 8.1.1.1). However, they affect the result differently. The
sensitivity coefficient for components affecting M is not unity (see above) and those
components affecting r and z appeary times (equation (8.2), so the net effect of each is a factor
greater. An example calculation is given in Table 8.6.
of 800 g using a single calibrated weight of 100 g and 8 build-up steps.
Load at 800 g with 1 calibrated weight
Component Values
Component (/ = 1,2, ... 6) U (mg) ki N Ui (mg) Vi Ui^ Ui^Ni
1. Mass value in 100 g 0.50 2.04 8 1.961 30 3.845 0.493

2. Mass instability in 100 g 0.12 1 8 0.960 4 0.922 0.212
3. Repeatability in r-z forj=8 2.97^8 1 1 8.400 9 70.567 553.3
4. Resolution of r forj=8 5^8 1.73 1 8.175 1000 66.825 4.466
5. Resolution of z forj=8 5l6 1.73 1 8.175 1000 66.825 4.466
6. Rounding of the final result 0.5 1 73 1 0.289 si 000 0.084 7.0E-06

Degrees of freedom Veff 78
Compare the expanded uncertainty at 800 g obtained here, 28.8 mg (round to 29 mg),
with that obtained using a single combination of calibrated weights (Table 8.5), rounded to
11 mg. It is clearly preferable to use a full set of calibrated weights when possible.
8.1.2 Uncertainty of weighing
8.1.2.1 The measurement
In this case, a calibrated balance is used to measure the mass value of an object. The model
that expresses this measurement is:
M=C + (r-z} (8.3)
where M is the mass value of the object placed on the pan (the measurand), C is the balance
correction at the particular load, r is the reading indicated on the balance when the test weight
is on the pan and z is the reading of the balance with no load on the pan.
Here, the sensitivity coefficient c, for converting uncertainties in C, r or z to components of

uncertainty for M is 1.
8.1.2.2 An example in calculating the uncertainty of weighing
This example refers to the same balance as that described in section 8.1.1.2. The uncertainties
of weighing at loads of 100 g and 800 g are calculated. The sources of uncertainty are:
(1) Correction to balance reading
The uncertainty in C, and its associated parameters (k and are given in Tables 8.4 and 8.5.
As components in this calculation, they may be expressed as follows.
For a load of lOOg: Ui = 8.05 mg with ki = 1.96, and v; = 2035.
For a load of 800g: Ui = 10.8 mg with ki = 1.99, and v; = 93.
(2) Repeatability
The repeatability of measurement is the standard deviation of the difference (r - z) as

discussed in sections 6.3.2. For the balance used in this example, it was found (in
section 6.4.1) to be 0.0 mg at 500 g and 4.2 mg at 1000 g with 9 degrees of freedom (i< = 9).
So, for the following calculation, the effect of balance repeatability on M will be taken as
0.0 mg for all loads up to and including 500 g and 4.2 mg for loads greater than 500 g. Being a
standard deviation, k^ = 1.
The least count of the balance is 10 mg, (see section 8.1.1.2) so the resolution uncertainty is ±5
mg (U, = 5 mg) and it affects both r and z. Since resolution errors have a rectangular
distribution, ki = 1.73, and as before put k = 1000.
The final measurement result (values of M and U) will be rounded to the nearest 1 mg, i.e.,
with a rounding uncertainty of ±0.5 mg. Thus, we put Ui = 0.5 mg, k, = 1.73, being
rectangular, and = 1000.
The effect of off-centre loading and hysteresis are negligible as explained in 8.1.1.2 item (6).
Table 8.7: Spreadsheet for calculating the uncertainty of weighing for an electronic balance at a load of
100 g
Load: 100g
Component (/ = 1,2,... 5) U, (mg) ki k/| Ui (mg) Vi uNvi

1. Uncertainty in C 8.05 1.96 1 4.107 2035 16.869 0.140
2. Repeatability in r - z 0 1 1 0.000 9 0.000 0.000
3. Resolution of r 5 1.73 1 2.890 1000 8.353 0.070
4. Resolution of z 5 1.73 1 2.890 1000 8.353 0.070

Table 8.8: Spreadsheet for calculating the uncertainty of weighing for an electronic balance at a load
of 800 g
Load: 800 g
Component Values
Component (/ = 1,2,... 5) U,(mg) ki Ui (mg) Vi Ui^ Ui*/Vi
1. Uncertainty in C 10.73 1.99 1 5.392 93 29.073 9.Q8Q
2. Repeatability in r-z 4.20 1 1 4.200 9 17.640 34.574
3. Resolution of r 5 1.73 1 2.890 1000 8.353 0.070
4. Resolution of z 5 1.73 1 2.890 1000 8.353 0.070

The uncertainty components for the calculation of uncertainty of weighing for 100 g and
800 g loads are shown in Table 8.7 and Table 8.8 respectively. The formulae used in these
spreadsheets are the same as those used in Tables 8.4 to 8.6. The expanded uncertainties of the
corrections and weighings across the full range of the electronic balance used in these
examples are summarised in Table 8.9.
Table 8.9: Expanded uncertainties for the corrections and weighings for the electronic balance
Uncertainty of Uncertainty of Uncertainty of Uncertainty of

Load correction weighing Load correction weighing
(g) (mg) (mg) (?) (mg) (mg)
1 100 8 12 700 11 16
200 8 12 800 11 16
300 8 12 900 11 16
400 9 12 1000 11 16
500 9 12 1100 11 16
600 11 16 1200 11 16 1
8.1.3 Limit of Performance for an electronic balance
The Limit of Performance F of a balance is a quantity that has been in use for many years and
is not an uncertainty in the normally accepted sense. Instead, it gives a very conservative
upper limit to the measurement error that may be expected when no corrections are made to
balance readings. Thus if r and z are the balance readings, the mass of the object lies in the
range (r - z) ± F. It is emphasised that the Limit of Performance is an upper and lower bound.
In many applications the balance may weigh much more accurately than is indicated by this
figure.
The Limit of Performance F is defined by the following equation:
^■ = 2.26»_+C_+C/(C_) (8.4)
where:
Umca is the maximum value of the repeatability of measurement of the balance or one-
third of the discrimination of the balance, whichever is greater.
Cmax is the magnitude of the maximum correction to balance reading C
G(Cmax) is the expanded uncertainty associated with Câx
In the example of section 6.4 (a balance with a range of 1200 g and discrimination of
0.01 g), the maximum value for the repeatability of measurement was found to be 0.0042 g at a
load of 1000 g. As this is greater than one-third of the discrimination, uâx is 0.0042 g.
The table in section 6.4.2 shows the maximum correction to the balance reading to be
-0.038 g (at a load of 800 g), while Table 8.9 shows the uncertainty of this correction as
0.011 g. Hence the Limit of Performance of the balance is
F = 2.26x0.0042 + 0.038 + 0.011 = 0.059 g

8.2 Single-pan two-knife-edge balances

8.2.1 Uncertainty of correction
The corrections to the balance reading at various loads are obtained by using combinations of
weights from a calibrated set. The measurement is expressed by the following ‘model’:
C = M-{r-z}-R (8.5)
Where C is the balance correction, the measurand (see Appenidix L for details) at a particular
load, M is the mass value of the calibrating weight(s) placed on the pan, (r - z) is the average
of n values of the difference (r - z), r is the optical scale reading and R is the dial setting when
the calibrating weight is on the pan and z is the zero reading of the balance with no load on the
pan.
The uncertainty in the correction C is determined by considering all uncertainty

contributions to all the values used in its calculation except R, i.e., in M, r and z of equation
(8.5). As R is a constant, it does not contribute uncertainty to the result.
The first step is the calculation of sensitivity coefficients, c,, for M and (r - z), needed
to convert their uncertainties into components of uncertainty for C. In each case the magnitude
of Ci is 1 (see Appendix L). The next step is to list all components and to assign values of
uncertainty, reducing factors and numbers of degrees of freedom. The uncertainty components
are illustrated in the following example.
8.2.1.2 An example in calculating the correction uncertainty
Consider a single-pan two-knife-edge balance with a full load of lOOOg and an optical scale of
range 100 mg and 10 divisions (1 div =10 mg) that can be read to 0.1 division (1 mg) using a
vernier scale (described in section 7.4). The correction to the balance has two parts, namely the
correction to the optical scale and the dial-settings, as discussed in Chapter 7. Double weighing
was used in the measurement.
A. Optical scale
The corrections to the optical scale were determined at 5 points across the full optical scale
with all dials set to zero. The balance has 0.97 mg at half scale load (50 mg) and 1.2 mg at
full-scale load (1 OOmg). Sample calculations are given for loads of 20 mg and 100 mg.
(Al) Mass value of calibrating weight(s)
A set of weights was used for the calibration of the optical scale of the balance. The
expanded uncertainty for each of the calibrating masses in the chosen set, as obtained from the
calibration report, is shown in Table 8.10. The report also gives 2.04 equivalent to 14 = 30,
for the coverage factor used in each. (The quantities ki and v; are explained in Appendix L)
Table 8.10: Calibration uncertainties for the set of weights

Nominal mass value (mg) Uncertainty (mg)
10 0.008
20 0.010
50 0.013
100 0.015
When used to calibrate the balance, the weights are used on their own or in combination
to make up the desired load. Since the mass uncertainties for weights from the same set are
highly correlated, the uncertainty of any one combination of weights should be added
(assuming for convenience, full correlation). For example, a load of 80 mg requires the three
weights, of mass 10 mg, 20 mg and 50 mg, and so the expanded uncertainty is 0.008 + 0.010 +
0.013 = 0.031 mg. The uncertainties for other combinations are given in Table 8.11. The
number of degrees of freedom for a combination is equal to that for the weight with the largest
contributing uncertainty.
Table 8.11: Uncertainty (tT,) of mass values of the combined calibrating weights used to determine the
optical scale corrections. The calibration uncertainties of the individual weights are from Table 8.10.
Load (mg) Weight used Combined uncertainty (mg)
20 20 mg 0.010
40 20 mg + 20' mg 0.020
60 50 mg + 10 mg 0.021
80 50 mg + 20 mg +10 mg 0.031
100 100 mg 0.015 1
(A2) Instability of calibrating weight
it is assumed that the uncertainties quoted in the reports of reference weights do not include
this component, which must therefore be included in these calculations. If it is clear from the
report on the reference weights that quoted uncertainties do include the effect of mass
instability, then the component discussed in this subsection should be omitted from the
calculations.
As the calibrating weights did not have an extensive history, the standard uncertainty
due to the instability of the calibrating weight was taken to be 8% of the maximum permissible
error (MPE) for the relevant OIML class (see section 5.2.2) with 4 degrees of freedom. In this
case, the weights are an OIML Fi set giving an instability uncertainty as shown in Table 8.12.
These values of uncertainty are standard uncertainties, so for each, k[ = 1.

i Nominal mass value (mg) Instability Uncertainty (mg)
10 0.002 0
20 0.002 4
50 0.003 2
100 0.004 0 1
Unlike the calibration uncertainties, the instability uncertainties are taken to be

uncorrelated. Therefore, when a combination of weights is used, their instability uncertainties
are combined in quadrature (equation L.3). This is done in the spreadsheets shown in Table
8.13 & Table 8.14.
(A3) Repeatability of measurement
The repeatability of the measurement is the standard deviation of the difference (r - z)

(discussed in sections 7.3.2). It was found to be 0.97 mg for loads below 50 mg and 1.2 mg for
loads above 50 mg with 9 degrees of freedom (v; = 9).
Since the repeatability of measurement is the standard deviation of the difference (r - z),
the standard deviation of the average of two values (double weighing), as used to determine C,
is a /2 smaller. So, for the following calculation, the effect of balance repeatability on C will be
taken as 0.686 mg for all loads up to and including 50 mg and 0.849 mg for loads greater than
50 mg. Being a standard deviation, k{ = 1.
(A4) Resolution of balance
For a single-pan, two-knife-edge balance with a vernier scale in the optical scale range, its
resolution is the readability of the vernier scale. For this example, in which the scale is read to
the nearest 0.001 g, an observed reading of 2.356 g indicates a true reading lying between
2.355 5 and 2.356 5 g with a rectangular probability distribution. We take the resolution
uncertainty as ±0.5 mg and k\ = 1.73, since the reducing factor for the semi-range of a
rectangular distribution is ^3 [4], i.e., 1.73. The resolution uncertainty has infinite degrees of
freedom, although, in practice any value greater than about 100 will do in the calculations that
follow (say, = 1000).
The resolution of the balance will affect both r and z, so for each put Ui = 0.5 mg and k{
= 1.73. Also, assuming as the worst case that resolution errors are systematic, their effects will
not be reduced by double weighing (by ^2).
Some balances have analogue scales, with mechanical or optical pointers that indicate
the deflection of the beam on a set of scale marks. Then, the resolution may be expressed as
scale ‘readability’, the degree to which the scale can be read. The readability is commonly one-
tenth or one-fifth of the smallest scale division. For example, take a = 1/10 of the division, k, =
1.73 and v, =1000, as above.
(A5) Rounding of the result
The final measurement result (values of C and U) will be rounded to the nearest 0.1 mg, i.e.,
with a rounding uncertainty of ±0.05 mg. Thus, we put Ui = 0.05 mg, = 1.73, being
rectangular, and 14 = 1000.
(A6) Other sources of uncertainty
included in the repeatability of measurement (section 7.3.4). Also, the effect of hysteresis
(section 7.3.5) would also be negligible when the balance is properly adjusted.
The calculation
The uncertainty components for the correction to the optical scale and the calculations are
shown in Table 8.13 for 20 mg and Table 8.14 for 80 mg.
The formulae used in the spreadsheets below are

