Professional Documents
Culture Documents
1.1 Traceability
Mass is unique amongst the base units of the international system of units in being
defined in terms of an artefact. A cylinder of platinum iridium alloy, held at the Bureau
International des Poids et Mesures near Paris, is defined as being exactly one kilogram
in mass. All mass measurements undertaken in the World should be traceable to this
artefact via National Measurement Institutes and accredited laboratories. The United
Kingdom’s national standard of mass, Kilogram 18, is held by the National Physical
Laboratory and is the basis of the entire mass scale in the UK. The NPL participates in
a wide range of international comparisons to ensure that measurements made in the UK
are equivalent to those made elsewhere. In the past there have been some problems
with organisations based in one country not accepting traceability to any NMI other than
their own (a prominent example of this is the American Federal Aviation Authority not
accepting measurements traceable to any source other than the American National
Institute of Standards and Technology - NIST), but this situation has now changed with
the advent of a structured approach to international comparisons that means that
measurements made throughout the world are accepted to be equivalent.
1.2 Weights
The use of appropriate mass standards and their correct treatment at all times is
essential to mass metrology. Weights are divided into classes from high quality
reference standards (Class E1) to those used in industrial settings (Class M3). These
classes were originally specified for legal metrology purposes, but they are now in
common usage throughout mass measurement. The International Organisation for
Legal Metrology (OIML) document OIML R111 specifies the properties of weights of
each class and specifies the tests that are necessary prior to carrying out a mass
calibration. A summary of the more important points is given here.
1.3.1 Construction
Weights should be made of a material that is chemically unreactive, non-magnetic, hard
enough to resist scratching and of a density that meets the OIML R111
recommendations for its class. Austenitic stainless steel is generally used in the
1
construction of Class E1 and E2 weights. Lower accuracy weights may be manufactured
from brass, iron or other suitable materials.
As with the other properties of weights the shape of weights for particular classes is
defined in OIML Recommendation R111.
It is common for weights to be submitted in a dirty condition - finger prints are a common
source of contamination. Dirty weights should be cleaned prior to calibration, but only
after the customer has given permission for this to take place. It is recommended that
weights should be calibrated both before and after cleaning.
1.3.3 Magnetism
There are two magnetic properties that must be measured to characterise weights,
permanent magnetisation and magnetic susceptibility. Magnetic susceptibility is a
measure of whether the weight can become magnetised by being placed in a magnetic
field (magnetism of this sort is transient) while permanent magnetism is a feature of a
weight that cannot be altered. It may be possible to de-Gauss magnetically susceptible
weights using a commercial de-Gausser, but such treatment has no effect on
permanent magnetisation. OIML Recommendation R111 sets out permissible limits for
these two properties for various classes of weights. Tables 1.1a and 1.1b shows these
limits.
Weight Class E1 E2 F1 F2 M1 M2 M3
Maximum Magnetization, µ0M (µT) 3 10 30 100 300 1000 3000
Table 1.1a: Maximum Permanent Magnetisation, µ0M (µT), for OIML Weights
Weight Class E1 E2 F1 F2
Nominal mass ³ 100 g 0.010 0.020 0.07 0.21
Nominal mass < 100 g 0.025 0.075 0.25 0.75
Nominal mass £ 1 g 0.12 0.37 1.2
Table 1.1b: Maximum Susceptibility, c
2
Tables 1.1a and 1.1b are taken from second draft of OIML R111, Feb 2000.
The susceptibility and magnetisation limits are specified such that weights with these
properties could cause an error in a balance reading of more than 10 % of the tolerance
of a particular weight class.
There are several methods available for testing the magnetic properties of weights
which range from the simple to the complex.
1.3.3a Compass
The most simple test is to bring a the weight under test into close proximity with a
compass. If the compass needle is deflected it indicates that the weight is magnetic.
However, this is a purely qualitative measurement and can only be used to indicate that
a weight is magnetic.
3
Figure 1.1: The Davis Susceptometer
4
Standard
Counter- Pivot
weight
Test-piece
Figure 1.2: Permeability Meter
Class E1 E2 F1 F2
Rz (µm) 0.5 1 2 5
Ra (µm) 0.1 0.2 0.4 1
Table 1.2: Maximum permissible surface roughness
5
rubbing), then solvent cleaning should be used as a last resort. In general this will
take the form of wiping the weight with a clean cloth that has been soaked in
solvent. Other more elaborate forms of cleaning include the use of ultrasonic baths
and soxhlet apparatus. After cleaning it is necessary to allow weights to stabilise
before calibrating them. The stabilisation time will vary according to the class of
weight and the extent of cleaning that has been undertaken.
The following handling methods are recommended in order to avoid the problems
mentioned above.
1.5.1 Gloves
Gloves should be worn whenever practicable (sometimes they restrict the operator’s
ability to manipulate tweezers et cetera). Chamois leather is an ideal material for such
gloves as it has good thermal insulation properties (so reducing thermal influences on
the balance during the weighing process) and affords a good grip on large weights when
manipulating them. Wicket keeper’s inner gloves, available from large sports retailers
make ideal gloves for this purpose. It is important that the gloves are cleaned prior to
use using a mild detergent to remove the natural oils from the material.
1.5.2 Tweezers
Tweezers, forceps or other specialist lifting devices should be used whenever
practicable to pick up weights and manipulate them inside the balance. It is important
that these tools have no sharp edges and should, when possible, not have metal to
metal contact with the weights. The greatest problems arise at each end of the mass
scale. It is often necessary to handle fractional weights with metal tweezers due to their
small size. In this case the tweezers should preferably be made of copper or brass
which are relatively soft. It is necessary to manipulate large weights using mechanical
lifting equipment. This type of equipment usually takes the form of a manually operated
fork lift. In this case it is good practice to cover the forks with corrugated cardboard and
acid free tissue paper in order to reduce the potential to mark or contaminate the
weights.
6
When manipulating weights with tweezers, or similar tools, it is good practice to keep a
gloved hand beneath, but not in contact with, the weight being moved. This hopefully
minimises the potential for the weight to fall to the ground.
The use of paper to stand weights on also has the advantage of allowing the operator to
write the identity of each weight close to where it is standing during the calibration. Care
must be taken not to place weights on top of writing and the top sheet of tissue should
be disposed of at the end of the calibration.
The major disadvantage with using boxes for storage is that they contact virtually the
entire surface of the weight so potentially causing contamination problems. It is
important that the interiors of weight boxes are kept clean to prevent a build up of dust
and other particles. A vacuum cleaner with a purpose built fine nozzle is ideal for this.
However, it is preferable for weights to be stored on acid free tissue paper under glass
domes. This is an ideal situation as only the base of the weight is in contact with other
surfaces. Regardless of how the weight is stored it is important to ensure that it is clearly
identified at all times so that it cannot get confused with similar weights.
7
BALANCE ASSESSMENT
Stuart Davidson
National Physical Laboratory
2. BALANCE ASSESSMENT
2.1 Introduction
The need for regular and appropriate assessment of the equipment used in a calibration
laboratory is vital both in providing traceability of results and in providing accurate and
reliable results. This chapter covers the routine assessment of standard balances and
mass comparators used for the calibration of standard weights and for other laboratory
based mass measurement. It is not intended to cover the assessment of specialist
balances or to provide details of the test performed as part of the Type Approval of a
balance.
· the balance assessment reflects the way in which the balance is used in practice
· the assessment takes place in the same conditions as the balance would normally be
used under.
It is essential that both these conditions are met if the results of the balance assessment
are to give an accurate reflection of the performance of the balance and therefore
provide a valid contribution to an uncertainty budget for measurements performed on
the balance.
