You are on page 1of 28

Applying the ISO Guide

to the Calculation of Uncertainty:



Temperature

Robin E. Bentley

A version of:

Publication number TIP P1358

National Measurement Laboratory CSIRO

This document is a trimmed version of the hard-copy booklet [1] TIP P1358, which is a supplement to the monograph:

"Uncertainty in measurement: the ISO Guide". Monograph 1: NML Technology Series, TIP P1337.

National Measurement Laboratory CSIRO, October 2001

Copies of the booklet TIP Pl:358 and Monograph 1 are available from the National Measurement Laboratory, Bradfield Road (PO Box 218), Lindfield, NS'vV.

(phone: G1-2-9413 77(1)

CSIRO Australia. 20Dl

ii

Contents

1 Calibration of a therrnocouple

2

2 Calibration of a temperature indicator

7

3 Calibration of a Pt resistance thermometer

10

4 Calibration of a digital thermometer

15

5 Calibration of a mercury in glass t.her morneter

19

6 Calibration of a block calibrator

23

NOTE 1

The approach taken in the application of the ISO Guide and the symbolism used are explained in the NML Monograph 1 ref. [2].

In all the examples thai follow:

NOTE 2

Type A components were known to be relatively small. Thus, they were evaluated using a simplified technique involving the parameter rlLp, whose values are tabulated in the Monograph [21, for example, Tnp Cl.1'3 forn 20 and a 95% coverage probability.

NOTE 3

The calculation of uncertainty for a measurement should lead to a measurement statement. It should include the value of uncertainty, the coverage probability (9.5%) and the coverage factor. For the following examples, an appropriate statement would resemble:

of colibraiioti

±O.6°C.

Theuncertaintsj has been calculated using the 180 Guide to the Expression of Uncertainty in Measurement and is by the stated interval. which is estimated to contain the mcosurarui with 95% probability. The coverage factor fOT the interval is 2.1".

2

1 CALIBRAI10N OF A THERl\lOCOUPLE

1 Calibration of a therrnocouple

The measurerrient

Consider the calibration of a length of type I( thermocouple, insulated in woven ceramic fibre and capable of use to at least 1000°C. I shall assume that its tip was wire-wrapped to that of a tYPE, H. reference thermocouple and placed in the uniform-temperature region of a suitable tube furnace. The inner Inconel liner of the furnace was earthed to minimise ac pick-Up. ThE! thermocouples were extended to O°C using an ice point (a typc-R extension lead being used for the standard) and then via copper leads through a low-thermal switch to a calibrated DVM.

At each of SE!Ven calibration temperatures over the range 0 to 1000°C, the measurement sequence was: the type R thermocouple, a short circuit Cu,ro') and then the type K. The sequence was repeated to five values of each emf, with the 10m V range of the DVM used for the reference thermocouple and for H1(, short (subtracted from the former reading to minimise errors from instrumental zero) and the 100rnV range for the type K thermocouple (larger signal). The outcome was a report with a tabulated set of calibration corrections given at equispaced temperatures over the range (taken from a best-fit curve, discussed in section 4.2 of [2]).

The model

In this example, the mean emf's of each thermocouple were converted to values of temperature using standard reference functions", expressed here as functions of E, f( E). Thus. the calibration correction (the measur and}, C, is:

C T;) T

fR(ER) - hdEg),

(1)

where T:) is the 'true' temperature derived from the reference standard (using fll) and T that from the type I{ thermocouple (subscript 'K'). TIlE: emf of the standard W'l..S calibration-corrected (item 1, below) and all emf's were corrected for the calibration of the DVM. Equation (1) is the model for this example.

The sensitivity coefficient for all uncertainty components that affect En and are,

en

DIR DEI{

(1. Cl96 '"C /;.N at 400°C Dh

DEK

-(1.024 (lC/ltV at 400°C.

(2)

and

CJ(

(3)

In each case, the sensiti vity coefficient is the inverse of the Seebeck coefficient (respectively,

104 and 42 at /lOO°C~).

1 A set of polynomials that express E as a function of T, the reverse of what is required here.

Temperature may be determined from this function by iteration or by simply looking up the value in the standard tables (generated from the reference functions).

Uncertainty components

The following is a list of error sources, with each uncertainty estimate calculated for a tip tomperature near 400°C.

Reference temperature

1. Calibration of standard: the type R standard thermocouple had been calibrated at NML with a reported calibration uncertainty of 9 p,v+ 0.0:34% of emf') where the bracketed expression is an expanded uncertainty ba .. sed on a coverage factor of 2.10 (equivalent to Vi. At :H)O°C (Ell :3400 ,uV) , the calibration correction IS 1.45 fLV with an expanded uncertainty of ±2.0G JiV ((Ii = 2.0G flY).