• Equation (L.2) to calculate the values of w,
• Equation (L.3) to calculate the combined standard uncertainty Uc
• Equation (L.6) to calculate the degrees of freedom Vgff
• Equation (L.8) to calculate coverage factor k
• Equation (L.7) to calculate the expanded uncertainty U
Table 8.13; Spreadsheet for uncertainty of corrections at 20 mg

Load: 20 mg
Component Values
Component (/ = 1,2,... 6) Ui (mg) ki Iq I Ui (mg) Vi uNvi
2. Mass instability 0.0024 1 1 0.002 4 5.8E-06 8.3E-12
4. Resolution of r 0.5 1.73 1 0.289 1000 0.084 7.0E-06
5. Resolution of z 0.5 1.73 1 0.289 1000 0.084 7.0E-06

Table 8.14: Spreadsheet for uncertainty of corrections at 80 mg

Load: 80 mg
Component Values
Component (/ = 1, 2,. .. 8) Ui (mg) ki Ig -I Ui (mg) V/' u,^ uNvi
6. Resolution of r 0.5 1.73 1 0.289 1000 0.084 7.0E-06
1. Resolution of z 0.5 1.73 1 0.289 1000 0.084 7.0E-06
8. Rounding of the final result 0.05 1.73 1 0.029 1000 0.001 7.0E-1Q

A summary of the correction uncertainty across the optical range is shown in Table 8.15.
Table 8.15: Summary table for the expanded uncertainties of corrections of the optical scale
Load (mg) 20 40 60 80 100 1
C/(mg) 1.7 1.7 2.0 2.0 ________
B. Dial settings
As discussed in section 7.4.6.2, if one uses the Simple Tolerance test to determine the
maximum correction of the balance, only the uncertainty of this correction, usually in the
largest decade, is needed. If all the dial-setting corrections are measured, all the corresponding
uncertainties should be calculated. However, for simplicity, only the imcertainty of the 9 of
each decade is calculated. The uncertainty components and procedures for the determination of
uncertainty for 0.9 g and 900 g are outlined here.
(Bl) Mass value of calibrating weight
A set of weights was used for the calibration of the dial settings of this balance. The
expanded uncertainty for each of the calibrating masses in the chosen set, as obtained from the
calibration report, is shown in Table 8.16. The report also gives 2.04 equivalent to = 30,
for the coverage factor used in each. When used to calibrate the balance, the weights are used
on their own or in combination to make up the desired load. Since the mass uncertainties for
weights from the same set are highly correlated, the uncertainty of any one combination of
weights should be added (assuming for convenience, full correlation). The uncertainties for
the combinations are given in Table 8.17.
Table 8.16: Calibration uncertainties for the set of weights

Nominal mass Uncertainty Nominal mass Uncertainty
value (g) (mg) value (g) (mg)
0.2 0.02 50 0.1
0.5 0.025 100 0.15
2 0.04 200 0.3
5 0.05 500 0.75
20 0.08 1000 _____ 1.667
Table 8.17: Uncertainties of the combined calibrating weights used to determine the dial settings
corrections
Load (g) Weight used Combined uncertainty (mg)
0.9 0.5 g + 0.2 g + 0.2' g 0.065
9 5g + 2g + 2'g 0.13
90 50g + 20g + 20'g 0.26
900 500 g + 200 g + 200' g 1.35
1 1000 lOOOg 1.667
(B2) Instability of weight(s)
As explained in section A above, the standard uncertainty due to the instability of the
calibrating weight was taken to be 8% of the maximum permissible error (MPE) for the OIML
class Fl set with the number of degrees of freedom equal to 4 (t' = 4) and is shown in Table
8.18. These values of uncertainty are standard uncertainties, so for each, = 1.

Nominal mass Instability Nominal mass Instability
value (g) Uncertainty (mg) value (g) Uncertainty (mg)
0.2 0.0048 50 0.024
0.5 0.0064 100 0.04
2 0.0096 200 0.08
5 0.012 500 0.2
20 0.02 1000 0.4 1
Unlike the uncertainties of the mass values of the calibrating weights, the instability
uncertainties among the weights are taken to be uncorrelated. Therefore, when a combination
of weights is used, their instability uncertainties are combined in quadrature (equation L.3).
This is done in the spreadsheets shown in Table 8.19 & Table 8.20.
(B3) Repeatability of measurement
The repeatability of the measurement is the standard deviation of the difference (r - z)

(discussed in section 7.3.2). It was found to be 0.97 mg for loads below 50 mg and 1.2 mg for
loads above 50 mg with 9 degrees of freedom (^ = 9).
no 7. CALIBRATION OF SINGLE-PAN, TWO KNIFE EDGE BALANCE
Since double weighing is used to determine C, the standard deviation of the average of
two values (double weighing) is ^2 smaller. In this example the lower end of the optical scale
(below 50 mg) is used, so the repeatability for loads is 0.686 mg for both 0.9 g & 900 g loads.
Being a standard deviation, k, = 1.
(B4) Resolution of balance
Using the same reasoning as is given in section A above, we have:

both r and z, U, = 0.5 mg, ki = 1.73 and v; = 1000.
(B5) Rounding of the result
The final measurement result (values of C and U} will be rounded to the nearest 0.1 mg, i.e.,
with a rounding uncertainty of ±0.05 mg. Thus, we put CZ, = 0.05 mg, = 1.73, being
rectangular, and v; = 1000.
(B6) Other sources of uncertainty
included in the repeatability of measurement (section 7.3.4). Also, the effect of hysteresis
(section 7.3.5) would also be negligible when the balance is properly adjusted.
The calculation
The uncertainty components and the calculations for the corrections to the dial settings are
tabulated in Table 8.19 for 0.9 g and Table 8.20 for 900 g. The formulae used in these
spreadsheets are the same as those used in Table 8.13 and 8.14.
Table 8.19: Spreadsheet for the uncertainty in the correction for a dial setting of 0.9 g, for a single-pan
Load: 0.9 g
Component Values
Component (/ = 1, 2,... 8) U, (mg) ki |g -| u, (mg) Vi Ui^
uNvi
1. Mass value for 0.9 g 0.065 2.04 1 0.032 30 1.0E-03 3.4E-08
2. Mass instability for 0.2 g 0.0048 1 1 0.005 4 2.3E-05 1.3E-10
3. Mass instability for 0.2* g 0.0048 1 1 0.005 4 2.3E-05 1.3E-10
4. Mass instability for 0.5 g 0.0064 1 1 0.006 4 4. IE-05 4.2E-10
6. Resolution of r 0.5 1.73 1 0.289 1000 0.084 7.0E-06
7. Resolution of z 0.5 1.73 1 0.289 1000 0.084 7.0E-06

Table 8.20: Spreadsheet for the uncertainty in the correction for a dial setting of 900 g, for a single-pan
Load: 900 g
Component Values
Component (/ = 1,2, ... 8) Ui (mg) ki |c,-| Ui(mg) Vi uNvi
1. Mass value for 900 g 1.35 2.04 1 0.662 30 0.438 Q.Q06
2. Mass instability for 200 g 0.08 1 1 0.080 4 0.006 1 .OE-05
3. Mass instability for 200* g 0.08 1 1 0.080 4 0.006 1.0E-05
4. Mass instability for 500 g 0.2 1 1 0.200 4 0.040 4.0E-04
6. Resolution of r 0.5 1.73 1 0.289 1000 0.084 7.0E-06
7. Resolution of z 0.5 1.73 1 0.289 1000 0.084 7.0E-06

Degrees of freedom Ve/r 41
Table 8.21: Summary table for the expanded uncertainty in the corrections for dial settings
i Dial setting (g) Uncertainty of correction
(mg)
0.1-0.9 1.7 2.1
1-9 1.7 2.1
10-90 1.7 2.1
100-900 2.2 2^
8.2.2 Uncertainty of weighing
In this case, a calibrated balance is used to measure the mass value of an object. The model
that expresses this measurement is:
M = Cop + Cdiai + R (8.6)
Where M is the mass value of the object placed on the pan, Cop is the optical scale correction,
Cdiai is the dial setting correction, r is the optical scale reading and R is the dial setting when
the test weight is on the pan and z is the zero reading with no load on the pan.
The sensitivity coefficient for converting imcertainties in r, C ox z to components of

uncertainty for M is 1.
8.2.2.2 The uncertainty of weighing: an example
Consider the same single-pan two-knife-edge balance as in section 8.2.1.4 with a range of
1000 g and a resolution of 0.001 g. Strictly speaking, uncertainty of weighing should be
calculated for each load. However, for simplicity, the uncertainty of weighing for each decade
of loads e.g. 0.1 g to 0.9 g, 1 g to 9.9 g, ...100 g to 1000 g is calculated. The uncertainty of
weighing of the largest load will be the uncertainty of weighing for that decade. Sample
calculations are given for loads of 0.9 g and 1000 g, the latter case being slightly more
complicated as it involves all dial settings (999.9) and the full optical range.
(1) Correction to optical scale
The uncertainty of the optical scale corrections can be obtained from Table 8.15 (rounded).
Becasue the uncertainty of weighing for a particular range applies to any load within that
range, the maximum uncertainty is taken, i.e., 2.04 mg with ki = 2.16 and v, = 13 (see Table
8.14).
(2) Correction to the dial settings
The uncertainty of the correction for the different dial settings can be obtained from Tables
8.19, 8.20 and similar.
Load of 0.9 g: t/, = 1.69 mg with ^, = 2.11, and v; = 17.
Load of 1000 g: It will be the combination of 4 dial settings 0.9 g, 9 g, 90 g and 900 g.
(3) Repeatability
The repeatability of the measurement Sr is the standard deviation of the difference (r - z)

(discussed in sections 7.3.2). It was found to be 0.97 mg for loads below 50 mg and 1.2 g
otherwise. As the uncertainty of weighing is to determine the uncertainty of one particular
weighing (a single weighing minus a zero reading), the repeatability component will equal Sr.
In this example, the maximum value of 1.2 mg will be used. Therefore:
For both loads, of 0.9 g and 1000 g: CT, = 1.2 mg with = 1, and k = 9.
As explained in 8.2.1.2 A, for both r and z, f/, = 0.5 mg, ki = 1.73 and k = 1000.
The final measurement result (values of C and U) will be rounded to the nearest 0.1 mg, i.e.,
with a rounding uncertainty of ±0.05 mg. Thus, we put t/, = 0.05 mg, k^ = 1.73, being
rectangular, and Vi = 1000.
The effect of off-centre loading and hysteresis are negligible as explained in 8.2.1.2 item (A6).
The uncertainty components for the calculation of uncertainty of weighing for 0.9 g and
1000 g load are shown in Table 8.22 and Table 8.23 respectively. The formulae used in these
spreadsheets are the same as those used in Tables 8.13 and 8.14.
Table 8.22; Spreadsheet for uncertainty of weighing for a single-pan two-knife-edge balance at 0.9 g
Load: 0.9 g
Component Values
Component (/ = 1, 2,... 6) U (mg) ki Iq I u, (mg) Vi /v,
2. Dial setting correction (0.9 g) 1.69 2.11 1 0.801 17 0.642 0.024
4. Resolution of r 0.5 1.73 1 0.289 1000 0.084 7.0E—6
5. Resolution of z 0.5 1.73 1 0.289 1000 0.084 7.0E-6