2.2 Preparation
8
used in a shop floor environment it should not be assessed in a temperature controlled
laboratory environment.
9
types rely on the comparison of two forces (an unknown weight with either an external
or internal weight). Despite the possibility of using these balances as direct reading
devices (applying an unknown weight and taking the balance reading as a measure of
its mass) single pan electronic balances will always perform better when used as
comparators, comparing a standard and an unknown in an ABA or ABBA sequence. As
with all the balance types discussed the method of assessment should reflect the way in
which the balance is used in practice.
2.4 Assessment
10
2.4.1g Sensitivity
Measuring the sensitivity of the balance allows a measured difference in terms of scale
divisions to be converted into a mass difference and is therefore vital to the use of the
balance. A suitable sensitivity weight, which can be easily transferred between the
balance pans without arresting the balance, is required. The value of this weight should
be enough to cause the pointer to move along the scale by between one quarter and
one third of the total scale length. The sensitivity weight should be calibrated against
suitable mass standards. The weight is then swapped between the balance pans (left to
right and then right to left) without arresting the balance. Arresting the balance will
change the effective rest point slightly and compromise the accuracy of the sensitivity
measurement. The average effect of swapping the sensitivity weight ( d ) is calculated
(the two figures should agree to better than 5%) and this value used to calculate the
sensitivity as follows:
2 ´ Ws
S=
d
The sensitivity should be checked at at least four points across the weighing range of
the balance.
11
2.4.2a Visual inspection and mechanical check.
Check knife edges and planes. Makes sure the released balance does not foul. Adjust
the zero of the balance.
2.4.2b Drift
This will give an indication of the balance’s stability and its sensitivity to changes in
environmental conditions, it will also give an indication of how long the balance should
be left to give a stable reading (the effectiveness of the damping). If the drift is linear it
can be eliminated by using suitable symmetrical weighing method (eg ABA weighing).
2.4.2e Hysteresis
This is checked by taking readings of increasing and decreasing load. Significant
difference may indicate mechanical problems within the balance.
12
2.4.2i Repeatability of measurement
This represents a series of actual comparisons using the balance in its normal weighing
mode. Statistical analysis of the repeatability of measurement provides a practical
assessment of the balance performance under normal weighing conditions.
2.4.3a Hysteresis
Hysteresis is checked by taking readings for increasing and decreasing loads.
Significant deviation between the two readings indicates a poorly adjusted or dirty
balance.
The weight is placed at each position on the pan twice and the weighing scheme is
symmetrical with time so any balance drift is calculated out.
13
readings taken before and after the test. In practice zero readings should be taken
between each scale point and the balance readings corrected accordingly (this is
particularly important if the balance shows significant drift in use).
Zero 0.00000
Zero 0.00000
Table 1: Scale linearity for an electronic balance with single point calibration
0.0007
0.0006
Scale error (g)
0.0005
0.0004
0.0003
0.0002
0.0001
0
0 20 40 60 80 100
14
These results show the typical performance of a balance with a single internal
calibration weight at full capacity (100 grams in this case). The maximum scale deviation
occurs at 50 grams half way between the two fixed points (zero and calibration at 100
grams). The small scale error at 100 grams may be due to an error in the internal
calibration weight.
If the balance is to be used as a direct reading device allowance should be made for the
scale errors measured either by making corrections to the balance readings or by
allowing for the (maximum) scale error in the uncertainty budget.
When using the balance as a comparator the absolute scale error is not significant since
it applies equally to both weights. The difference in the slope between the scale error
line and the horizontal is however significant. Graph 1 shows that between 40 and 50
grams the error is smallest since the slope of the line is near horizontal (interestingly,
the maximum scale error occurs at 50 grams). The maximum deviation from horizontal
occurs between 70 and 80 grams which is where the most significant errors will occur
when using the balance as a comparator. In practice the scale error between these two
points is equal to the difference in the two measured scale errors ie;
Thus comparing a 70 gram weight with an 80 gram weight would introduce an error of
0.191 grams. On a more realistic level, calibrating a 0.7 Newton weight (nominal mass »
71.36 grams) against a 70 gram mass standard would introduce an error of;
. ´ (7136
0191 . - 70)
mg
10
which is equivalent to 0.026 mg. It can be seen that such an error may be significant and
should be taken into account when producing an uncertainty budget for work done on the
balance in question. Larger errors may be introduced when comparing components which
are further from nominal (for example comparing a piston weighing 75 grams with a mass
standard of either 70 or 80 grams will introduce an error of 0.091 mg).
The second set of data shows results for a balance which is calibrated at two points in
its range (half and full load).
15
Nominal Applied Balance Scale
load load reading error
(g) (g) (g) (g)
Zero 0.00000
Zero 0.00000
Table 2: Scale linearity for an electronic balance with two point calibration
0.0005
0.0004
0.0003
Scale error (g)
0.0002
0.0001
0
-0.0001 0 20 40 60 80 100
-0.0002
-0.0003
-0.0004
Balance reading (g)
It can be seen a two point calibration has the benefit of reducing the maximum scale
error (in general the maximum error should be about one quarter of that achieved with a
single point calibration). This is useful when the balance is used as a direct reading
16
device. When used as a comparator the deviation in the slope of the scale error line
compared with the linearity line will introduce errors into the comparison. The graph
shows a maximum deviation between 30 and 40 grams. Taking the example of a 0.3 N
weight compared with 30 gram mass standard the error introduced is;
Which gives an error of 0.011 mg. It can be seen that this error is of equivalent magnitude
to that obtained for the 0.3 N weight calibrated on the previous balance (0.37 ppm in both
cases) and thus significant in terms of an error budget. The results show that while a two
point calibration will generally improve a balance’s performance as a direct reading device
it will not make it any more accurate as a comparator. Generally, for this type of balance,
the maximum deviation of slope will occur between 40 and 60 grams where the scale
error graph crosses the X-axis.
Given the above scale error measurements illustrated, more detailed assessment of the
ranges 70 to 80 grams for the first balance and 30 to 40 grams for the second balance
may be undertaken. The results of these extra tests will help to quantify the maximum
potential error due slope error for inclusion in an overall uncertainty budget.
2.4.3d Repeatability
The repeatability of an electronic top pan balance can be assessed by repeated
application of a mass standard. In general ten measurements will give enough data to
analyse. The repeatability should be measured at at least two loads (usually half and full
load) and at any other load where the balance will be normally used. Generally the
following data should be calculated for a repeatability assessment:
17
significantly the re-assessment period may be extended. In any case a balance should be
re-assessed if it is moved, modified or changed in any other way.
18
END USER REQUIREMENTS FOR ASSESSMENT
M J Buckley
Standards Officer and General Manager
South Yorkshire Trading Standards Unit
This section deals with the practical information that end users should obtain when
assessing balances and follows on from the assessment of balances.
2.6.1 General
When assessing balances we need to accumulate information that enables us:
It is necessary to carry out the routine tests referred to in Para 6.2: linearity, eccentricity
and repeatability are essential. In addition we need to check on other things such as the
effect of electrical disturbances or local magnetism. We may also have to assess local
effects such as long term vibration or temperature problems.
The results of these tests may not however be truly indicative of long term performance.
Short term repeatability problems may ‘colour’ the results of our measurements. It is
19
therefore necessary that a log is kept of all balance calibrations. We can also calculate
a long term process standard deviation for each balance.
So we must also carry out short term testing. We may need to carry out some form of
daily or weekly testing. We also need to monitor the short term repeatability of each
weight calibration. Different users may produce different repeatability values, perhaps
because of slight variations in operating techniques, or due to heavy breathing (more
about this later!).
