Beruflich Dokumente
Kultur Dokumente
© Mettler Toledo, A. Reichmuth Uncertainty & Specs 1.1 Prtd.: October 27, 2000
—1—
© Mettler Toledo, A. Reichmuth Uncertainty & Specs 1.1 Prtd.: October 27, 2000
—2—
© Mettler Toledo, A. Reichmuth Uncertainty & Specs 1.1 Prtd.: October 27, 2000
—3—
© Mettler Toledo, A. Reichmuth Uncertainty & Specs 1.1 Prtd.: October 27, 2000
—4—
3) ”SPC” stands here, and in all following instances, for the value of a
property's specification as given in the data sheet.
4) i) If there is no tare weight (such as a beaker, boat, or other
container), the reading is taken with empty pan instead.
ii) If the balance is re-zeroed at any load, then the first reading is
zero, by definition. Consequently, the reading with the sample is
then equal to the second reading (of the tare and sample weight),
and the difference need not be calculated by the operator, as the
balance ”took” the first reading and has already subtracted it from
the second reading.
© Mettler Toledo, A. Reichmuth Uncertainty & Specs 1.1 Prtd.: October 27, 2000
—5—
n
x = n1 Σ
i =1
xi is the mean of this measurement series (or
weighings, i.e., differences of pairs of readings);
n is the number of measurements (or weighings, i.e.,
differences of pairs of readings) 5).
Non-Linearity
The non-linearity of a balance can be read from the data
sheet, too. This specification describes the largest deviation
between the actual and the ideal, i.e., linear characteristic
curve 6)
SPC NL max yNL .
As the characteristic curve of an individual balance, although
a systematic deviation, is generally unknown to the user, we
have to treat the actual deviation for any given load as a ran-
dom contribution.
Because the non-linearity specification only gives the limits
within which the linearity deviation lies, we have to make an
assumption about its random distribution to determine its
variance. For lack of further knowledge, we assume here a
uniform distribution of the non-linearity within the specified
limits:
pNL x = 1 within –SPC x SPC
NL NL
2 SPC NL
With this assumption we are able to evaluate an equivalent
variance of the non-linearity:
SPC NL SPC NL
s 2NL = x2 pNL x dx = x2 1 dx =
–SPC NL –SPC NL
2SPC NL
© Mettler Toledo, A. Reichmuth Uncertainty & Specs 1.1 Prtd.: October 27, 2000
—6—
= 1 1 x3 SPC = 1
NL
x3 SPC = NL
–SPC
2SPC NL 3 –SPC 6 SPC NL
NL
NL
Sensitivity (Deviation)
With the assumption that the balance’s sensitivity was ad-
justed with an internal or external calibration weight, we first
have to deal with the weight’s tolerance. Usually its deviation
is given in the data sheet as a tolerance band
SPC CAL max m CAL ,
and we find ourselves in the same situation as we were when
deriving a variance for the non-linearity. With the same rea-
soning we can write for the variance of the calibration weight
deviation
2
sCAL = 13 SPC CAL 2 . unit: [g2]
Furthermore, as calibration can only take place through a
weighing of the calibration weight, strictly speaking, we
would have to consider the repeatability of this calibration
weighing. However, calibration is a special case insofar as a
special signal processing is applied to it, i.e., usually a
stronger filtering and a longer measurement interval, thereby
improving the repeatability of the calibration weighing. For
practical reasons, this contribution may therefore be ne-
glected, which we will do here.
A further complication occurs if the calibration weight does
not amount to the balance’s full weighing capacity. When the
calibration weight is smaller than the weighing capacity, the
obtained calibration measurement is extrapolated to the
corresponding calibration value at full capacity.
Unfortunately, this calculation increases a potential linearity
deviation occurring at the load of the calibration weight by
the same factor. Particularly with precision balances,
possessing weighing capacities of several kilograms, it is for
practical reasons not always possible to build in a calibration
weight equal to its full capacity. With analytical balances, on
the other hand, the built-in calibration weight usually
embraces the full weighing capacity of the balance.
© Mettler Toledo, A. Reichmuth Uncertainty & Specs 1.1 Prtd.: October 27, 2000
—7—
Temperature Coefficient
The temperature coefficient of sensitivity may also be taken
from the data sheet (if this item is specified). This specification
describes the largest static sensitivity deviation caused by a
change in ambient temperature.
SPC TCS max TC S .
Again, we use the same procedure to obtain the variance
from the band limits (see derivation under the linearity devia-
tion)
2
sTCS = 13 SPC TCS 2 . unit: [1/K2]
About the properties of the temperature excursion we can
only speculate here. Unless there is additional knowledge
available about the course of ambient temperature, the
following assumptions seem reasonable:
• If the temperature at the location of the weighing is con-
stant, we can drop the influence of the temperature (coeffi-
cient) altogether.
• The temperature excursion at the location of the weighing
stays within a band of ±d t degrees.
• If an automatically induced calibration (for example
”FACT”) is active, then it is realistic to assume that a
maximum temperature change of ±2°C may occur, before
the balance gets adjusted.