Table 8.23: Spreadsheet for uncertainty of weighing for a single-pan two-knife-edge balance at 1000 g
Load: 1000g
Component Values
Component (/ = 1, 2,... 9) Ui (mg) ki Ui (mg) Vi Ui^ u^/vi
2. Dial setting correction (0.9 g) 1.69?; 2.12 1 0.801 «17 0.642 0.024
3. Dial setting correction (9 g) 1.70 2.12 1 0.802 17 0.643 0.024
1. Resolution of r 0.5 1.73 1 0.289 1000 0.084 7.0E-06
8. Resolution of z 0.5 1.73 1 0.289 1000 0.084 7.0E-06
1
The uncertainty of weighing across the full range of the single-pan two-knife-edge
balance is given in the summary Table 8.24.
Table 8.24: Single-pan, two-knife-edge balance - summary table for uncertainty of weighing
Load (g) Uncertainty of weighing (mg) ki
0.1 -0.9 3.6
1-9.9 3.9 2.0
10-99.9 4.2
100-1000 4.7
The uncertainty of weighing at lOOOg was calculated assuming that all the dial settings of the
balance (999.9 g) and the full optical range were used.
8.2.3 Limit of Performance for a single-pan, two knife edge balance
The Limit of Performance F of a balance is a quantity that has been in use for many years and
is not an uncertainty in the normally accepted sense. Instead, it gives a very conservative
upper limit to the measurement error that may be expected when no corrections are made to
balance readings. Thus if r and z are the balance readings, the mass of the object lies in the
range (r - z) ± F. It is emphasised that the Limit of Performance is an upper and lower bound.
In many applications the balance may weigh much more accurately than this figure.
The Limit of Performance F is defined by the following equation:
P= +C^, (max) + U(C^ (max)) + (max) + C/(C„^ (max)) (8.7)

where;
Wmax is the maximum value of the repeatability of measurement of the balance or one-
third of the resolution of the balance, whichever is greater.
Cop(max) is the magnitude of the maximum correction to the optical scale reading
C/(Cop(niax)) is the corresponding expanded uncertainty of Cop(max)
QiaKîax) is the magnitude of the maximum correction to dial reading, whether obtained in
a complete calibration of the dial settings or in a tolerance test
U{Cdia^ra&x)} is the corresponding expanded uncertainty of Q^Xmax)
In the example of section 7.4 (a balance with a range of 1000 g and resolution of
0.001 g) the maximum value for the repeatability of measurement was found to be 0.0012 g at
the high end of the optical scale. As this is greater than one-third of the resolution, uâx is
taken to be 0.0012 g.
The table in section 7.4.2.1 shows the maximum correction of the optical scale to be
0.00052 g and the maximum correction of the built-in weights (section 7.4.2.2) is 0.0008 g,
using the Simple Tolerance test, hence from equation (8.7), the Limit of Performance of the
balance is
F = 2.26 X 0.0012 + 0.00052 + 0.002 + 0.0008 + 0.0022 = 0.0082 g

Definitions
The following definitions are provided to assist the users of this book to interpret the
information relating to mass and balance calibration. Therefore, the definitions are particular
to the context in which these terms are used in the book.
However, there are more general definitions of some of the terms in the following list. The
broader definitions, which apply to all fields of metrology, can be found in references [4,11].
Arrestment Mechanism for lifting the knife edges off the planes and keeping the beam
and pan steady.
Buoyancy Force on an object due to the fluid in which it is immersed, usually air in the
context of this book. It is normally expressed in units of mass.
Combined The standard uncertainty of the value obtained for the measurand. It is found
standard by combining the standard uncertainties associated with the input quantities.
uncertainty
Confidence The interval about a measurement within which the true value of the
interval measurand is expected to lie with a high level of confidence. See also
“Expanded uncertainty”
Conventional A very widely used mass value that usually avoids the need for buoyancy
mass corrections in measurement. It is approximately equal to true mass. The
precise definition is: The conventional mass of an object is equal to the true
mass of a weight of density 8000 kg.m'^ that at a temperature of 20
balances the object in air ofdensity 1.2 kg.m'^.
Correction A value that must be added algebraically to the uncorrected result of a
measurement to compensate for systematic error. A typical example would
be adjusting balance readings using corrections from a calibration report.
Coverage factor The factor by which the combined standard uncertainty is multiplied to
obtain the expanded uncertainty.
Damping Means by which a swinging balance is brought to rest.
Decade Group of four (or more) weights from which any integral value from 1 to 10
may be formed using various combinations. For weight sets, the most
common group is 5, 2,2' and 1
115
116 DEFINITIONSAND SYMBOLS
Degrees of freedom The number of degrees of freedom for a standard uncertainty is a measure of
the reliability of that estimate. For the standard deviation of a set of n repeat
measurements, the number of degrees of freedom is n - 1.
Dial reading or The reading on a mechanical dial used to indicate the mass of the setting
setting weight lifted off the beam for a single-pan, two-knife-edge balance.
Discrimination The smallest change in mass value that can be indicated by a balance.
Double weighing Substitution weighing carried out twice. On the second occasion, the order
in which the objects are placed on the balance pan is reversed. Also called
double substitution weighing.
Drift A progressive unidirectional change in the reading of a balance.
Dual or multi- Balance with two or several measurement ranges. The ranges may differ
range balance only in resolution or they may have different measuring systems.
Expanded The expanded uncertainty U is so defined that ± U is an interval about a
uncertainty measurement within which the true value of the measurand is expected to lie
with a high level of confidence. The internationally accepted level of
confidence used is 95%.
Hysteresis The difference between the indications of a measuring instrument when the
same value of the quantity measured is reached by increasing or decreasing
that quantity.
Input quantities Quantities used in the determination of the measurand, such as those
measured during the ‘measurement’, calibration data and reference values.
(See also “Measurand” and “Model of a measurement”.)
Instability (of a The tendency for the mass of an object to change with time.
mass value)
Mass (or True Mass is a measure of an object’s resistance to acceleration, or equivalently, it
mass) determines the gravitational force experienced by an object. This is an
intrinsic property of an object and is independent of physical changes, such
as gravity, temperature etc. The unit of mass is the kilogram.
Measurand A quantity to be determined by measurement, which often cannot be
measured directly but is obtained from input quantities, some of which are
measured directly. In this book, the measurand is usually a mass value and
the most important input quantities are balance readings and a reference mass
value. (See also “Model of a measurement”.)
Model of a An expression or set of equations giving the measurand in terms of the input
measurement quantities.
Optical scale A device that indicates the deflection of a balance beam by optically
projecting a pointer on to a scale. See also “Scale”.
Reading of a The indication of a balance after the beam has stopped swinging and hence
balance the display is changing only very slowly if at all.
DEFINITIONSAND SYMBOLS 117
Repeatability The closeness of the agreement between the results of successive

measurements of the same quantity carried out by the same method by the
same observer at quite short intervals of time. It is measured by the standard
deviation of several such successive measurements.
Residual The difference between the calculated value and the observed value of a
quantity.
Resolution The smallest difference in an instrument’s indication that can be reliably
read.
Rest point The reading of a balance.
Scale An ordered set of marks, together with any associated numbering, forming
part of the displaying device of a measuring instrument. For example, the
means by which an optical pointer indicates the deflection of the beam in a
single-pan, two-knife-edge balance.
Scale division or The interval between two adjacent scale marks.
interval
Sensitivity A measure of balance response to changing load. It is measured as the
change in the balance reading per unit mass. As most balances read directly
in mass units, the sensitivity is usually dimensionless. For a well adjusted
balance it will be exactly equal to one.
Sensitivity The reciprocal of Sensitivity, i.e,, the change in load that must be applied to a
reciprocal or SR balance to change the reading by unit mass (or by one division, for balances
that do not read directly in mass units). As most balances read directly in
mass units, the SR is usually dimensionless and for a well adjusted balance it
will be exactly equal to one.
Stability (of a The ability of a balance to maintain a constant reading with time.
balance)
Standard deviation A measure of the spread of possible values of error or in a fluctuating
variable. The standard deviation s of a sample of repeat measurements is:
r
-
where
n = the number of individual results x,
X = the arithmetical mean of the individual results x,
Standard An uncertainty expressed as a standard deviation.
uncertainty
Substitution Weighing procedure in which a reference weight is replaced by the unknown
weighing object. The mass difference between the weight and the object is the
difference between the two readings.
118 DEFINITIONSAND SYMBOLS
Tare Facility in a balance to enable the balance reading to be made equal to zero
with an object on the pan.
True mass See ‘Mass’
Uncertainty A parameter that expresses the extent of error in any given measurement or
estimate, usually expressed in the form ±U.
Variance Square of the standard deviation.
Weighing The process by which the mass value of an object is determined.
Weight An artefact that serves as a standard of mass.
Symbols
a Measured mass difference
Semi-range of a rectangular distribution
Ci a sensitivity coefficient in uncertainty calculations
C Correction
A check weight
Cm Dial setting correction at load M
Correction to balance reading at load M
Cdia\ Dial setting correction
Cop Optical scale correction
F Limit of Performance of a balance
g Local acceleration due to gravity
k Coverage factor
ki Reducing factor that converts an initial uncertainty estimate into a standard deviation
m True mass of any object
rnp True mass of a reference weight
niT True mass of a test weight
M Conventional mass of any object
Me mass value of a check weight
Mr Conventional mass of a reference weight
DEFINITIONSAND SYMBOLS 119
^sens Conventional mass of a sensitivity weight

Nif Conventional mass of a test weight
Me Corrected conventional mass obtained using a calibrated balance
n number of measurements
r A balance reading
A balance reading with the reference weight on the pan
Resolution of a balance
^^sens
Change in balance reading during measurement of sensitivity reciprocal
R A reference weight
•Sm Standard deviation of a balance for a single mass comparison
Sr Standard deviation of mass measurements on a balance
Sm Standard deviation of mass values
S Sensitivity reciprocal of a balance
t A balance reading with the test weight on the pan
T A test weight
u Standard uncertainty
Ub Standard uncertainty of a measured mass due to buoyancy
Uc Combined standard uncertainty
u Expanded uncertainty
z Zero reading of a balance
a/ A correction to conventional mass
Difference in optical scale readings with and without load on the pan
V Number of degrees of freedom
Number of (effective) degrees of freedom for Uc
P Density
Pa Density of air
pR Density of a reference weight
Pt Density of a test weight
^Pt Total width of OIML density limits for the class of test weight T
120
“Vm here to calibrate the weights

Appendix A
Use of balances
There are a number of factors affecting the performance of a balance and the balance reading.
They can be divided into environmental and human factors.
A.l Environment
A balance performs best in an ideal environment. Although this can be defined as one where
further improvements make no change to the performance of the balance, it is almost
impossible to quantify and the requirements change with the sensitivity and type of balance.
In general the more sensitive the balance the better the environment that is required. This
appendix gives a general guide to the principles involved in deciding whether the location of
a balance is satisfactory. The general constraints of economy and space will sometimes be the
overriding consideration. The principles are for the ideal case. It may be necessary for the
balance user to compromise with some of the conditions depending upon the type of balance
and the accuracy required. As a guide it can be stated that to obtain weighing accuracy
required for OIML class E2 or better, there is almost no room for compromise.
Al.l Temperature
The most important environmental factor to consider is temperature, or rather temperature

gradients. These not only cause changes in lever ratio and other electrical and mechanical
properties of electronic balances, and variations in the arm length ratio of single pan balances,
but also set up air currents within the balance case.
The room temperature should be stable. The actual value of the temperature is
relatively unimportant; it is the stability that matters. Small fluctuations around a mean have
little effect on most balances. However, continual increase or decrease in temperature over
several hours causes temperature gradients in the balance mechanism, resulting in drift in the
readings and sometimes changes in the sensitivity.
If possible the temperature of the balance room should not change by more than ±2 or
3 °C during any eight-hour period. If the room cools down at night but is stable during the
day, then the balance will spend most of the day reaching equilibrium with the room, and
hence the reading will drift even though the room temperature is stable. Therefore, it is
recommended that air-conditioning be left on throughout the night for high-accuracy
measurement. The 2 or 3 °C stability is what is ideally required for balance calibration, but
less stringent conditions can be tolerated for weighing depending upon the accuracy required.
121
122 A. USE OF BALANCES
Convectional air currents due to temperature gradients probably cause the greatest
disturbance to balance readings. These often result from the operator sitting in front of the
balance or from placing a hand inside the balance case. When this happens time must be
allowed for the balance and the environment to stabilise. To avoid the problem of hand
heating, use long-handled weight lifting devices where possible, instead of a gloved hand, for
accurate weighing.
Before making mass measurement, make sure the balance has been switched on for the
warm-up period recommended by the manufacturer, or at least 30 minutes if this period is not
known. It is also recommended that the balance be “exercised” by placing a weight, which is
used for the mass measurement, on the balance a few times before real measurement is made.
Direct sunlight should not fall anywhere near a balance used for accurate weighing.
Warm objects should be kept well away from balances in use, and the lighting should be
fluorescent. A drift in the balance indication is much more likely to be due to temperature
effects than anything else.
Al.2 Humidity
Humidity is relatively unimportant, but conditions should never be allowed to reach the point
where condensation is likely to occur. For high-precision weighing relatively stable relative
humidity, i.e. constant to ±5% RH, is desirable to minimize surface effects. The range 45-
60% is recommended.
Al.3 Pressure
For mass verification work at sea level, the atmospheric pressure is not an important factor,
provided the density of the weights satisfies OIML requirements. At altitudes greater than
300 m, the atmospheric pressure should be measured (an aneroid barometer of reasonable
quality will suffice) and buoyancy corrections may be necessary (See sections 2.4 and 5.2.4,
and Appendix H). Assuming that the laboratory temperature is in the vicinity of 20°C,
accurate mass verifications should not be conducted if the pressure is below 960 hPa, unless
buoyancy corrections are made (see eqn. 2.7).
An air-conditioned room will often be at a slightly higher pressure than ambient