· Check that the balance is complete, clean and undamaged. If the balance
is electronic, has it been connected to the mains supply for sufficient time
(the ideal is to leave it permanently on, but in standby mode, if there is
such a facility, when the balance is not in use)
· Remember that dust or other material may get trapped below the balance
pan. Give it a clean
· Check that ancillary equipment works correctly. If the doors are motorized,
do they function correctly, opening and closing without banging. What
about auto-zero and stability settings? Have they been set for optimum
performance where the balance is located, and have the settings been
altered
· Check for any localized vibration or magnetism that was not previously
present
Now we can carry out an assessment of the balance. Many balances require some
preliminary weighings to condition the balance. Some larger capacity balances work
best if they are preloaded when not in use.
We should first zero the balance then, if this is the normal calibration practice, check
(and set) the calibration interval using the internal or external calibration cycle. On large
capacity balances the internal weight may be proportional to the displayed value – with
the result that any minor error in the value of the internal weight produces a large error
in the displayed value. In this case use an external weight. Remember that the
calibration cycle only works if after loading the internal or external weight, zero is
regained on removing the weight. If the weighing machine has an in-built calibration
cycle, check that it has worked by putting a similar value weight on the load receptor
(with the machine out of the calibration cycle).
20
Now we need to carry out a linearity check. Sometimes this need only be for part or all
of the electronic range. Other balances, particularly used for direct reading, may require
the whole range calibrating, together with the value of any range or dialled-on weights.
At the conclusion of the test recheck zero – to assess hysteresis (if this is a problem a
full linearity test with descending loads may be more appropriate). If you rely on part of
the electronic range to define the difference between the standard and test weight you
will need to assess the uncertainty arising from this procedure and include it in your
uncertainty budget. If you use mechanical balances you will need to perform a linearity
test in a similar way, calibrating the optical or mechanical scale as well as any in-built
weights.
Next we need to perform an eccentricity test – that is to establish if there is any variation
in the indicated value if we place a weight in the middle of the load receptor or in a
position which is off-centre. Where possible, however, we should try and design the
effects of eccentric loading out of the weighing machine, with auxiliary hanging
suspension, or by the use of specially designed load receptors. It is necessary that we
check that the eccentric pan is actually functioning as designed.
Now we need to assess the repeatability of weighing the same weight a number of
times. This test should be conducted using the same technique as when the balance is
used for normal calibration. Even if you do not normally carry out as many as ten
difference weighings when calibrating weights, it is convenient to assess balance
performance by weighing the standard and test weight ten times, using the aba or abba
technique normally used in calibration. When making these repetitive measurements we
may get different standard deviations if we use different operators to perform this
operation, or if the environment is different (using solid rather than adjustable weights
when the air pressure is changing illustrates one example of possible variation in
repeatability). Furthermore, placing the weights used in this test in a different position
from normal may result in different repeatability values – some parts of a balance are
hotter than other parts. Finally, local effects such as the effects of the operator breathing
may influence the local environment and so cause changes in repeatability.
If more than a single measurement is made when calibrating weights (and a single
measurement is not in itself a very safe procedure) then we can evaluate the
measurement series and monitor the standard deviation of this measurement against a
21
process standard deviation. Regular monitoring of performance will provide information
as to the condition of the balance (and ensure by maintenance that a balance does not
suddenly become unusable).
22
WEIGHING TECHNIQUES
Ian Severn
National Physical Laboratory
3. WEIGHING TECHNIQUES
3.1 Introduction
There are many weighing techniques currently employed in mass metrology. This is
indicative of the wide range of processes that rely on weighing and the uncertainty
demands that are required by different industrial sectors and end-users.
As with any form of mass calibration it is essential to have the balance calibrated on a
regular basis. It is recommended that the balance undergoes a full assessment and
calibration, by a suitably accredited body, on a periodic basis which will be influenced by
the application and frequency of use. It is important that the balance scale is tared
before use. If there is an internal balance calibration feature this should be used prior to
making measurements on the balance. A reading with zero load should be taken (z1)
followed by the reading with the load on the pan (r1) and a final zero reading (z2). This
allows the user to compensate for any drift in the instrument. The drift corrected reading
(rd) is given by:
z1 + z 2
rd = r1 -
2
3.3.1 Example
An example of how a table of balance reading corrections may look for a 500 g capacity
instrument is shown in Table 3.1. It would be usual for only one of the fourth or fifth
column to appear in such a table of results, depending on the preference of the
calibration laboratory so the user must be aware of both possibilities. The column
marked correction indicates the correction a user must apply to the balance reading to
23
take account of the errors in the scale linearity while the error indicates the offset from
the correct value that was observed by the calibration laboratory. The calibration
correction has the same magnitude as the correction but has the opposite sign.
If when an object is weighed on this balance the initial and final zero readings are 0.000
and 0.006 g respectively, while loaded the reading is found to be 303.235 g, the actual
corrected value of this object would be 303.230 g - (0.000 + 0.006)/2 g = 303.227 g. If a
reading is between two calibration points it is usual to apply the correction of the point
closest to the indicated reading. In the case of a reading of 242.342 g with zero readings
of 0.000 and 0.002 g the calibration correction to be applied would be -0.006 g giving a
drift corrected result of 242.336 - (0.000 + 0.002)/2 g = 242.335 g.
A more robust method of carrying out measurements of this type is to have two similar
containers, one to fill with the test substance and one to act as a reference. The
weighing should then be carried out as follows:
1. The reference container is placed on the balance pan and the reading noted (ref1).
2. The empty container (to be loaded with the test substance) is placed on the balance
(test1)
3. The container is filled with the test substance and the balance reading taken (test2)
4. The reference container is put back on the pan and the reading taken (ref2).
The weight of substance added to the container (wt) may then be calculated as follows:
24
w t = ( test 2 - test1 ) - (ref2 - ref1 )
This weighing scheme eliminates drift and should allow better weighing uncertainties to
be achieved.
When weight A (the first weight to be place on the balance pan) is compared with weight
B (the second weight used) there will generally be a difference (Dm) between the
readings with each of the weights on the pan. This may be expressed in terms of the
following equation:
A = B + Dm
There are two popular techniques for comparative calibration ABA and ABBA
calibrations.
A1 B1 A2 B2 A3
where Ai and Bi represent the balance reading for the ith application of weights A
and B respectively
The difference between weights A and B is then calculated from the equation:
25
A 1 + A 2 + A 3 B1 + B 2
Dm = -
3 2
å Ai åB i
i =1 i =1
Dm = -
n+1 n
Two quantities, the first differences and the second differences, followed by the average
difference may be calculated as the calibration progresses. This process is shown for an
ABABA weighing.
In some applications the sample standard deviation of the final column is taken to
indicate the standard deviation of the weighing process. This is not statistically correct
as some of the readings contribute to more than one of the average values (so meaning
that they are not truly independent measurements). Hence this type of pseudo-statistical
analysis should be avoided.
It is much better to make several totally independent ABA based comparisons and look
at the statistical agreement between them rather than attempt to make an invalid
statistical analysis on a single set of comparisons. In the case of an E2 calibration made
at NPL four or five ABABA comparisons would be made, the results for each calculated
as shown on the previous page, and the sample standard deviation of all four or five
results evaluated.