© Mettler Toledo, A. Reichmuth Uncertainty & Specs 1.1 Prtd.: October 27, 2000
—8—
© Mettler Toledo, A. Reichmuth Uncertainty & Specs 1.1 Prtd.: October 27, 2000
—9—
Repeatability
Repeatability is an absolute deviation and by definition was
determined from the difference of pairs of readings.
Therefore, its contribution to one weighing (i.e., difference of
two readings) is the simple variance
s 2RP . unit: [g2]
Non-Linearity
Non-linearity is an absolute deviation. It occurs when weigh-
ing the tare, as well as the sample (tare and weighing ob-
© Mettler Toledo, A. Reichmuth Uncertainty & Specs 1.1 Prtd.: October 27, 2000
— 10 —
8) With the exception of weighings that include either zero load, full
capacity, or weighings with zero sample weight.
All three cases are but of academic interest, therefore they are not
considered here.
9) There may be a dependence of the readings for small samples. If
there is knowledge about such a correlation, it can be used; here,
we do not consider it for the sake of simplicity.
© Mettler Toledo, A. Reichmuth Uncertainty & Specs 1.1 Prtd.: October 27, 2000
— 11 —
Measurement Uncertainty
It is reasonable to assume that a balance’s combined mea-
surement deviation resembles a normal distribution. As one
of its justifications we mention the fact that some contributions
themselves are normally distributed already (e.g., repeatabili-
ty). A second reason is that there are multiple contributing
sources of independent deviations which favors a normal
distribution of their combined deviation.
From the combined standard deviation we can determine an
uncertainty interval from the laws of normal distribution,
provided a confidence level is given. We first derive from the
confidence level the expansion (or coverage) factor k, i.e.,
the quotient relating uncertainty to the standard deviation
u
kP = ,
s
© Mettler Toledo, A. Reichmuth Uncertainty & Specs 1.1 Prtd.: October 27, 2000
— 12 —
© Mettler Toledo, A. Reichmuth Uncertainty & Specs 1.1 Prtd.: October 27, 2000
— 13 —
Remark:
It can be seen from this example that with small sample
weights 13) the contribution to uncertainty originating from
the balance’s sensitivity (calibration weight tolerance and un-
© Mettler Toledo, A. Reichmuth Uncertainty & Specs 1.1 Prtd.: October 27, 2000
— 14 —
© Mettler Toledo, A. Reichmuth Uncertainty & Specs 1.1 Prtd.: October 27, 2000
— 15 —
© Mettler Toledo, A. Reichmuth Uncertainty & Specs 1.1 Prtd.: October 27, 2000
— 16 —
Approximations
Revisiting the figures of the previous example, we recognize
that with a sample weight of 1/200 of the weighing capacity,
the variance of the sensitivity deviation, calibration weight
tolerance and temperature coefficient, equals 1.8 10 –12 , an
amount negligible compared to the variance due to
repeatability and non-linearity ( 2.2 10 –9 ). It may be presumed,
that this be the case for all small sample weights. We will
show that this is true.
Taking the radicand of the relative uncertainty
1 SPC 2+2 SPC 2 + 1 SPC 2 1
+ SPC d 2
2 RP 3 NL 3 CAL,rel 3 TCS t
m
and expanding it by m 2, we get
2
SPC 2+ 2 SPC 2 + m SPC
RP 3 NL
2 1
+ SPC
CAL,rel d 23 TCS t
3
According to the assumption, that m is small (minimum
sample weight!), a fact which is even more true for its square
(m 2), we may drop the second term in favor of the first, and
we get as an approximation
© Mettler Toledo, A. Reichmuth Uncertainty & Specs 1.1 Prtd.: October 27, 2000
— 17 —
14) usually larger than 1/10 to 1/4 of the balance’s weighing capacity,
yet independent of total load
© Mettler Toledo, A. Reichmuth Uncertainty & Specs 1.1 Prtd.: October 27, 2000
— 18 —
© Mettler Toledo, A. Reichmuth Uncertainty & Specs 1.1 Prtd.: October 27, 2000
— 19 —
Conclusion
In many instances, the weighing result needs to be qualified.
To this end, the measurement uncertainty accompanying the
weighing process is required, but usually not readily avail-
able, not least because it is dependent on the application at
hand. At other times, the operator needs to know the mini-
mum amount of mass he/she is able to conformally weigh to
a required relative uncertainty and confidence level
(minimum weight).
This paper explains how uncertainty and minimum weight
can be estimated from the data sheet specifications of a
balance. The assumptions and restrictions, under which this
deduction is valid, as well as when, and under which
conditions, neglections can be made, are discussed. Two ex-
amples with actual data from analytical balances are given
as illustrations.
The theory and examples provided enables the user to
estimate the appropriate figures of uncertainty or minimum
sample weight for his/her balance application.
© Mettler Toledo, A. Reichmuth Uncertainty & Specs 1.1 Prtd.: October 27, 2000