(approximately 40 Pa (0.3 mmHg)) and opening the door will cause a pressure pulse that may
disturb the balance. This pressure pulse can also occur in a reasonably well-sealed room when
the door is opened. In these situations the door should be locked during weighing. Normal
variations in atmospheric pressure should not be a problem.
Al.4 Air currents

An air current of velocity 4.6 cm.s’’ impinging normally on a pan of area 78 cm^ (10 cm
diameter) produces a force equivalent to a mass of 1 mg on the pan. The air speed in an air-
conditioned laboratory is typically about 30 cm.s'\ The effect is directly proportional to the
squares of both the pan diameter and the air speed, i.e. doubling either of these parameters
produces four times the force on the balance pan.
A.l ENVIRONMENT 123
Thus draught shields are essential for reasonably precise weighing as air currents can
easily cause errors of several parts per million. After loading weights on to a balance and
closing the balance draft shield, a little time up to about 10 seconds must be allowed for
conditions within the shield to stabilise.
Al.5 Gravity
Electronic balances match the gravitational force acting on the weights with a non-
gravitational force. Hence the scale value will vary if the balance is moved from one place to
another, making a readjustment necessary. (It is not recommended that a correction be
applied to balance readings to account for changes in gravity.) The scale value is readjusted
using a standard mass according to the manufacturer’s instructions. On some balances, this
process is largely automated. The value of g, local gravity, varies with latitude, height and
local anomalies. If the balance has been transported to a different building or up a flight of
stairs, it should be re-calibrated as a matter of course.
A1.6 Vibration
All balances are susceptible to vibration to a greater or lesser extent. Electronic filtering and
the appropriate choice of time constants (often via a microprocessor) can help minimise the
effects of vibration on the display of force-compensation balances. Most manufacturers
incorporate these options into the electronics associated with the balance. For knife-edge
balances, severe vibration will damage the knives.
Single-pan beam balances measure tilt and hence are very sensitive to change in level
of the balance. Therefore the use of anti-vibration mounts to eliminate vibration is not always
a satisfactory solution (Al. 1.7 (c)). This is particularly so for balances with a weighing
capacity of 1 kg or more, because moving masses around on the bench may change the
balance reading. This applies even to electromagnetic-force-compensation balances, which
are not quite so sensitive to change in level. The only real solution is a solid bench in a
vibration-free environment.
Al.7 Balance bench
The following is a list of the desirable qualities for the construction and siting of a balance
bench.
(a) The bench should be made of stone, e.g., marble, granite, slate, terrazzo, etc,, and be a
minimum of about 40 mm thick. It should not be made of reinforced concrete because of
possible magnetic effects of the reinforcing,
(b) The bench should preferably be mounted on brick pillars on a concrete floor and free
from any wall. The bench should not be made of timber or mounted on a wooden floor.
The floor should be the ground or basement floor of the building. It should not be a
suspended floor,
(c) The bench should be placed directly on top of the pillars without anti-vibration
mountings (see section Al, 1,6), At most, sheet lead or thin cork could be used.
124 A. USE OF BALANCES
(d) The pillars should be spaced so that one balance is placed mid-way between two pillars.
It is better if the bench tops for each balance are separate slabs of material.
(e) The bench should be sited in a room free from vibration due to machinery, passing traffic,
etc. A lot of vibration is transmitted through the floor, and if this is the case an
independent foundation should be made for the bench. This will mean going to a depth
of about one metre, preferably on to rock. Care should then be taken to ensure that this
foundation is completely free from the floor.
(f) For temperature stability an internal room with artificial lighting may be better than a
room with air-conditioning. Failing an internal room, choose a room on the south side
(in the southern hemisphere). If neither of these options is available then the windows
should have an outside sun shield. In general it is better not to air-condition if adequate
temperature stability can be obtained without it.
(g) The flow of air from any air-conditioning must be directed away from the balance bench.
(h) Any surfaces, such as concrete, that are likely to give off dust should be sealed.
A.2 Laboratory practice

A2.1 Electrostatic effects
Use the manufacturer's draft shield as far as possible. If a new or modified draft shield is
needed, conductive material should be used. To check a balance for sensitivity to external
electrostatic effects a glass or plastic object charged by rubbing with wool or a synthetic fabric
can moved around near the balance.
Weights are metallic and hence not normally subject to electrostatic charges, but non-
conductive samples or weighing vessels (glass, plastic) can become electrostatically charged
through friction resulting from handling (eg. pouring from one container to the other). This can
be prevented by increasing the relative humidity in the room to 45-60% or by placing weighing
vessels in an electrically conductive container. A humidifier, an anti-static gun or an ionizing
blower may eliminate the electrostatic charge.
A2.2 Magnetic effects
Magnetic forces on an object being weighed may interfere with the balance readings, though
these effects should always be negligible for properly constructed weights and balances. If it is
necessary to weigh magnets or magnetic materials then some special precautions should be
taken. These could include demagnetizing the object or placing it in a container made of
Mumetal. The magnetic force can also be minimized by keeping the object at a distance from
the weighing pan by means of a hanger. Magnets or magnetic material should be kept at a
distance from the balance chamber. The presence of magnetic forces is revealed by poor
reproducibility and a strong dependency on the position and location of the object on the
weighing pan.
A.2 LABORATORY PRACTICE 125
A2.3 Loading the weights on the balance pan
It is important to load weights on to a pan as centrally as possible. The shift or comer-load

error that can result from off-centre loading is dealt with in chapters 6 and 7. However, poorly
positioned weights can damage any reasonably sensitive balance. For single-pan beam
balances, pan swing, increased knife wear and errors resulting from knife-edge imperfections
can result.
When using sensitive balances, place the weights gently on the pan without scraping
them across its surface. If the balance needs to be released and arrested then these operations
should be done as smoothly as possible.
A2.4 Levelling
Most balances are provided with a level indicator and the balance should be carefully levelled
prior to its first use or after being moved. The level should also be checked from time to time.
Small changes in level of electronic balances alter the zero reading and the scale value. For
single pan beam balances, imperfect levelling can lead to significant errors.
A2.5 Cleanliness
Balances used in a good environment should not need cleaning very often, but a loss of
repeatability could indicate the need for it. Using a camel hair brush, it may be possible to
clean the knives and bearing planes while the balance is arrested.
126
‘‘Air currents can cause serious problem in vêighing

Appendix B
Use and handling of weights

This book uses and recommends the OIML classification of weights. When buying new
weights, it is convenient to seek quotations in terms of these classifications. When using old
weights, the user needs to estimate which classification they best fit, as this influences their use
and calibration.
B.l Types of weights

The calibration of balances requires weights of adequate accuracy for the balance being
examined. Therefore once a set of weights has been obtained and calibrated it is important that
it be handled properly. The aim of this appendix is to give some guidance as to how this may
be done.
The OIML classification divides weights into 7 categories: Ei, E2, Fi, F2, Mi, M2 and
M3. For details, refer to OIML Recommendation Rill [3].
(a) Class E weights are integral weights made of non-magnetic stainless steel with glossy
appearance. Class Ei weights are used mostly at national measurement institutes such as
NML. They are working standards used for the calibration of E2 weights and are
traceable to national secondary standards of mass (see Appendix C). Class E2 weights are
used by laboratories requiring very accurate reference standards (see Appendix C) and for
the calibration of high-precision balances.
(b) Class F weights may be non-integral or two-piece made of non-magnetic stainless steel or
drawn brass with glossy appearance. They may have a small adjusting cavity, which shall
be closed by a lifting knob. Class F weights are used as reference standards, for the
calibration of lower class weights, and for the calibration of balances.
(c) Class Ml weights may be two-piece brass or gray cast iron weights depending on the mass
value. The surface should be smooth.
(d) Classes M2 and M3 weights may be two-piece brass or gray cast iron weights with smooth
surface or suitable coating. Class M weights are used for the calibration of lower class
weights and for the calibration of balances and industrial weighing machines.
The National Measurement Laboratory will calibrate weights of classes Ei to Mp

Although the Laboratory does not prescribe recalibration periods, it recommends that one-
127
128 B. USE & HANDLING OF WEIGHTS
piece stainless steel weights be recalibrated at least every five years, stainless steel of more
than one piece every three years, and others every 2 years.
B.2 Handling of weights

Weights of class E and F should never be touched with bare hands. The small weights should
be handled with bone- or plastic-tipped tweezers and the large weights with clean gloves
(chamois, cotton or plastic) or with a lifting tool. Stainless-steel tweezers, handled carefully,
can be used to pick up the fractional (milligram) masses. For precision measurement, one
should avoid putting hands inside the balance chamber during loading and unloading of the
weights on the balance as the resulting air currents and temperature effects can severely disrupt
weighing.
B.3 Care of weights

(a) When not in use the weights should always be kept in the box or container provided.
(b) Weights from two or more sets should never be mixed.
(c) Weights should never be dropped; they should be placed on the pan or a clean surface. If
dropped, they must be recalibrated.
(d) Weights should never be placed on a dusty or dirty surface, or slid across a surface,
especially on a balance pan.
(e) Weights should not be allowed to clink together. Sometimes it is necessary for them to
touch when putting a few on a pan, but this should be done carefully.
(f) If the weights appear dirty or have been touched, they should be dusted with a soft brush
or lightly wiped with a clean chamois. Classes E and F weights should then be
recalibrated.
(g) Weights should not be cleaned with solvent unless absolutely necessary. However, for
weights that are still too dirty to be used after brushing or cleaning with a chamois,
cleaning with solvent, e.g., distilled water, is needed. Sufficient time should be allowed
for the masses to stabilize before use. Typical stabilization times after cleaning are given
in Table B 1.
Table Bl: Stabilization time after cleaning

Weight class El E2 Fl F2 to M3
After cleaning 7-10 days 3-6 days 1-2 days 1 hour
with alcohol
After cleaning 4-6 days 2-3 days 1 day 1 hour
with distilled
water
(h) For any calibration involving weights, the weights should be in thermal equilibrium with
the ambient conditions of the laboratory, and with the balances as well for Ei to Fi
B. 3 CARE OF WEIGHTS 129
calibration. The times required for the weights to reach temperature equalization are
summarised in the Table B2.
Table B2: Thermal stabilization in hours
AT Weight Size E2 F, F2
5°C 10-50kg 12 6 3
1-5 kg 6 2 1
100-500g 3 2 1
10-50g 2 1 1
<10g 1
0.5°C 10-50kg 3 1 0.5
1-5 kg 1 1 0.5
100-500g 1 0.5 0.5
<100g 0.5
AT is the initial difference between weight temperature and ambient temperature
B.4 Estimation of the OIML class for old and/or special weights
Some weights were manufactured before 1994 when OIML Rill [3] came into effect and
other weights have a special design or a non-standard nominal value. The calibration
laboratory or general user may assign an OIML class based on the history of stability of the
weights or they may be classified according to the criteria given in section B.l. (a to d). OIML
Rill provides more technical details relevant to the classification of weights.
Be careful!
130
Appendix C
Traceability of Australian mass standards