26
Example
The calibration of a 10 g F1 weight against two standards:
Repeat
A1 A2 A3 B1 B2
9.999 978 9.999 978 9.999 980 9.999 894 9.999 896
Mean 9.999 979 9.999 895
Difference 0.000 084
Repeat
A1 A2 A3 B1 B2
9.999 980 9.999 980 9.999 982 9.999 910 9.999 912
Mean 9.999 981 9.999 911
Difference 0.000 070
27
3.5.2 ABBA Calibration
The ABBA calibration method is potentially more efficient than the ABA method in terms
of the number of weight applications required to produce a calibration result of a
particular uncertainty. In this case the weights are applied to the balance pan in the
following order:
A1 B1 B2 A2
where An and Bn represent the balance reading for the nth application of weights
A and B respectively.
A 1 + A 2 B1 + B 2
Dm = -
2 2
28
Example
This example illustrates the calibration of a 1 g E2 weight by carrying out two ABBA
comparisons against each of two standards.
A1 A2 B1 B2
-0.024 1 -0.024 0 -0.017 5 -0.017 3
Mean -0.024 0 -0.017 4
Difference -0.006 6
A1 A2 B1 B2
-0.024 0 -0.024 3 -0.017 5 -0.017 7
Mean -0.024 2 -0.017 6
Difference -0.006 6
A1 A2 B1 B2
-0.024 2 -0.023 8 -0.026 2 -0.026 1
Mean -0.024 0 -0.0262
Difference +0.002 2
A1 A2 B1 B2
-0.023 7 -0.024 1 -0.025 7 -0.026 1
Mean -0.023 9 -0.025 9
Difference +0.002 0
29
Cyclic weighing gets its name because the first weight weighed is also weighed last so
that it is possible to account for drift in the measurements. This first weight is normally a
mass standard (S1) that is used to assign values to the test-weights (T1...Tn). A second
standard (S2) is also included in the weighing scheme so that the values assigned using
the first standard may be checked against an independent standard. At its most simple
a cyclic weighing scheme for four test-weights would consist of the following
measurements:
In this type of cyclic weighing it is imperative that all of the measurements are made in
quick succession with an approximately equal time between each. From the above it
may be seen how the value of each weight is calculated from that of the previous one.
The value of S1 is used to assign the value of the first test-piece. Theoretically the
measured differences should sum to zero, but in practice there will be some drift in the
measurements. This drift is eliminated by applying a correction (equal to the sum of the
measured differences divided by the number of measurements) to each of the
measured differences. The second stage of the process involves summing all of the
calculated mass differences (SDmi) and then calculating the correction to be applied to
each difference in order to compensate for drift. In this simple example the sum will be
equal to R7 - R1 (ie the difference between the two readings made on S1). This figure is
then divided by the number of calculated differences (in this case six) to obtain a
correction (Corr) to be applied to each difference. As a test of the correction applied and
the arithmetic used the corrected differences should be summed. If this quantity does
not sum to zero, within the uncertainty of the arithmetic used, there is an arithmetic
problem and the calculations must be repeated. As a final check on the calculations the
value of S1 should be calculated using the last difference. It should agree, to within the
accuracy of the calculations, with its quoted value.
The calculated value of S2 should be compared with its standard value. If the calculated
value of S2 is higher than its certificated value by a quantity ES, then the values
calculated should all be reduced by ES/2. The uncertainty in the weighing scheme is
given by ES/Ö3.
30
Example
A set of four 50 g test-objects to be calibrated against S1 = 50.005 g and S2 = 50.001 g.
From the equations in the right hand column the weights would be assigned the
following values from that of S1.
In this case the assigned value of S2 is 0.002 g higher than its quoted value. Hence the
assigned values of all of the test-objects should be reduced by 0.002/2 g to take
account of this discrepancy. The uncertainty from this weighing scheme is 0.002/Ö3 g =
0.0012 g.
The simple cyclic weighing scheme described above is only suitable for low accuracy
applications as it is extremely sensitive to balance drift and only involves one application
of each test-piece onto the balance pan (this means that there is no check that the
correct weight has been applied).
31
Comparison Measured Corrected Using S1 value Using S2 value
Difference Difference
S1 v T1 d1 d1 - Corr T1= S1-d1+Corr T1= S1*-d1+Corr
T1 v T2 d2 d2 - Corr T2=T1-d2+Corr T2=T1-d2+Corr
T2 v T3 d3 d3 - Corr T3=T2-d3+Corr T3=T2-d3+Corr
T3 v S2 d4 d4 - Corr S2*=T3-d4+Corr S2=T3-d4+Corr
S2 v T4 d5 d5 - Corr T4=S2*-d5+Corr T4=S2-d5+Corr
T4 v T5 d6 d6 - Corr T5=T4-d6+Corr T5=T4-d6+Corr
T5 v T6 d7 d6 - Corr T6=T5-d7+Corr T6=T5-d7+Corr
T6 v S1 d8 d8 - Corr S1=T6-d8+Corr S1*=T6-d8+Corr
Sum d1+ d2+...+ d8 0
Corr sum/8
A standard is compared with the first weight under test. The first test weight is then
compared with the second test weight et cetera until the final weight is compared with
the standard. A second standard is used at the centre of the weighing scheme in the
same manner as one of the test weights.
This is shown in the third column above. If the correction has been calculated and
applied correctly the sum of the corrected differences will be zero (within the precision of
the arithmetic used). If this is not the case the correction must be re-calculated and
applied.
The values for each of the test-pieces may be calculated from that of the first standard
in the manner shown in the fourth column. In this manner the value of test-piece one is
calculated from that of standard 1 and the first corrected mass difference and the value
of the second is calculated from that assigned to test-piece 1 and the second corrected
difference et cetera. It may be seen from this that a value will be calculated for the
second standard that is based on the weighing scheme and the value of the first
standard.
The final value assigned to each test-piece is the mean of the values obtained from S1
and S2. It is only necessary to calculate the value based on S1 (or S2 if the calibrator
prefers) and then to consider the offset between the assigned value of the other
standard (S2*) and its quoted value (S2). If a weight has been assigned a value of Tx
using S1 then its value taking into account both standards is given by:
S2 - S2 *
T = Tx +
2
Alternatively the values of the test-pieces may also be calculated based on the value of
S2 before taking a mean of the two results. In this case the process starts by assigning
T4 a value based on S2 and the corrected difference d5 - Corr. The procedure then
assigns values to all of the weights in turn with S1 being given a value based on that
assigned to T6. This assigned value of S1 is then used to give a value to T1 with the
32
process continuing until S2 is assigned a value from T3. However, this is a much more
laborious method of carrying out the calculation.
The uncertainty in the weighing process may be calculated by looking at the difference
between the assigned value of one of the standards (eg S2*) and its quoted value (S2):
S2 - S2 *
Uw =
3
The denominator reflects the fact that this uncertainty has a rectangular distribution (ie it
is based on two points which represent the extremes of a range).
This is most easily illustrated by considering how values would be assigned to a weight
set using a single standard. In reality the weighing scheme would be extended to involve
one or two other standards. The standard is compared with any weights from the set of
the same nominal value and also with various combinations of weights from the set that
sum to the same nominal value. A check-weight, which is a standard treated in the
same manner as any of the test-weights, is added in each decade of the calibration so
that it is possible to verify the values assigned to the weight set. In the case of a 1 kg to
100 g weight set the following minimal weighing scheme may be used:
1000 =1000S
1000 = 500 + 200 + 200D +100C
1000S = 500 + 200 + 200D +100
500 = 200+ 200D+100
500 = 200+ 200D+100C
200 = 200D
200 = 100 +100C
200D = 100 +100C
100 = 100C
In a simple case such as this it is possible to calculate the values of the test weights
manually, but in a realistic situation when several standards are used and information is
required about the weighing scheme uncertainty it is necessary to undertake a least
squares analysis using a computer.
33
a) it minimises use on (and hence wear on) standards
b) it produces a set of data which provides important statistical information about the
measurements and the day to day performance of the individual balances
c) there is a redundancy of data (ie more measurements than unknowns).
3.8 Make-weights
Make-weights are weights that are added to a load in order to make it approximately
equal in weight to the object it is being compared with. Make-weights are added so that
only a small part of the comparators scale is used during a comparison. For example if a
100 g scale had a 1 % error it would equate to a 0.2 g error if there were a 20 g
difference between the loads under test, but this could be reduced to a 0.01 g error if a
make-weight is used to balance the loads to within 1 g.
A + M w = B + Dm
Example
A 1.000 023 g make-weight is added to weight A when it is compared with standard B.
The calculated difference between the readings indicates that A together with the make-
weight is 0.000 546 g heavier than B. Therefore B is heavier than A by 1.000 023 -
0.000 546 g = 0.999 477 g.
34
In using a two pan balance it is usual for the rest-point (the orientation of the loaded
balance if its swing was allowed to fully decay) to be calculated from a set of three or
more turning points (the extreme points of the beam’s oscillation). If five turning-points
were taken the rest-point would be:
æ æ tp + tp 3 + tp 5 ö æ tp 2 + tp 4 ö ö
rp = ç ç 1 ÷ +ç ÷÷ ¸ 2
èè 3 ø è 2 øø
The most common methods to use in comparing a pair of weights on a two pan balance
are double-double or double substitution weighing.
ær + r r +r ö s
Dm = ç 1 4 - 2 3 ÷ ×
è 2 2 ø 2
where s is the balance sensitivity
A more simplistic variation of this, the double weighing, involves just a simple exchange
of the weights and the calculation of the difference between them. However, such a
measurement regime does not take any account of drift in the balance.
35
pans (for example the right hand pan) while the left hand pan is loaded with a counter-
poise weight that remains in the same position throughout the entire weighing. Care
must be taken to ensure that the counter-poise is heavier than either of the weights
under comparison. An ABBA or ABABA weighing scheme is recommended for this type
of comparison.
A = B + (VA - VB )(r a - 12
. ) + Dm
where ra is the air density in g cm-3
Typically for every 1 mm difference between the heights of the centres of gravity of two
1 kg weights there is approximately a 0.3µg correction to be applied. Therefore if a
10 kg standard with its centre of gravity at height 10 cm is compared with a 10 kg slab
weight with its centre of gravity at 3 cm there will be a 0.21 mg correction to be applied
to the mass difference.
It should also not be necessary to make corrections when a balance is used in a direct
reading mode if it has been calibrated in the location where it is being used. If a balance
has been calibrated at a different location (not best practice) or weights are being used
in a dead-weight tester (for example a pressure balance or a force machine) it may
prove necessary to make a correction for gravity.
36
g = ge (1 + 0.0053024 sin2 f - 0.0000058 sin2 2f ) - 3.088H m s-2
where ge = value of gravity at equator
g = local gravity
f = latitude of location
H = Height of location above mean sea level
This typically has a 20 parts per million error associated with it due to geological
differences.
The British Geological Survey is responsible for the maintenance of a grid of local
gravity reference stations around the UK and is able to provide data relating to a
particular location. Alternatively a useful internet site containing data relating to locations
around the United Kingdom is run by the Bureau Gravimetrique International
(http://bgi.cnes.fr:8110).
37
ESTABLISHING A MASS LABORATORY
M J Buckley
Standards Officer and General Manager
South Yorkshire Trading Standards Unit
4.1 Construction
This section will deal with the basic requirements for establishing a mass laboratory, in
terms of the construction of the facility and the equipment needed for calibrating
weights.
Consideration also needs to be given to the needs of the end user – what are the typical
sizes and classes of weights that need to be calibrated. For lower class weights there
may be as much demand for sandblasting and painting cast iron weights as there is for
calibrating them. You may also need to consider whether an adjustment service is to be
provided.
Whilst there are only a small number of weights in Imperial units still in use in industry,
there are weights which are used for force measurements, or as part of some other
measuring equipment, such as a pressure balance. Many of the newer mass
comparators cannot accommodate these weights because of their physical size or their
non-nominal mass value. Furthermore, for force weights you may need to know about
the difference between mass and conventional mass, and you may need to have a good
working knowledge of the technical units of force, and their relationship with the SI unit
of mass.
Although many of the basic requirements apply to all types of mass laboratory, it is true
that the conditions for higher class weights are more onerous – the costs are also
greater. If the basic facilities are right, however, you should be able to make higher
precision weighings in the same facilities at a later date if you need to.
4.1.1 Location
The choice of location of a laboratory is important. A mass laboratory should not be
located adjacent to railway lines, major roads, heavy industry or canals. All of these
locations are unsuitable due to possible localised vibration. If a site is chosen where
vibration is present, no amount of pneumatic or hydraulic isolators will enable good
38
quality mass measurements to be made. Likewise overhead electric cables, substations
and radio transmitters should be avoided.
If you are likely to need to calibrate a significant number of large weights (over 50 kg)
you will have to consider access – but large goods vehicles near the calibration
laboratory, or overhead cranes installed in an adjacent area can also cause problems.
For laboratories undertaking high precision weighings, it is often recommended that part
of the laboratory should be below ground level – whilst this could give better
temperature stability, it is nowadays possible to build above ground facilities which have
similar levels of isolation from environmental effects. What is not recommended is that
the laboratory should be on the upper floors of a building (although some mass
calibration laboratories have been built like this). Mention has been made of cranes for
unloading – it is also recommended that the laboratory is not near a person or goods lift,
used as a throughway to some other room, used for other purposes (eg equipment
storage) or be adjacent to heavy traffic areas such as canteens.
4.1.2 Construction
The construction of a mass calibration laboratory should ensure that stable conditions
can be maintained without compromising access for weights trolleys or equipment.
Small test rooms can lead to temperature gains if too many people and weighing
machines are located in a small area. If possible allow a minimum of 10 m² for each
person.
39
It is worth installing a secondary ceiling, using a grid system and non- fibrous wipeable
tiles. The gap between the primary and secondary ceiling can be used for service
cables. If air conditioning is installed, conditioned air can enter the gap between ceilings,
and then be allowed to diffuse through perforated tiles fitted with removable filter
materials (the filter material should be dust free). This will allow better thermal stability in
the calibration area and ensure that pressure effects caused by the air conditioner do
not affect the balances.
There are problems in choosing a good floor material. It must be easy to clean and yet
not subject to dust formation. Sealed vinyl sheet such as used in operating theatres and
clean rooms is best. Although it will require cleaning this should be by damp mopping
rather than using large quantities of water or cleaning fluids which will change the
humidity levels in the laboratory. Chemical cleaning fluids may also affect the stability of
your standard weights.
4.1.3 Services
When constructing a laboratory you will need to consider the provision and location of
services. At all costs avoid water or heating pipes from being located in the laboratory or
immediate vicinity of a calibration laboratory, as they may affect the temperature of the
test area.
There should be sufficient power points for the balances, environmental measuring
equipment and computers. Ideally, a stabilised supply with battery back-up should be
provided to ensure a continuous stabilised power supply to the mass comparators. By
stabilising the power supply, voltage fluctuations and lightning strikes will not affect the
balances used for calibration. Furthermore, any power outages will be contained, with
the balances being kept thermally stable by being fed from the battery back-up system.
When calibrating weights it is important to record the balance readings accurately. This
is best undertaken by interfacing the balance to a computer so that data can be
transmitted automatically, and manual transcription errors can be avoided. However,
this may not be possible at first. Nonetheless, it may be expedient to install the trunking
and sockets (even if blanking plates are used at first) when establishing a laboratory, so
that building alterations do not have to be made if a data recording system is installed at
a later date.