The Australian primary standard of mass is copy No.44 of the international prototype
kilogram. It is normally kept under two bell jars in a safe at NML.
Copy 44 was manufactured in 1938, and although not delivered to NML imtil 1949, it
has a measurement history extending back to 1939. Since delivery it has been returned to the
BIPM on five occasions. Table C.l shows the values obtained for the mass of copy 44 since
1939. The standard uncertainty of these BIPM calibrations used to be 8 pg, but as a result of
improved balance technology and weight washing techniques, the standard uncertainty of the
1991 calibration was 2.3 pg.
Table C.l: The mass of copy 44 of the International Prototype Kilogram obtained in comparisons
carried out at the BIPM
Date Mass
1939 1kg + 269 pg
1946 + 270
1947 + 283
1954 + 270
1964 + 262
1968 + 297
1979 + 283
1991 + 287
The Australian secondary standards consist of eight high-quality 1 kg stainless steel

weights. They are calibrated every two years by comparison with copy 44 using a highly
automated 1 kg balance that has a standard deviation in the range 1 to 2 pg and is able to inter-
compare up to four weights at a time. The mass uncertainty of the secondary standards at the
95% confidence limit is 30 pg. Most of this uncertainty arises because of the difference in
buoyancy between copy 44 (p = 21527 kg.m’^) and the stainless steel secondary standards (/?=
8000 kg.m-’).
The third level of standards consists of two sets of OIML class E] weights, which extend
from 20 kg to 1 mg and are known as the working standards. These weights are used for the
calibration of class E2 weights.
131
132 C. TRA CEABILITY OF A USTRALIAN MASS STANDARDS
The primary standards of the state and territory trade measurement laboratories and of
some NATA laboratories form the fourth level of Australian standards. The quality of these
sets is approximately class E2.
Figure 1 shows the traceability of the Australian standards. For legal purposes,
measurement traceability is documented using a special type of report called certificate of
verification. The National Measurement Laboratory (NML) issues certificates of verification
of State Primary Standards to the appropriate State verifying authorities under Section 9 of the
National Measurement Act 1960 and Regulation 18 of the National Measurement Regulations
1999. (These authorities are the Trade Measurement Laboratories of the States and
Territories.) The accuracy of the standard is expressed as an uncertainty, at the time of
verification, not exceeding the permissible uncertainty stated in Schedule 5 of the National
Measurement Regulations 1999. The NML, National Standard Commission (NSC) and
verifying authorities in Australia appointed under Regulation 73 may also issue certificates of
verification of reference standards of measurement under Regulation 13 of the National
Measurement Regulations 1999. All the uncertainties are calculated at 95% confidence level.
The values and uncertainties of relevant influence quantities, at the time of verification, will
also be stated in the certificates.
Figure C.l: Traceability of Australian Standards

Appendix D
Least-squares analysis of weighing schemes

Weighing schemes are analysed using linear regression (least-squares analysis). This involves
finding values for the test masses that best fit the available data, which include the value of the
reference mass and the results of the mass comparisons. Since all measurements are subject to
uncertainty, it will not generally be possible to find test mass values that fit the measurements
exactly.
As a simple example, consider the closed cycle scheme given by equations (4.3) in
section 4.2.1. After the measurements have been made, one can write down four equations:
(D.l)
Al ~ A2 ~ ^2
A2 ~ P^R ~ ^3
where Îr is the mass of the reference weight Ri, jUi is the mass of Ti and JU2 is the mass of Tj.
The symbol // represents the ideally correct value of these masses, which can never be known
as all measurements contain uncertainty. Hence the use of approximation signs in
equation (D.l)
The last three equations express the results of the mass comparisons made in the closed
cycle. The first equation states that the best available mass value for the reference weight is uq
(obtained, presumably, from a report). Equations (D.l) can be written in the form shown in
equations (D.2), where the quantities e, are error terms.
A/e ^0 +ô
At ? ~Ai fli +ei

(D.2)
A1-A2 ^2 +^2
A2 “At ? +^3
Although it is not possible to evaluate the masses or error terms in equation (D.2), one
can calculate mass values Mr , Mi and M2 that fit the equations best by giving the smallest error
terms. We call Mr , Mj and M2 the best-fit or least-squares mass values.
133
134 D. LEAST- SQUARES ANALYSIS OF WEIGHING SCHEME
When these best-fit masses are substituted into equations (D.2), one can calculate the
resulting error terms, which are called the residuals 7?,.
7?o M Uq
= M-flj
(D.3)
7?2 = Af, -
-Mj^ - Uy
The best-fit masses are determined by finding the mass values that give a minimum value for
the sum of squares of the residuals S:
S = (A/, - a„)" + - Jl/, - a,)’ + (ji/, - A/, - a Y (D.4)
One can use a well known rule of calculus that states that, for the minimum value of S, the
partial derivatives of 5 with respect to Mr , Mi and M2 will all be zero:
— = 0 = 2(A/,-a,) + 2{M,-M,-a,}
^S
=Q= .M,-a,} + 2(M, (D.5)
dS
— = 0 = -2{M,-M,-a,} + 2(M,-M, -aj
So that
3Mj^-My-M2 = 00+05-03
-Af^+2Mj-M2 = -«i+«2 (D.6)
- M-M^+ IM^ = - O2 + O3
Solving these equations (which are called the normal equations) leads to
R ~
M, = a.i (D.7)
° 3
Notice that the value of the reference mass is unchanged, as Oo is the value assigned to it in
equation (D.l). This will always happen in weighing schemes that have only one reference
weight as the measurements tell us nothing new about the mass of the reference.
D. LEAST-SQUARES ANALYSIS OF WEIGHING SCHEME 135
Thus simple weighing schemes like closed cycles can be analysed using ordinary
algebra. Decade schemes can also be treated in this way, but the algebra is much more
complicated. Unfortunately, every scheme will have a different solution, in other words, the
equations corresponding to equations (D.7) will be different.
Using matrix algebra it is possible to derive a set of equations that will be the same for
all weighing schemes. The analysis of weighing schemes using matrices is outlined below for
those readers who are familiar with matrix algebra. The procedure is similar to that described
above in ordinary algebra.
It is assumed that the reader knows the basic rules for the addition and multiplication of
matrices. The transpose A^ of a matrix A has the rows and columns interchanged. For
example:
a ^21
^11 ^12 ^13
«12 ^22 (D.8)
^21 ^22 ^23 y
a ^23 >
When a matrix A is multiplied by its inverse A'\ the result is the unit matrix T.
A 0 0"
AA-^ 0 1 0 (D.9)
<0 0 ly
(where is taken to be a 3x3 matrix). It can be shown that for all non-singular matrices
(^7?)' =B' J' and (^')’' = )' (D.IO)
Equation (D.2) can be written in matrix form as follows:
^1 0 0"
a, 1 -1 0
(D.ll)
«2 0 1 -1
<F2,
<-l 0 ly
In symbolic form, equation (D.l 1) can be written as
a = Xfi-e (D.12)
where a is called the observation vector (a vector is a matrix with only one row or column),
and X is the design matrix. Any weighing scheme can be represented by this equation, though
each scheme will have its own design matrix. If M is the vector of best-fit mass values, and R
is the residuals vector, then for the above closed cycle.
f 7^ A
'A//
3/ (D.13)
^2
J J
and corresponding to equation (D.3) we have
/? = (D.14)
and the sum of squares of residuals is
S = R’R = {XM-tt')‘{XM-a)=(M'X'-a'\xM-a) (D.15)
^=0 3^ = 0 dM
Since and 0
^^R ^^R
differentiating equation (D.15) gives
as
^^R
0 (D.16)
and differentiating with respect to Mj and M2 will give similar equations with (1 0 0)
replaced by (0 1 0) or (0 0 1). These three vectors combine to form the unit matrix /
(see equation D.9), so combining equation (D.16) with the two similar equations, we have
rX'(XM-a) + (XM-a)'XI = 0 (D.17)
The two terms in equation (D. 17) are both vectors and the second is just the transpose of the
first. Therefore if they add to zero, then both must be zero. So
IX‘[XM -a) = X XM-X a = 0 (D.18)

D. LEAST-SQUARES ANALYSIS OF WEIGHING SCHEME 137
From which it follows that
M = (x'xYx'a (D.19)
Equation (D.19) gives the best-fit mass values in terms of the design matrix and the
observation matrix. Although this result was derived using a closed cycle as an example,
equations such as (D.12), (D.14), (D.15) and (D.19) will apply to all weighing schemes. So a
computer program based on equation (D.19) can be used to analyse the measurements obtained
from any weighing scheme.
Until now, this treatment of linear regression for mass metrology has two important
defects. Firstly it assumes that all the observations (i.e., all the components of the vector a)
have the same standard deviation, whereas this is seldom the case. Secondly, the standard
deviations of the components of M must be determined so that the uncertainties of these mass
values can be calculated. To rectify this situation, an understanding of covariances and the
variance-covariance matrix is needed.
For a single quantity x, only the standard deviation must be known in order to
calculate its uncertainty. However if two or more quantities are involved, such as in the simple
sum: X + y, one must know not only and Sy but the covariance as well, because the
standard deviation of x + y is given by
=7^x +-’’+2C^ (D.20)
The covariance takes into account any correlation between a measured value of x and a
measured value of y. The uncertainties associated with the measurement of x are frequently
independent of those associated with the measurement ofy, in which case will be zero. See
reference [4] section C.3.4 for a mathematical definition of covariance.
The square of a standard deviation is called a variance, thus s/ is the variance of x. For
a vector such as a in equation (D.l 1), the variances and covariances are collected into a matrix
called the variance-covariance matrix. The variance-covariance matrix of a is
<52
•’0 C„. c'^02 c'^03
c'^01 C.2 c
'^13
V, = (D.21)
c'^02 c1^12 c
’^23
c c'^13 ^c^23
<^03 •^3 )
where sq is the standard deviation of ao, si is the standard deviation of ai, etc, and Cqi is the
covariance between ao and ai etc. Notice that the variances form the diagonal terms of the
matrix and the covariances are the off-diagonal terms.
Provided the buoyancy uncertainty is small, the results of substitution weighings are
independent of each other and of the mass adopted for the reference weight. So for the
weighing schemes described in chapter 4, matrix Va will be diagonal. (Covariances can arise
for schemes with more than one reference weight.) Hence, Va is easily determined from the
standard deviations of the reference mass and of the balance.
To find the standard deviations of the best-fit mass values, one needs to calculate the
variance-covariance matrix Vm for the vector M. It can be shown that if two vectors Y and Z
are related by the equation: Y = AZ, where A is any matrix, then the variance-covariance
matrices Vy and Vz are related by Vr = A VzA‘. If we apply this rule to equation (D. 19), then:
= (x'x)-' x‘v^x{x'x)'' (D.22)
In the derivation of equation (D.19), it was assumed that all the variances sf, S2^ and
are the same (equal to say) and that all the covariances are zero. Therefore K is the
diagonal matrix given by . Substituting this value of into equation (D.22), one
obtains
V„=(x-xf'«’ (D.23)
from which the uncertainties of the best-fit masses can be evaluated.
But what if the variances are not all equal (or if some covariances are not zero), so that
Va is not a simple diagonal matrix? It follows from the definition of the variance-covariance
matrix (equation D.21) that K is a symmetric matrix (i.e., Va = Va) and all symmetric matrices
can be written as the product of a matrix and its transpose, so
(D.24)
The observations (i.e., vector a) are now transformed by multiplying equation (D.12) by (A‘)'^:
{A'ya = {A')-'x,,-{A')-'e
or b = Un-e' (D.25)
where = and tZ = (/4')’’Ar (D.26)
The variance-covariance matrix for b is given by
Fi,=(A‘yvÂ-^ = (a 'Pa ‘AA-' =I {B2I)
So for the transformed observations b, all covariances are zero and all variances are equal. So
equations (D.19) and (D.23) are valid:
D. LEAST - SQUARES ANAL YSIS OE WEIGHING SCHEME 139
M = [v'u]~'Vb = = (x'F;‘x)’'x'r;’a
(D.28)
= (xr/xf (D.29)
Equation (D.28) gives the best-fit mass values M, and equation (D.29) the variance-covariance
matrix of J/no matter what the form of the observational variance-covariance matrix
MiillAifli
“24 computer can make matrix calculation as easy as falling out of a tree ”
Appendix £
Least-squares calibration of weights installed in

balances
This Appendix describes the application of the method of least squares calibration of the
masses installed in a single-pan, two knife-edge balance.
Each decade in the balance generally has four masses to make up the nine, or more,
combinations for the dial settings. To make use of this information in the least-squares
analysis, the appropriate combination for the balance under test must be obtained either by
observation or from the manufacturer. The least-squares method requires the observations
obtained in section 7.3.3.2.2. The calculations are relatively involved but can be performed
using a computer. From the analysis it is possible to calculate the uncertainty for each dial
setting. Because the combinations of masses differ for different balances it is not possible to
give a completely general treatment, but an example can be used to describe the method in
sufficient detail to cover all cases.
Consider a balance that has been calibrated by the least-squares method. Let the
measured values for the dial settings for a particular decade be aj, a2,... ap, and let A, B, C, D
be the mass values for the weights in the balance The nominal values of A, B, C and D are
such that
A:B:C:D = 1: 1:2:5.
These values, and the equations given in (E.l), will be different for various
combinations of the masses in different balances. The mass loading arrangement is such that
nine equations with four unknowns A, B,C and D can be written as follows:
A = a]
A + B = a2
A + C = as
A + B + C = a4
D = as(E.l)
B + D = 0(5
A + B + D = a-j
B + C + D = as
A B + C D = ag
141
142 E. LEAST-SQUARES CALIBRA TION OF MASSES INSTALLED IN BALANCES
This set of equations can be treated like the weighing schemes of Chapter 4 in order to
find the best-fit values of A, B, C and D (see Appendix D for details). We therefore use
equation (D.28).
M = ( X' VI^X a (E.2)
where vector M contains the best fit mass values of A, B, C and D {sqq equation E.4). The
observation vector a contains measurements aj to a9, and the design matrix X is obtained from
the structure of the left-hand side of equation (E.l); see equation (E.4). For the purpose of
solving equation (E.2) for M, the variance-covariance matrix Va can be taken to be a unit
matrix as the repeatabilities of a single-pan two knife-edge balance at different loads are
normally equal because this is a constant load balance. Therefore equation (E.2) can be
rewritten in the form of equation (D.19):
M = (X'Xy^X‘ a (E.3)
The observational matrix a, the design matrix Xfor (E.l) and the solution matrix M are
shown below:
0 0 0
1 0 0
0 1 0
1 1 0
0 0 1 a «5 (BA)
1 0 1
1 0 1 a?
1 1 1
^8
1 1 1
^9
Using equation (D.14)
R = XM-a (B.5)
for the residuals vector and equation (D.15)
S = R‘R (E.6)
for the sum of squares of the residuals S, the value of S is obtained.
The number of degrees of freedom v of (E.l) is the number of equations minus the
number of unknowns (in this case, 4), i.e., v= 9 - 4 = 5.
The observational variance is Sobs^ = — (E.7)