40
4.1.4 Balance plinths and other furniture
Balances need installing on suitable tables with (ideally) only one balance on each
table. Do not use ordinary tables, or anti-vibration tables designed for other types of
measurements. Plinths should be strong, not transmit vibration from the floor or
surroundings and not exert any influence on the balance through movement, flexing or
magnetism (through metal reinforcements).
The best and most economic solution is probably the construction of solid brick piers (in-
filled with concrete) supporting a cross table of granite. To avoid dust generation the
bricks should be sealed with a non-flaking paint. Polished granite is suitable as a
balance top (100 mm to 200 mm thick depending on the size of balance). Do not use
reconstituted marble or stone as the dry atmosphere of the calibration area will cause
the glued stone to disintegrate.
When installing the brick support piers please ensure that they are not mounted onto
the laboratory floor, otherwise vibrations will be transmitted to the balance when
operators walk across the floor.
Apart from balance tables you will need to provide cupboards which are dust proof for
housing the standard weights (they should be stored under glass rather than contained
in their transportation boxes - this will avoid instability due to contamination from the
materials from which the box is made).
Depending on the overall temperature control required it may be necessary to install air
conditioning. Although measurements are referenced to a temperature of 20 °C it is not
necessary to make measurements precisely at this temperature, particularly if making
lower grade measurements or calibrating stainless steel test weights using stainless
steel standards. However it is very important that the temperature remains stable for
long periods, with little or no change in temperature.
Annex C to the draft revised OIML Recommendation R 111 specifies that temperatures
in mass calibration laboratories calibrating higher class weights should generally be
within the limits 18 °C to 27 ° C, and 40 % rh to 60 % rh. (Lower limits might apply when
certain mass comparators are installed which have reduced operating specifications).
The rate of change of temperature (per 12 hours) and humidity (per 4 hours) should not
change by more than:
41
Class E1 Class E2 Class F1 Class F2 Class M1
0.3 °C / 12 hours 0.7 °C / 12 hours 1.5 °C / 12 hours 2.0 °C / 12 hours 3.0 °C / 12 hours
If air conditioning is installed it must be arranged in such a way that the conditioned air
does not cause a pressure effect on the load receptor. Furthermore there needs to be a
uniform temperature throughout the test area. This can often be achieved through
blowing the conditioned air into the partition between the primary and secondary ceiling,
allowing it to diffuse through perforated ceiling panels fitted with a dustless filter
material.
4.2 Equipment
In the range 20 kg to 1 mg you need to decide whether you will only be calibrating
weights that comply with R 111. If you intend to calibrate other weights you should
remember that they may be of varying physical forms and dimensions, and may be of
non-nominal values.
When providing balances or mass comparators, you should bear in mind the uncertainty
that you want to achieve in calibrating weights, and the practicality of calibrating the
number of weights that are likely to be submitted to you for testing.
If you intend to calibrate higher accuracy weights you may need to have a larger number
of balances or comparators, with extended resolution. If you want to calibrate a large
42
number of weights of all classes then you may need to provide many balances, some
with the same resolution and some of the same capacity but with a different resolution,
size of load receptor or other operating characteristics. For work of the highest
accuracy, such as the calibration of Class E1 weights, some comparators will be
provided with weight handlers or robots to enable weights to be interchanged
automatically, data being collected by computer, with the whole calibration being
undertaken without a human observer. This leads to greater efficiency, enables more
measurements to be made, and allows for better environmental conditions during
measurement by carrying out measurements in a ‘closed’ environment.
Most modern machines used for calibration are now electronic, and do not use a
mechanical beam or lever system with knife-edges and bearings. They have the benefit
of a digital display and can be interfaced to computers for data capture. For low
accuracy weights, electronic balances of general design may be quite satisfactory, and
are relatively inexpensive. For higher accuracy you may need to use a mass
comparator, a high accuracy electronic balance. Unlike the normal balance they may
have higher resolution, but may have several discontinuous weighing ranges or
windows. They may also require to be installed in more environmentally stable
conditions.
4.2.2 Weights
Most standard weights are cylindrical in shape with a knob and a recessed base.
Weights of nominal value less than 1 g are of polygonal wire or sheet metal. For the
highest precision weights are solid and of one-piece. However lower grade weights may
be made with a screw knob which protects an adjusting cavity (this facilitates adjustment
of the weights to compensate for wear, but means that the weights are less stable than
solid weights and cannot be used for the highest precision calibrations). Small weights
should preferably be made of wire rather than sheet, as they have better stability. If
sheet metal weights are used they must be kept very clean as more of the weight is in
contact with the load receptor than is the case with wire weights.
Generally, sets of weights come in the series 5, 2, 2, 1, although older sets may have a
different sequence or contain values which are no longer in normal use, such as 5, 3, 2,
1. Sets of weights are supplied in wooden transit boxes, but they should be stored in the
laboratory out of their boxes under glass.
43
Virtually all weights are made in compliance with the OIML International
Recommendation R 111, which prescribes constructional requirements and specifies
materials, tolerances, surface conditions, density and markings. The highest precision
weights, Classes E1 and E2 are solid stainless steel, with a prescribed density. They
have no weight markings, as the marking would itself attract dirt. Lower grade weights
are usually adjustable, and may be of other materials including chrome-plated brass,
brass or painted cast iron. Except for the lowest grade of weights, stainless steel is the
preferred material, because of its stability, ability to be polished during manufacture and
for its density. Smaller weights may be of stainless steel, german or nickel silver, or
aluminium. Because of its density, aluminium should only be used for weights up to 10
mg.
When calibrating weights over 50 kg it is normal to calibrate the larger standards using a
series of weights of the same value, which can be calibrated on an automatic mass
comparator and then stacked to form a weight equivalent to the larger standard. Whilst
this facilitates quite low uncertainties, it is a very expensive process requiring
considerable investment in stacking weights and also special mass comparators.
For most calibration work, the standard weights will be at least one class better than the
weights to be calibrated. However, high accuracy standards require more care in
storage, handling and use if they are not to change significantly between calibrations.
As a general rule, the value of a standard weight should not change by more than 50%
of the calibration uncertainty between calibrations. Ideally, this rate of change should be
much less than this. Except in special circumstances laboratories should not buy Class
E1 weights for normal use as it is very difficult to maintain them.
To determine air density you can use either the BIPM formula or, for lower grade
measurements, the simplified NIST formula. For the highest accuracy it is also possible
44
to measure air density using special weights which have different volumes. Other
experimental ways of measuring air density are also being researched.
In general, for low grade F2 and M1 measurements we only need to ensure that the
temperature remains within the limits specified. For higher accuracy measurements we
may also need to measure air pressure and humidity. We also need to monitor the rate
of change. This is best measured using a recording thermograph (not a recording
thermometer which may only measure the temperature once every few minutes). In an
automated laboratory an environmental measuring and recording system can be
installed with all environmental data being obtained automatically, twenty four hours a
day.
When measuring air pressure there are two convenient devices, mercury barometers
being generally unsuitable as they are difficult to use accurately and difficult to calibrate.
In previous times the precision aneroid barometer was used – these can often be
bought second hand from specialist repairers. The most practical instrument is now the
electronic pressure transducer (battery or mains operated). These transducers can be
bought in a barometric pressure option. They will need fairly frequent calibration at first
to ensure that they are stable. They have the advantage of being very easy to use.
45
For density, there are probably three practical ways of measurement. For laboratories
working at the highest levels an automated density comparator is probably the best.