E.l NUMERICAL EXAMPLE 143
The observational variance is a measure of the uncertainty of the calibration of the

masses. From this value it is possible to calculate the uncertainty for each of the dial settings,
but because the masses are not used an equal number of times their uncertainties are not equal.
However as the differences are not large the standard deviation of each dial setting can be
estimated to sufficient accuracy by multiplying 2/3. Here the “2/3” is a factor inserted to
approximate the more rigorous calculation.
2/3 Soi)g (E.8)
To obtain the actual uncertainty Sj must be combined with the uncertainty of the masses
used in the calibration. This is difficult to do rigorously, but a reasonable approximation is to
calculate the standard deviation, S2 of the masses used for each dial combination, and then
combine it with sj using
/ 2 , 2
5 = +52 (E.9)
If S] > 4s 2, then S2 may be ignored. The largest value of s should be chosen as the
uncertainty for the decade. This will almost certainly occur at dial setting ‘9’, so that only one
value need be calculated.
This procedure is repeated for all the decades of the balance.
E.l Numerical example

Consider the calibration of the 100 g decade of a Mettler B5C 1000 single-pan, two knife-edge
balance. The observations, which are a different combination to those given in equation (E.l),
are as follows:
(1) A = 10.000 02
(2) C 19.999 98
(3) A + C = 29.999 9
(4) A + B + C = 40.000 15
(5) D = 50.000 37 (E.10)
(6) A + D = 60.000 46
(7) + C + D = 70.000 6
(8) A + C + D = 80.000 54
(9) A + B + C + D = 90.000 63
The design matrix X and the observation matrix a are shown below:
10.00002
1 0 0 0
0 0 1 0 19.99998
1 0 1 0 29.9999
1 1 1 0 40.00015
0 0 0 1 50.00037 (E.ll)
1 0 0 1 60.00046
0 0 1 1 70.0006
1 0 1 1 80.00054
1 1 1 1
90.00063
A spreadsheet program can be used for the calculation.
Table E.1: Spreadsheet for calculating the solution matrix M and residuals R
X a
1 0 0 0 10.00002
0 0 1 0 19.99998
1 0 1 0 29.9999
1 1 1 0 40.00015
0 0 0 1 50.00037
1 0 0 1 60.00046
0 0 1 1 70.0006
1 0 1 1 80.00054
1 1 1 1 90.00063
M RES S
9.99997211 -4.789E-05 3.81737E-08
10.0001489 4.211E-05
20.0000221 9.421E-05
50.0004937 -6.842E-06
1.237E-04
5.789E-06
-8.421E-05
-5.211E-05
6.842E-06
E.l NUMERICAL EXAMPLE 145
The steps in obtaining the solution matrix M are:

1. Enter the design matrix in a 4x9 array and name the selected cells as X (as described in
Chapter 4, section 4.3)
2. Enter the observation matrix a in a 1x9 matrix and name it a.

3. Highlight a 1x4 column and type:
=MMULT(MMULT(MINVERSE(MMULT(TRANSPOSE(X),X)),TRANSPOSE(X)),a)
then press [Control]+[Shift]+[Enter] and name this vector as M
4. Highlight a 1x9 column and type:
=MMULT(X,M)-a
then press [Control]+[Shift]+[Enter] and name this vector as RES
5. Highlight one cell in the spreadsheet and type:
=MMULT(TRANSPOSE(RES),RES)
then press [Control]+[Shift]+[Enter] and name this cell as S
The solution matrix M has 4 elements that give the masses of A, B, C and D
A = 9.999 972
B= 10.000 149
C = 20.000 022
D = 50.000 494
The values of the dial setting are then found by using these values and substituted in equations
(E.IO). They are shown below:
Dial setting (g) Actual mass value (g)
10 9.999972
20 20.000 022
30 29.999 994
40 40.000 143
50 50.000 494 (E.12)
60 60.000 466
70 70.000 516
80 80.000 488
90 90.000 637
The sum of squares of residuals, S, is given by:
S = 3.82x 10'® g2
The number of degrees of freedom r, in this example is the number of equations minus the
number of unknowns (in this case, 4), i.e., v = 9 - 4 = 5.
Thus, si^ = [S / (9 - 4)]= 2/3 [ 3.82 X 1 O'® / 5]*^
so Si = 0.000 058 g.
The uncertainty of the calibrating masses (2s2 with 30 degrees of freedom) is

Nominal mass Uncertainty
50 g 50 gg
20 g 20 |ig
20* g 20 |Lig
lOg 10|Ltg
For the masses used in the calibration of the dials, the standard deviations are combined
according to equation (5.3), i.e., simple addition. The uncertainty for dial (9), 90 g, is 90 pg.
Therefore, the maximum standard deviation of the dial values is
s =[0.000 058^ + 0.000 0452]’^ = 0.000 073 g
Using the Welch-Satterthwaite formula (L.6), the number of effective degrees of

freedom of the dial values is 12 and the coverage factor k is 2.18. Thus, the expanded
uncertainty of the dial values for the 10 to 100 g decade is
U = 2.18 X 0.000 073 g = 0.000 16 g.

Appendix F
Sample report for the calibration of weights
A weight calibration report should include the description of the weights, the client, the date of
calibration, the identity of the weights, the mass value of each and its uncertainty, the
environment in which the calibration had taken place and the assumptions of calculation. The
following sample report uses the example detailed in sections 4.4.3 and 5.4.2.

Bradfield Road, West Lindfleld, NSW 2070, Australia
Phone: (612) 9413 7000 Fax: (612) 9413 7631
MEASUREMENT REPORT ON
A Set of Weights
Report no: RN 12345
Particulars of weights: 5 stainless steel integral weights 100 g - 10 g
Serial no: 24653
Client: CSIRO-NML, Bradfield Rd., West Lindfield, NSW 2070
Calibration date: 14 December 2001
Date of Issue of report: 20 Dec 2001
Temperature during test: 20.4°C
Examined by: E. Morris
Calibration Result
Nominal value and/ or markings Mass (g) Uncertainty (±) (g)
(g)
100 g 100.000 29 0.000 11
50 g 49.999 72 0.000 104
20g 19.999 89 0.000 044
20* g 20.000 04 0.000 044
lOg 10.000 18 0.000 028
147
148 F. MINIMUM REQUIREMENTS FOR REPORTS
Notes
1. This set of weights has been examined at the National Measurement Laboratory on the basis of
weighings made in air against standards of known mass. The calibration performed was
appropriate for an OIML class Fi set.
2. The set of weights was calibrated according to procedures detailed in The calibration of
weights and balances by E C Morris and K Fen, National Measurement Laboratory,
Lindfield, NSW, Australia. To obtain the results, which are given above, it was assumed that
the densities of the weights lie within the required range for weights of OIML class Fj. The
set of weights was calibrated in air of density 1.2 ± 0.034 kg.m'\
3. The mass value assigned to each weight represents, to within the corresponding uncertainty,
the mass of a hypothetical object of density 8000 kg.m'^ that, in air of density 1.2 kg.m'\
would balance that weight.
4. In use, the mass of each weight should be taken as equal to the value given and buoyancy
corrections, if required, should be computed on the assumption that the density of the weight
is 8000 kg.m’^. If this procedure is followed, any error arising from the effects of air buoyancy
on the weights will not exceed one half of the OIML F, uncertainty, provided the density of
the air in which the weighings are made lies within the range 1.13 to 1.27 kg.m’\
measurand with 95% probability. The coverage factor is equal to 2. The tabulated mass
uncertainty is at the time of test, with no allowance for the contribution due to drift. When
the weights are used it will be necessary to include a contribution to the uncertainty due to
drift.
Dr N Bignell
For Dr B D Inglis
Appendix G
Statistical tests
This Appendix describes two statistical tests that may be used in conjunction with mass and
balance calibrations, and in metrology generally. Fluctuations occur in all measurements, but
one must sometimes decide whether or not an apparently large fluctuation is a chance variation
or the result of an error or deficiency in the measurement system. Statistical theory allows us
to determine the probability that a fluctuation is a chance variation, and the lower this
probability the more concerned we should be that an error or deficiency has arisen. The tables
for t- and F-tests given below are based on a 5% probability, a commonly accepted level.
Hence, chance is accepted as the explanation of a large fluctuation only if the probability of
chance is 5% or greater.
G.l The t-test

The t-test indicates whether or not a new measurement Q' of some quantity Q is statistically
consistent with a reference value Qref, which is normally the average of n previous
measurements. It will be assumed that each of the n measurements and the measurement Q'
involve the same process, for example, each may be the average of five weighings of the same
check weight.
We use the quantity t, defined by:
(G.l)
where Sref is the standard deviation of the n previous measurements of which Qref is the
average. The number of degrees of freedom v for /is n - 1.
The new measurement Q' is consistent with the earlier ones if |/| (the magnitude of t,
i.e., its value without the sign) is less than or equal to the value tâx given in Table G1.
Sometimes tmax is simply taken to be 2, which is acceptable, insofar as it makes the test
more conservative, but it is not recommended for fewer than 15 degrees of freedom. (Note
that the tmax values given in Table G1 are equal to the coverage factors used for calculating
95% confidence limits.)
149
150 G. STATISTICAL TESTS
Table G.l: Maximum values of |Z| for use in the t-test, for various degrees of freedom v.
U-1
^max
12.71
V
7
^max
131
V
13
^max
2.16
V
30
^max
2.04
2 4.30 8 2.31 14 2.15 40 2.02
3 3.18 9 2.26 15 2.13 50 2.01
4 2.78 10 2.23 17 2.11 60 2.00
5 2.57 11 2.20 20 2.09 120 1.98
6 2.45 12 2.18 25 2.06 oo 1.96 1
G.2 An example of the t-test

In a laboratory that frequently calibrates 100 g weights by direct comparison, the mass of a
check weight is regularly measured in the same way as the masses of test weights. On a given
occasion, the mass of the check weight was found to be 99.9993 g. The previous 12
measurements were:
99.9998 100.0003 99.9999 100.0006 100.0001 100.0001

100.0004 99.9996 100.0003 100.0001 100.0003 99.9999
The 99.9993 g measurement looks rather low, so the operator applies the t-test. From the 12
previous measurements one finds that Qref= 100.000117 g and Sref= 0.000286 g with 11
degrees of freedom. Therefore
99.9993-100.000117 -0.000817
(G.2)
0.000286-71.083 0.000286x1.041
From table Gl, one finds that |/|, in this case 2.74, should be less than 2.20, so we have a
strong indication that something has gone wrong. The problem could lie with the balance, the
reference weight, the check weight, the laboratory environment or the operator’s technique.
All possibilities should be investigated.
G.3 The F-test

The F-test indicates whether or not a measurement process has degenerated in some way.
When applied to weighing, it may show that a problem has arisen with the balance, its
environment, the manner of taking readings or anything else that could influence the results.
The quantity F is defined by:

T2 _ new
(G.3)
G.3. THE F-TEST 151
where is the latest measurement of standard deviation for a balance and Sref is a reference
value for that balance, based on earlier measurements of standard deviation. The larger the
degrees of freedom and Vref, the more confidence one can have in the test.
Although this test works for any combination of and Vref, the only situation
considered here is the one in which a laboratory always measures the standard deviation of a
given balance with the same number of degrees of freedom, The reference standard
deviation may be the most recent previous measurement or it may be obtained by averaging n
previous measurements in the following way:
= -X )/ n Vn, (G.4)
1=1
This is called the pooled standard deviation and it has the advantage that vê/can be quite
large.
In practice, there is reason for concern only when Snew is greater than Sref, so one is
interested only in cases where F is greater than one. Provided F is less than or equal to the
value given in Table G2, one can be reasonably confident that the value of s„ew does not
indicate any deterioration in the balance, its environment or operator technique.
G.4 Examples of the F-test