Here a mass comparator with a handler device for placing the standard volumes and the
test weights in a controlled temperature bath of FC40 liquid allows the unknown weight
to be calibrated against a standard with a known density and volume. The flourinert
liquid FC40 is used in preference to water as it is less susceptible to the formation of air
bubbles - the downside of this is that the density of the FC40 must be measured
regularly.
Another system for making density measurements involves hanging a pan below a
balance (a hydrostatic balance) and immersing the test weight in distilled water. This is
a fairly common technique.
Finally, you could use a pyknometer to immerse the weight in water and weigh it on a
conventional balance or mass comparator. These pyknometers are now commercially
available - they are generally best suited for large weights of nominal mass 5 kg or
more. With some experience they can be used to determine whether the density of the
test weight falls within the limits detailed in R 111.
The magnetic susceptibility of weights is also specified in R 111. There are a number of
ways of measuring this, including the use of a mass comparator and a magnetic
standard, a test method which was developed at BIPM. The most practical methods are
either with the attracting beam balance (which is commercially available in the United
Kingdom) or by an electronic measuring device which can use standard reference
samples supplied and calibrated by NPL.
46
AIR DENSITY MEASUREMENT AND BUOYANCY CORRECTION
Stuart Davidson
National Physical Laboratory
5.1 INTRODUCTION
The measurement of air density is necessary in the field of mass measurement to allow
buoyancy corrections to be made when comparing weights of different volume in air. It is
particularly important when comparing weights of different materials or when making mass
measurements to the highest accuracy.
5.2 DEFINITIONS
A conventional mass value for an artefact can be calculated from a true mass value
using the following equation:
æ æ 1 1ö ö
. ´ 10 -3 ÷
M c = M t ç1 + ç - ÷12
è è 8 rø ø
47
5.2.3 Buoyancy correction
This is a correction made when comparing the mass of artefacts of different volumes. It is
equal to the difference in the volumes of the artefacts multiplied by the density of the
medium in which they are compared (usually air). When comparing true mass values the
buoyancy correction to be applied between two artefacts can be given by the following
equation:
BC = (V1 - V2 ) ´ rair
M t 1 = M t 2 + BC
. ´ 10 -3 )
BC = (V1 - V2 ) ´ ( rair - 12
The buoyancy correction is applied with the same convention as for the true mass
correction, ie:
M c1 = M c 2 + BC
The equation for conventional mass buoyancy correction is more complicated than for
true mass and care must be taken with the sign of the correction. If weight 1 has a greater
volume than weight 2 and is compared in air of density greater than 1.2 kg/m3 the
correction (BC) will be positive.
48
Material compared with Buoyancy correction
Stainless steel (ppm)
Platinum Iridium 94
Tungsten 88
Brass 8
Stainless Steel 7.5*
Cast Iron 24
Aluminium 294
Silicon 365
Water 875
*This is the result of comparing two types of stainless steel, with densities 7.8 and 8.2
g/cm3
The table shows that even when comparing weights of nominally the same material (such
as stainless steel) attention must be paid to buoyancy effects when the best uncertainty is
required. When comparing weights of dissimilar materials the effect of air buoyancy
becomes more significant and must be applied even for routine calibrations when true
mass values are being measured.
When working on a conventional mass basis the buoyancy corrections become smaller.
The OIML recommendations R 33 use a range for air density of 1.1 to 1.3 kg/m3 (ie.
approximately ± 10% of standard air density) meaning the corrections are about one tenth
of the true mass corrections. This together with the limits specified by OIML R 33 for the
density of weights of Classes E1 to M3 mean that the maximum correction for any weight
is one quarter of its tolerance. This is generally not significant for weights of Class F1 and
below (although allowance should be made for the uncertainty contribution of the un-
applied correction) but for Class E1 and E2 weights buoyancy corrections need to be
applied to achieve the uncertainty values required for weights of these Classes.
49
Routine Measurement Best Capability
Uncertainty ppm Uncertainty ppm
Temperature (oC) 0.1 360 0.01 36
Pressure (mbar) 0.5 500 0.05 50
Humidity (% RH/oC dew pt.) 5% 350 0.25oC 58
CO2 content (ppm) - - 50 21
CIPM Equation 100 100
Total (x 10-3 kg/m3) 0.86 720 0.16 133
Table 2: Routine and Best Achievable Realisation of Air Density using the
CIPM Formula
50
UNCERTAINTY IN MASS CALIBRATION
UNCERTAINTY IN MASS CALIBRATION
Pauline Leggat
National Physical Laboratory
6.1 Introduction
Although perhaps an obvious point, before starting it is worth confirming precisely what
the measurements are aimed at determining. In this example it is the conventional mass
of a weight and the following need to be considered:
· Which measurements and calculations will be required to enable you to establish the
mass value and the uncertainty in its determination? For example, will you need to
determine air density?
· How many measurements do you need to take? (The more measurements you take
the more representative the mean (average) value becomes, although there is a
reduction in benefit as the number of measurements increase beyond a certain
point. Ten measurements is a common choice and statistically valid but not always
practical – eg for economic reasons. One way of dealing with this problem is to use
some data from previous measurements to determine the performance of the
balance and then take a smaller number of measurements for the particular
calibration in hand. This is dealt with in paragraph 6.3.3 under repeatability.)
· How to take your measurements and calculate a mass value.
Once you have determined the mass value you are ready to start calculating the
uncertainty in the measurement. The following table is a typical layout for an uncertainty
budget. It can be in the form of a computer spreadsheet – to make repeated
51
calculations easier - or it can be a paper table completed by hand using a calculator.
Each column in the table is dealt with separately below.
6.3.1 Symbol
The symbol used in to denote the input quantity or the influence factor.
The sources of the uncertainty are dependent on the measurement process and
equation used - as defined in your procedures and by your laboratory environment.
Here we consider the most common sources.
Each of the input quantities in the measurement equation used to calculate the mass
value has an uncertainty. For example if the equation you are using is:
Wx = Ws + DW + Ab
then there is an uncertainty associated with each of the input quantities Ws, DW and Ab. The
uncertainty in DW depends on uncertainty due to other influence factors:
52
Another influence factor to be considered is
6.3.3 Values
The values associated with the sources of uncertainty are either measured, calculated
or come from a priori (previous) knowledge.
In our example:
53
Ds Uncorrected The uncertainty quoted on the mass standard’s calibration
drift of the certificate will not include any contribution for drift in its
standard mass value. The evaluation of this effect is normally the
responsibility of the weight’s owner as they are best placed
to evaluate how much its mass changes between
calibrations. Drift is usually determined by considering how
much a particular artefact has changed its mass value over
a recent period and extrapolating the figure to cover the
period up to its next calibration. In this example no previous
calibration knowledge is assumed and the uncertainty in the
current mass value calibration is also used to estimate the
limits of drift.
dId Digital rounding Each reading is subject to a rounding error. It is taken to be
error ± half the resolution of the comparator. Such errors occur in
the comparator reading of the standard mass and the
unknown mass.
WR Repeatability This is an uncertainty component which is a measure of the
‘spread’ of the repeated readings. It is estimated by
determining the experimental standard deviation of the
mean (see Further reading). In this example a previous
evaluation of repeatability of the measurement process
(from ten comparisons between a mass standard and an
unknown mass) were used to establish a standard deviation
of 0.00017 mg which was then divided by Ön, where n is the
number of readings in the current measurement – in this
case three).
54
6.3.4a Normal distribution
Normal distribution
This represents a group of measurements
where the values are more likely to fall 0.45
0.4
closer to the mean value than further 0.35
away from it. Repeated measurements 0.3
0.1
deviation of ±1. 0.05
-4 -2 0 2 4
Rectangular distribution
This represents a group of
measurements where the values are
0.12 evenly spread between two limits and
never fall outside these limits. An
0.1
example is when using an assumed air
0.08 buoyancy correction (as opposed to a
0.06 measured or calculated value). The
graph shows a rectangular distribution,
0.04
again with a mean value of zero, but
0.02 limits of ±5. In this case one standard
0 deviation is ±2.89.
-6 -4 -2 0 2 4 6
0.05
0
-6 -4 -2 0 2 4 6
55
6.3.5 Divisor
In order to eventually sum all the individual input quantities they must be quoted with the
same confidence level. This is done by establishing a number by which the input
quantity uncertainty value is divided to convert it to one standard deviation and is
dependent on the distribution as shown in the table below.
Distribution Divisor
Normal 1 or 2
Rectangular Ö3
Triangular Ö6
The normal distribution has a divisor of either 1 or 2 depending on the confidence level
of the value quoted. For example on a certificate of calibration the uncertainty might be
quoted as ‘k = 1’ (~68%) or ‘k = 2’ (~95%) in which case the divisor is the ‘k’ number.
Other values are sometimes used, for example k = 3.
This is a multiplication factor which converts the uncertainty in the value of an input
quantity to a corresponding uncertainty in the output quantity (it sometimes has to
convert both quantity – such as temperature or pressure - to mass and also the right
units). In the example being discussed all the input quantities are already expressed in
the quantity mass and using the sub-unit milligram so the sensitivity coefficient is 1.
In order to add all the components together we need them in the same units and the
values for this column are simply calculated from
The number of degrees of freedom is “…in general the number of terms in a sum minus
the number of constraints on the terms of the sum”[1]. Before considering this further it
is necessary to first appreciate that measurement uncertainties are considered to fall
into one of two categories, known as Type A and Type B.
56
· Type A uncertainties are those that are evaluated by statistical methods. For
example, uncertainty due to less-than-perfect repeatability of a measurement can
be reduced by calculating a mean value from several measurements.
The degrees of freedom, vi, for individual uncertainty contributions are given by:
The resulting probability distribution will be a normal distribution unless one rectangular
distribution is much larger than the other components.
In general the effective degrees of freedom, veff, will not need to be calculated if the type
A uncertainty is less than half of the combined standard uncertainty, there is only one
type A component and at least three measurements have been taken. Otherwise the
effective degrees of freedom will have to be calculated to ensure that the k-factor of 2
will indeed give a confidence level of ~95%.
57
The effective degrees of freedom for the combined standard uncertainty will depend on
the magnitude of the degrees of freedom for the type A contributions in relation to the
type B. If the type B uncertainties are all taken to have infinite degrees of freedom the
relationship is shown using the simplified Welsh-Satterwaite equation
u c4 (y )
v eff =
æ u i4 (y ) ö
çç ÷÷
v
è i ø
0.0454 4
veff =
æ 0.00014 ö
çç ÷
è 9 ÷ø
veff = 3.8E+11
veff in this example is a very large number which can be taken to be infinity. If this value
had been less than 100 a k-factor would have been calculated from a distribution other
than a normal distribution. More information about this can be found in the further
reading list but for our example the k-factor is two.
The expanded uncertainty U(Wx) is the combined standard uncertainty u(Wx) multiplied
by a k-factor which will give an uncertainty value with a confidence level of
approximately 95%, in this case 2.
58
Symbol Source of Value Probability Divisor Sensitivity Standard vi or
uncertainty ±mg distribution coefficient uncertainty veff
± mg
Ws Calibration of 0.0500 Normal 2 1 0.0250 ¥
standard
weight
dC Comparator 0.0200 Rectangular Ö3 1 0.0115 ¥
linearity
Ab Air buoyancy 0.0216 Normal 1 1 0.0216 ¥
Ds Uncorrected 0.0500 Rectangular Ö3 1 0.0289 ¥
drift of the
standard
dId Digital 0.0050 Triangular Ö6 1 0.0020 ¥
rounding
error
WR Repeatability 0.0001 Normal 1 1 0.0001 9
u(Wx) Combined Normal 0.0454 >500
standard
uncertainty
U(Wx) Expanded Normal k=2 0.0908 >500
uncertainty
In order to calculate the uncertainty in the air buoyancy correction for entry into the main
uncertainty budget an additional uncertainty budget has to be completed. Air buoyancy
(Ab) is dependent on the volumes of the weights and the air density with the following
relationship:
where (Vs - Vx) is the difference in volume between the standard and the unknown
weight
(ra - 1.2) is the difference between measured density of the air and the standard
air density
There are two ways to calculate an uncertainty value for Ab, either working in relative
values or calculating the sensitivity coefficient directly. In our example the volumes have
not been measured so the value of (Vs - Vx) is taken to be the largest difference
possible according to the OIML recommendations [4] when comparing E2 and F1
weights – that is 1.3 cm3 with an uncertainty of ±1.3 cm3. This uncertainty is treated as a
rectangular distribution because the real value may lie anywhere between these limits
and thus the standard uncertainty (u(V)) is equal to ±(1.3 ¸ Ö3). In this example the air
59
density, ra, has been measured as being 1.22 kg/m3 with an uncertainty of ±10% - thus
the uncertainty in (ra – 1.2), that is u(r), is ±0.012 kg/m3. Thus:
2 2
u ( Ab) æ u (V ) ö æ u ( r ) ö
= çç ÷÷ + çç ÷÷
Ab è (Vx - Vs ) ø è a ( r - 1.2) ø
2
æ 1.3 ¸ 3 ö æ 0.012 ö 2
u ( Ab) = Ab çç ÷ +ç
÷ ÷
è 1.3 ø è 1.22 - 1.2 ø
u ( Ab) = 0.0216
This method of calculation works well for an equation where the only operators are
multiplication or division.
The other method, partial differentiation, sounds more complicated but is actually quite
straightforward for this type of equation.
Using the same equation, Ab = (Vs - Vx)(ra - 1.2), and the same values as above, we
calculate the sensitivity coefficients - the numbers by which values of u(V), expressed
here in cm3, and u(r), expressed here in kg/m3, should be multiplied to calculate their
effect on the output quantity expressed in grams.
The partial derivative of a simple equation, such as the one we are looking at, is simply
the multiplier for the term for which we wish to calculate the partial derivative. For
example, if our equation is A = B ´ C then the partial derivative of B is C and the partial
derivative C is B.
In our air density problem the partial derivative of (Vs - Vx) is (ra - 1.2) and the partial
derivative of (ra - 1.2) is (Vs - Vx). In the correct terminology this is expressed as :
60
¶ (Ab)
= ( r a - 1.2) = 0.02
¶ (Vs - Vx )
¶ (Ab)
= (Vs - Vx ) = 1.3
¶ ( r a - 1 .2 )
The values 0.02 and 1.3 are entered directly into the sensitivity coefficient column of the
uncertainty budget and the remaining calculations are the same as for the main
uncertainty budget.
Partial differentiation is more difficult when the equation is more complex [1].
100.000 71 g ±0.10 mg
and would be accompanied by a statement explaining how the uncertainty and its
confidence level are calculated such as:
The uncertainty has been rounded to 0.10 mg; uncertainties should always be rounded
up rather than down to ensure that the value remains within the 95% confidence limit.
61
6.6 Further reading
[1] BIPM, IEC, IFCC, ISO, IUPAC, IUPAP, OIML. Guide to the Expression of
Uncertainty in Measurement, International Organisation for Standardisation,
Geneva. ISBN 92-67-10188-9
[3] NPL Best Practice Guide No 11. A Beginner’s Guide to the Uncertainty of
Measurement. (included in the accompanying documentation)
62