1. The standard deviation of a balance is always measured with 9 degrees of freedom. When
averaged according to equation (G.4), 10 measurements made over the last three years give
Sref — 50 pg. A new measurement then gives 74 pg. Has something gone wrong?
The value of F is 74^50^ = 2.19. This is greater than the limiting value of 1.99 obtained from
table G2 (for = 9 and n = 10), so the test indicates that something has gone wrong.
2. An experienced technician shows a novice how to use a balance and measure its standard
deviation. The technician then measures the standard deviation with four degrees of freedom
and obtains a value of 50 pg. He asks the novice to do the same, whereupon the novice gets
115 pg. The technician tells the novice that he must have done something wrong and needs
more practice, but is this a fair assessment? (Assume that nothing could have gone wrong with
the balance or the laboratory environment in such a short time.)
In this example, Sref is taken as the single measurement made by the technician, so n is
equal to one in table G2. The value of F is 115^/50^ = 5.29. However the limiting value of F
from the table is 6.39, so we can reasonably conclude that the technician’s assertion is not fair.
Based on the above results, there is a good chance that the novice’s technique is satisfactory.
What if the technician were to repeat his measurement and get much the same result as he did
the first time? The technician’s measurements could then be pooled. This would lead to a
value of F that is still in the vicinity of 5.29, but n would now be equal to two, so the limiting
152 G. STATISTICAL TESTS
F value would be 3.84. It would now seem likely that there is something wrong with the
novice’s measurement.
Table G.2: Maximum values of F for use in the F-test. = number of degrees of freedom for
measurements of balance standard deviation, n = Number of pooled measurements used to calculate
the reference standard deviation, Sref.
Vm
n 4 9 19 1
1 6.39 3.18 2.17
2 3.84 2.46 1.87
3 3.26 2.25 1.77
4 3.01 2.15 1.73
5 2.87 2.10 1.70
6 2.78 2.06 1.68
8 2.67 2.01 1.66
10 2.61 1.99 1.64
12 2.57 1.97 1.63
15 2.53 1.95 1.62
20 2.49 1.93 1.61
25 2.46 1.92
30 2.45 1.92
40 2.43 1.91
50 2.42 1.90
Appendix H
Air density measurement
H.l Air density equation

The density of moist air can be determined from measurements of temperature, pressure and
humidity using the equation [12]:
3.4510x10'^/7
Pa (1-O.378xj (H.1)
ZT
where
pa - density of moist air (in kg.m’^)
p - air pressure (in Pa)
T - temperature using ITS 90 (in K)
Xv - mole fraction of water vapour
Z - compressibility factor
The mole faction of water vapour is given by:
(H.2)
IUU77
where
h - relative humidity (expressed as a percentage)
f- enhancement factor
/= 1.00062 + 3.14x10'^/7 + S.bxlO'"^ i (H.3)
t is the temperature in degrees Celsius
Psv - saturation vapour pressure of water (in Pa)
Psv = exp(AT^ + BT C + —) (H.4)
T
where
=1.2378847 x 10'^ K'^
5 = -1.9121316x lO'^K"’
C= 33.93711047
£> = -6.3431645 X lO^K
153
154 H. AIR DENSITY MEASUREMENT
Compressibility factor Z:
Z 1 + a, A + 0-21 + (bg + bjt)x^ + (cq + Cj 2

(H.5)
where
«(,= 1.58123x 10-® KPa’
£Zy =-2.9331 X 10’8 Pa’
a2= 1.1043 X IO’’”K’’ Pa’
= 5.707 X 10-®KPa’
Z>; =-2.051 X 10’8 Pa’
Co= 1.9898 X 10-^KPa’
cy =-2.376 X W® Pa’
d= 1.83 X 10'” K^Pa'^
e =-0.765 X 10’8 K^Pa'^
H.2. Measurements of temperature, humidity and pressure

To obtain the air density with an uncertainty of ±1%, the following instruments are needed.
Temperature:
To obtain an uncertainty of ±1 K, a calibrated mercury in glass thermometer or a calibrated
platinum resistance or thermistor thermometer with moderate accuracy will be adequate.
Pressure:
A calibrated digital or aneroid barometer with an uncertainty of ±500 Pa will generally be
suitable.
Humidity:
A calibrated hiunidity probe or sensor with an uncertainty of ±10% will be adequate for
normal weight calibration.
Appendix I
Mass and density limits for the OIML

classification of weights [3]
Table LI: Maximum permissible errors (the difference between actual mass and nominal mass) for
the OIML weight classes.
Nominal E, El Fl Fl Ml M2 M3
Values *n»g ±mg ±mg ±mg ±mg ±mg ±mg
20 kg 10 30 100 300 1000 3000 10000
10 kg 5.0 15 50 150 500 1500 5000
5 kg 2.5 7.5 25 75 250 750 2500
2 kg 1.0 3.0 10 30 100 300 1000
1 kg 0.50 1.50 5.0 15 50 150 500
500 g 0.25 0.75 2.5 7.5 25 75 250
200 g 0.10 0.30 1.0 3.0 10 30 100
100 g 0.050 0.15 0.50 1.5 5.0 15 50
50 g 0.030 0.100 0.30 1.0 3.0 10 30
20 g 0.025 0.080 0.25 0.80 2.5 8.0 25
lOg 0.020 0.060 0.20 0.60 2.0 6.0 20
5g 0.015 0.050 0.15 0.50 1.5 5.0 15
2g 0.012 0.040 0.12 0.40 1.2 4.0 12
1g 0.010 0.030 0.10 0.30 1.0 3.0 10
500 mg 0.008 0.025 0.080 0.25 0.80 2.5
200 mg 0.006 0.020 0.060 0.20 0.60 2.0
100 mg 0.005 0.015 0.050 0.15 0.50 1.5
50 mg 0.004 0.012 0.040 0.12 0.40
20 mg 0.003 0.010 0.030 0.10 0.30
10 mg 0.002 0.008 0.025 0.080 0.25
5 mg 0.002 0.006 0.020 0.060 0.20
2 mg 0.002 0.006 0.020 0.060 0.20
!J2£. 0.002 0.006 0.020 0.060 0.20
155
156 I. MASS AND DENSITY LIMITS . . .
Table L2; Minimum and maximum density limits (in kg.m'^) for the QIML weight classes Ei to Mi
1 Nominal
Value E, E2 Fl F2 Ml
>100g 7934 .. 8067 7810 .. 8210 7390.. 8730 6400.. 10700 >4400
50 g 7920 .. 8080 7740 .. 8280 7270 .. 8890 6000.. 12000 >4000
20 g 7840 ..8170 7500 .. 8570 6600.. 10100 4800 .. 24000 >2600
10 g 7740.. 8280 7270.. 8890 6000.. 12000 >4000 >2000
5g 7620.. 8420 6900.. 9600 5300.. 16000 >3000 1
2g 7270.. 8890 6000.. 12000 >4000 >2000
1g 6900.. 9600 5300.. 16000 >3000
500 mg 6300.. 10900 >4400 >2200
200 mg 5300.. 16000 >3000
100 mg >4400 >2300
50 mg >3400
20 mg >2300
Appendix J
Buoyancy uncertainty in mass calibration

When measuring the conventional mass of a test weight by substitution weighing, the
buoyancy correction given in equation (J.l) may need to be applied (see section 2.4):
<a/=(p„ -1.21——- \m
(J.l)
\Pt Pr y
Here Pa is the air density at the time of the weighing, is the density of the test weight, /?/{is
the density of the reference weight, and is to be added to the measured conventional mass
M of the test weight in order to obtain the correct conventional mass. If this correction is
applied, the residual uncertainty (given by equation 5.6) will normally be negligible.
In order to calculate values of the three densities appearing in equation (J.l) are
required, but these are often not available. Furthermore the correction is normally small
within 300 m of sea-level. For these reasons, the correction is usually applied only when
high quality weights are calibrated at elevated laboratories. Consequently it is often
necessary to estimate the uncertainty that arises from the failure to make this buoyancy
correction in substitution weighing.
If one has no knowledge of pr or /Tr , then the best available value of is zero (since
Pt is less than pR, or vice versa, with equal probability) and the buoyancy standard
uncertainty Ub resulting from putting = 0 is the standard deviation of the quantity on the
right in equation (J.l), given by
(J.2)
In this equation, R is the root-mean-square value of (Pa - 1-2) and Sz is the standard deviation
of Z, where Z = (p- p~^).
The value of R that can be expected near sea-level probably varies somewhat from one
geographical region to another, but based on measurements of air density made at NML over
a long period, a value of 0.02 kg.m’^ seems a reasonable estimate for an air-conditioned
laboratory in any region, for the purpose of estimating this uncertainty. Combining the
temporal variations in air density with the variation with altitude, the values of R given in
Table J.l are obtained. At altitudes above 100 m, the temporal variations are relatively
157
158 J. BUOYANCY UNCERTAINTY IN MASS CALIBRATION
unimportant, so that the R values given in the table are just the mean values of (1.2 - /?«) to be
expected under laboratory conditions.
The standard deviation sz is obtained as follows:
1.56x10 ®+ 52 (J.3)
where St and Sr are the standard deviations of pr and pR respectively and both pr and pR have
been put equal to 8000 kg.m’^.
Table J.l: The quantity 7? to be used in equation (J.2). The second column gives the mean air density
Pa to be expected under laboratory conditions. For altitudes above 100 m, R is approximately equal to
the mean value of (1.2 - pj
1 Altitude (m) Air density (kg.m’) R (kg.m-’)

Within 100 m of sea
level 1.191 (at sea-level) 0.020
200 1.164 0.037
400 1.137 0.064
600 1.112 0.089
800 1.086 0.114
1000 1.062 0.138
1200 1.037 0.163
1400 1.014 0.186
In order to progress further, it is necessary to make some assumptions about the weights and
their densities:
(i) The weights conform to the OIML recommendation Rill [3]

(ii) Although the permitted density range for weights of class F2 and below is very
wide, in practice the densities of these weights lie within the range 7000-
9000 kg.m’\
(iii) For classes Ei to Fi, the density distribution for weights selected at random is
rectangular between the OIML density limits.
(iv) Weights of any class are calibrated using references of the next higher class, eg
Fl test weights are calibrated using E2 references.
Classes Ei to Fi calibrations
It follows from the third assumption that
Sj. = = 0.289A/77. (J.4)

2V3
J. BUOYANCY UNCERTAINTY IN MASS CALIBRA TION 159
where Apr is the full width of the OIML allowable density range for the test weight. There is
a similar equation for sr . As Sr is nine times smaller than st , it will be ignored in
equation (J.3), so that (from equations J.2, J.3 and J.4),
Apj.RM (J.5)
Classes Fi and M calibrations
As a consequence of the second assumption, w/, will be significant for class F2 calibration
only at very high altitude. It will never be significant for class M calibration.
“ According to my calculations, you shouldfloat noyv... I think... ”

160
Appendix K
A brief outline of balance construction
In this Appendix, two types of balances, electronic balances and single-pan two-knife
edge balances, are discussed.
K.1 Electronic balance

All balances of this type measure the total gravitational force, or weight. Because of their
construction most are top-loading balances. Basically, there are four different types of
weighing principles: electromagnetic force compensation (EFC), strain gauge (SG), oscillating
string and capacitance. In this appendix, only EFC & SG, which are widely used, are
discussed.
Electromagnetic force compensation (EFC) balance
All balances of this type measure the total gravitational force, or weight, rather than compare
forces. Fig. K.l shows the principle of an EFC balance. A coil, rigidly attached to the balance
pan, is placed in the annular gap of a magnet. When a mass is added to the pan a position
sensor detects that the pan has been lowered and causes a current through the coil to be
increased, providing a magnetic counter-force that returns the balance pan to the original
position. The compensating current is measured as a voltage across a resistor R and then is
read out on what is effectively a digital voltmeter. The compensating current is in direct
proportion to the mass on the pan, and hence the actual value of the mass may be obtained.
High resolution and high accuracy can be achieved using this principle of construction.
Strain gauge
A number of large capacity balances, particularly platform scales, have strain-gauge

load cells as the sensing mechanism. Generally, high resolution and high accuracy are not
required. These are the most common form of force transducers. The material used is a
conductor, which changes electrical resistance while undergoing mechanical stress. Four strain
gauges are normally attached to the surface of a highly durable elastic component, which is the
load receptor. The electric signal is detected through a Wheatstone bridge. The resulting
voltage output is amplified and displayed. At present these balances are limited to an accuracy
of about 1 part in 10"^.
161
162
However, because electronic balances can be considered as black boxes the method of
calibration is independent of the system used to detect the mass.
WEIGHING PAN
Figure K.1: Diagram of an electromagnetic force compensated balance
K.2 Single-pan, two knife-edge balances
These instruments can be divided into two categories. In one, the load of the balance is
suspended below the beam and the beam is arrested during loading and unloading of the pan.
Fig. K.2 shows a balance with the load suspended below the beam. In the other type, which is
much less common, the load is supported above the beam by a parallelogram linkage and there
is usually no arresting mechanism. Both types of balance are almost always critically damped.
Most balances have built-in masses attached to the pan assembly so that whenever a load is
placed on the pan an equivalent mass is lifted from the pan, thus ensuring that the reading
remains within the optical range of the balance. This means that the mass to be supported by
the knife edges in the balance is fairly constant, so balances of this type are often referred to as
constant-load balances. This means the sensitivity is unlikely to vary with load. The traditional
optical display is sometimes replaced with an electronic digital display, but this does not affect
the method of testing and use.
Due to their construction it is often very difficult to measure the parallelism of the
knife edges of these balances. The knife edges are usually glued to the beam with no
adjustment provided, the alignment being pre-set by the manufacturer.
J. BUOYANCY UNCERTAINTY IN MASS CALIBRATION 163
Figure K.2: Single-pan, two knife-edge balance
“ When a balance is overloaded, it may fail to operate ”

164
Appendix L
An introduction to uncertainty calculation

L.l Uncertainty calculation
This Appendix gives a brief description of uncertainty calculation for mass & balance
calibration. For a more complete treatment the reader is referred to ISO ‘Guide to the
Expression of Uncertainty in Measurement’ [4] or an NML monograph [5]. As far as possible,
symbols and terminology in this book follow the ISO Guide.
The measurand y is the desired quantity, which may not be measured directly, but may
instead be obtained from the measurements of other parameters, the ‘input quantities’. The
function or equation relating the measurand and the input quantities is the model of the
measurement:
y= Xj, ... Xn}, (L.l)
where Xj is theinput quantity of the n needed to evaluate y.

All measurements are subjected to various kinds of imcertainty and, in general, each
input quantity would be affected by more than one imcertainty source. Thus, there would be
more than n components of uncertainty, say N of them, and each would be characterised by 3
parameters:
(1) U, is the initial estimate of the uncertainty of the component. For instance, Ui may
be an expanded uncertainty from a calibration report, the semi-range of a rectangular
distribution or a standard deviation during the measurements. Each of these will be
distinguished by an appropriate reducing factor (see below).
(2) A reducing factor or divisor ki needed to convert U, to its equivalent standard

deviation, U,/ ki, referred to in the ISO Guide as a standard uncertainty, and
(3) The number of degrees of freedom, v;, a measure of the ‘accuracy’ of CT, (of how well
it was estimated).
The ISO Guide requires that each component of uncertainty be converted to an effect on
the measurand y, expressed as a standard uncertainty, where:
I I
(L.2)
165
166 L. AN INTRODUCTION TO UNCERTAINTY CALCULATION
and Ci is the sensitivity coefficient obtained from (L.2a). This is the partial differentiation of
the model, equation (L.l)
c, =3^ (L.2a)
axJ
Ci is the sensitivity coefficient for the input quantity Xy, the one affected by the component
Ui (which may be only one of its uncertainty components). For example, for the model: y =
11.4 xy + 7.8 X2, where xy has 2 uncertainty components, Ui and U2,, and X2 has 3 uncertainty
components, U3, U4 and U5.
Therefore, ci=C2=---- =11.4 andc2=C4=C5=
dxj dx2
Notice that the sensitivity coefficient represents the sensitivity of the measurand y to a small
change in the relevant input quantity (assuming the others remain unchanged).
For mass calibration, the measurand is the mass value of the test mass. For balance
calibration, the measurand is the correction to the balance reading, and when using the balance
it is the mass value obtained in a mass measurement after the correction has been applied.
As mentioned in [4], there are Type A & Type B evaluations of standard uncertainty.
Type A contributions are those for which m , is a standard deviation calculated from
measurements, using equations (3.7) for mass calibration or (6.2) and (7.2) for balance
calibration. Type B contributions are obtained in some other way: quoted from a report,
estimated from manufacturer’s specifications, estimated from experience, etc. The
Type A/Type B classification is for information only and does not affect equations such
as (L.3), (L.6), (L.7).
The net effect of all components of uncertainty is the combined standard uncertainty
Me- It is calculated by adding the squares of the contributing standard deviations (variances):
(L.3)
1=1
Associated with each m , is a parameter known as the number of degrees of freedom i<.
This quantity is a measure of the relative uncertainty with which m , has been estimated; the
larger v;, the smaller the relative uncertainty. If m , is a standard deviation obtained from
several measurements, v; is simply equal to the number of measurements minus one.
Otherwise, K may be obtainable from a report or some other document, but often it has to be
estimated from experience. If the estimated relative uncertainty of m , is ± R%, then
If an uncertainty component is considered reliable to ± 25%, for instance, this equation gives
From Vi we calculate the number of effective degrees of freedom v^ff associated with Uc.
It is found from the Welch-Satterthwaite formula:
L.l. UNCERTAINTY CALCULATION 167
(L.5)
4
in other words: (L.6)
In a report, a measurand value should be stated together with its expanded uncertainty U.
Currently the accepted practice is to evaluate U for a 95% coverage probability (i.e., ±C/is the
95% confidence interval). U is given by:
(L.7)
where k is the coverage factor and depends on It can be obtained from Table L.l or
calculated using equation (L.8), from reference [5]. (Notice that Table L.l is the same as
Table G.l.)
k = coverage factor corresponding to Veff
- , 2.5 2.3 2.2 3.1
= 1.96 +----- +----- ^ +---- - +---- - (L.8)
I'e#
Table L.l: The coverage factor k for various degrees of freedom v and a coverage probability of 95%
V k V k V k V A
1 12.71 7 2.37 13 2.16 30 2.04
2 4.30 8 2.31 14 2.15 40 2.02
3 3.18 9 2.26 15 2.13 50 2.01
4 2.78 10 2.23 17 2.11 60 2.00
5 2.57 11 2.20 20 2.09 120 1.98
6 2.45 12 2.18 25 2.06 oo 1.96
L.2 Rounding of results

This section gives a brief description of rounding procedure for the final result and its
uncertainty. For more details, the reader can refer to NML monograph [5]. The recommended
rounding procedure is as follows.
• Rounding once only: when the final measurement result is being expressed.
• Express the uncertainty to only 1 significant figure when the most significant digit
(the first non-zero digit from the left) is 5 or more and to 2 significant figures,
otherwise. Example of the former are 0.07 and 0.005 mg, and of the latter, 12 and
0.35 mg.
• Uncertainty values should be rounded up, except where rounding down involves a
change of only a few percent, for example, 7.4 and 7.8 becomes 8 whereas 7.1 is
rounded to 1.
• Next, the measured value is rounded to match that of the uncertainty.
168
Hf?&£N
“ You see Tinny, 1 kg of iron is not always equal to 1 kg offeathers.
168
Bibliography
[1] D. Prowse. The calibration of balances. CSIRO Australia, reprinted 1995.
[2] E. C. Morris. Balances and weighing. CSIRO National Measurement Laboratory

Technical Memorandum no. 83, 1992.
[3] OIML Rill. Weights ofclasses Ej, E2, Fj, F2, Mi, M2, M3. Organisation Internationale de
Metrologie Legale Edition 1994 (E).
[4] ISO Guide to the Expression of Uncertainty in Measurement. International Organization

for Standardization, Geneva, 1993. ISBN 92-67-10188-9.
[5] R. E. Bentley. Uncertainty in Measurement: The ISO guide. Publication TIP Pl337
CSIRO, Sydney, Monogr.l: NML Technology Transfer Series, 2001.
[6] Technical Notes: Worked Examples - Assessment of uncertainties of measurement for

calibration and testing laboratories. National Association of Testing Authorities,
Australia, 1999.
[7] Guide to the measurement of mass and weight. The Institute of Measurement and Control,
National Physical Laboratory, Teddington, 1998.
[8] ASTM E319-85. Standard practice for the evaluation ofsingle-arm mechanical balances.
American Society For Testing and Materials, Philadelphia, 1993.
[9] J. W. Humphries. The calibration of weights built into a balance. Aust J. App. Sci 11-360
1960.
[10] G. A. Bell. The calibration of the weights in balances with automatic weight loading.
NSL Technical paper no. 6,1955.
[11] ISO/IEC International vocabulary of basic and general terms in metrology. Adopted as
Australian Standard AS 3807-1998 ""Vocabulary of basic and general terms in
metrology ”.
169
CSIRO Library Network System
170 3 9009 00541018 4
[12] R. S. Davis. Equation for the determination of the density of moist air (1981/91).
Metrologia 29\ 67-70,1992.
D 2 3 5 6 8 9 10 11 12 13 14 15 16 17 18 19
MORRIS, E C
The calibration of weights and balances /

Edwin C. Morris and Kitty M.K. Fen. —
Lindfield, N.S.W.: National Measurement
Laboratory CSIRO, 2002
170 p.
LOCATION: P 531.752 Mo
Cover: aerial view of NML, Sydney

Of And: The Calibration Weights Balances

Uploaded by

Copyright:

Available Formats

Of And: The Calibration Weights Balances

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Of And: The Calibration Weights Balances

Uploaded by

Copyright:

Available Formats

The Calibration of Weights

Monograph 4; NML Technology Transfer Series

Monograph 4: NML Technology Transfer Series

National Measurement Laboratory CSIRO

National Measurement Laboratory CSIRO, March 2002

Copies of the monographs are available from

© CSIRO Australia, 2002

Cartoons by Christophe Granet: Copyright Hagen Cartoons 2002

2. Air buoyancy and conventional weighing 5

3. Mass calibration by direct comparison

4. Mass calibration using weighing schemes 19

5. Uncertainty estimation for mass calibration 31

6. Calibration of electronic balances 47

7. Calibration of single-pan, two-knife-edge balances 71

8. Uncertainty for balance calibration 95

Definitions and symbols 115

1.1 Outline of the book

Air buoyancy and conventional weighing

2.1 Air buoyancy

Xr=-P.yTg = -p.^ and X,=-p^V,g = -p^^

2.2 Conventional weighing

2.3 Definition of conventional mass

The OIML (International Organisation for Legal Metrology) recommendation covering

2.4 Buoyancy correction for conventional mass

2.5 Measurement of true mass for volume or density determination

• Measurement of the density of a liquid or the volume of a container.

Fortunately, no significant difficulty is raised by this, except for very accurate

Mass calibration by direct comparison

A requirement of substitution weighing is that the test weight or object must be

3.2 Substitution weighing

The first two weighings give

M^ = M + 5(Zi + Zj — ^*1 — ^2)/ 2 (3.3)

3.3 Extended substitution weighing

where n is an even numbe?. The measured mass of T is

3.4 Checking mass measurements

In the absence of significant drift, n could be odd.

If a balance is limited by its resolution, the measured value of may be somewhat

3.5 Measuring balance sensitivity

(i) Place weight A of mass on the pan, rf,

where sens = |r,

3.6 Measuring the repeatability of mass comparison

One can obtain a value of Sm from a set of measurements such as an extended

An E2 reference weight R has a mass of 1000.001 1 g. It is used to calibrate a Fj test weight T

From equation (3.6), the sensitivity reciprocal is given by

M 7. =1000.0011 + (- 0.0003 - 0.0002 - 0.0009 - 0.0010)/2 = 1000.0011 - 0.0012 = 999.9999 g

3.7.2 Extended substitution weighing and a calculation of repeatability of mass

A weight T is calibrated against a reference R having a mass of 199.999 94 g. Extended

rj 199.999 89 r3 199.999 87 rs 199.999 90

From equation (3.5), taking S to be 1,

= 199.999 94 + = 199.999 94 + 0.000 36 = 200.000 30 g

3.7.3 Repeatability of mass comparison (Sm)

0.000 6 0.000 7 0.000 6 0.000 4 0.000 7 0.000 6 0.000 6 0.000 8 0.000 8

-0.000 1 0.000 2 0.000 1 -0.000 2 0.000 0 -0.000 1 0.000 1 -0.000 2 0.000 1

Mass calibration using weighing schemes

In general, schemes have several advantages:

However there are also some potential disadvantages:

4.2 Weighing schemes

A+B C+D+E ~ (4.2)

4.2.1 Closed cycle