2. Use of standard: in 100h of careful USE', a type R thermocouple may vary in calibration by up to about (page 97 in ref. [31), a reasonable estimate (Vi. 8, from Note I on page I). At 400°C this amounts to U, 1.5fLV. NML thermocouple calibrations include the effect of inhomogeneity, They will thereby apply to a use at any immersion within that length of thermocouple specified in the NML reports (typically, lOOOmm). Thus, no additional uncertainty component is required to cover inhomogeneity (i.e. cover the effect of a different temperature profile) in the present use. On the other hand, if inhomogeneity had not been COVElrE,d in the calibration (or covered, but not over a sufficient length) a significant component would then be needed (0.02% if new, otherwise, it could be excessive and would need a thermoelectric scan to determine a suitable value).

3. Extension lead for standard: NML calibrations for Pt-based thermocouples are usually done with the cold junction (C.J) end at ambient temperature. This suits the requirements of most clients and avoids inhomogeneity problems for the flexible end, if placed in an ice pot. So, the NML client (laboratory conducting the calibration in this example) would need an extension lead. Let us assume this lead, aIm pair of type R wires, protected from cold work over the length to be, used in the ice-pot, had been calibrated at NML to give a correction at 21°C of -0.6 ± 0.5 fiV (k 2.10 and Vi 18). 'rhus, trw corresponding uncertainty in ER is ±0.5{tV. Without calibration, the uncertainty component for the lead would have been about ±1.5pV, if new, otherwise it cannot be estimated and depends on the degree of handling it had received.

1. CJ temperature for standard: the C.J ends of the extension lead for the standard were located in an ice pot and were within ±O.02°C of O°C (a 'reasonable' estimate:

Vi == 8). If the temperature at this end had increased by 0.02°C, thermocouple emf would have decreased 0.111t V (Seebeck coefficient at O°C is 5.:3 It VI<-l).

5. Calibration of DVM for standard: the calibration had been clone at an ambient temperature of 20 ± 2°C and applies to an integration time of 100 power line

The expanded uncertainty reported for the calibration is lIt V 50 pprn for both

the l.Grn V range and the 100 III V and the coverage factor was not given. assume the value 2 (k;) and use the conservative value, Vi = 20

At 400°C, ER is about :3400{lV and so 1.17 pV.

So, let's

in [2]).

6. Use of DVM: in the present example the DVM was used at an ambient temperature of 20°C with an integration time of 100 power line cycles, the same conditions as those applying during calibration. Hence, no additional errors from these sources will need to be covered. The only remaining contribution to the 'use'

CALIBRATION OF A THERJVJOCOUPLE

uncertainty is drift/change in calibration since the DVM was last calibrated. This was estimated from manufacturer's specifications as ±25ppm, i.e, 0.09 tlV, and was considered 'reasonable' (Vi = 8).

7. Type A uncertainty in rnean reading for standard: at a calibration temperature of 400°C the emf range for five readings (mp = 0.51: tabulated in [2]) was 1. 7 ftV (zero corrected) for the type H t.herrnocouple and 9 pV for the type K. 'rhus, the tYPE~ A expanded uncertainty for the standard is 0.51 x 1.7 0.871lV (note that

for u, 4, ki 2.78).

8. Resolution of DVl\.1: the least count on the lClmV range was 0.1 ltV, so the

resolution uncertainty is 0.05 /t V and is considered 'rectangular' with ki = 1.7:3

and Vi 100 (Note 1 on page 1).

Type K thermocouple

9. Inhornogeneity: the type K thermocouple was new and we can assume a level of inhomogeneity equivalent to about ±O.l % of Emlf (page 13 in [:3, 4]), which is a 'reasonable' estimate (Vi cc 8, from Note 1 on page 1). For a tip temperature of

100°C (E}{ 16400 It V) this represents t/ie. 1 GAIN.

10. Drift during calibration: the calibration run from O°C to 1000°C took about 9h to complete and would have affected the calibration of the type I{ thermocouple. The change could be as much as OJ % of emf (the in-situ drift in 10 h for O. fnnm wires [51) and is inversely proportional to diameter. This represents about 16.4 /i V at 400"C and being roughly linear with time, its effect will be assumed 'rectangular: with it half-width of 8.2 ltV and k; Ln. The estimate is reasonable (Vi 8). Note, because of this change, the calibration will apply in a subsequent use only if the thermocouple" is used at deeper immersion. It will, however, apply to other thermocouples cut adjacently from trw same supply reel.

11. Temperature of thermocouple head: in this example the type K thermocouple was a single continuous pair of wires in flexible insulation with its C.J end at O°C. Thus, there was no extension lead error. If, instead, it had been in two sections (their join being the 'head'), as in the example on page 15, allowance should be made for variations in head temperature.

12. CJ temperature of type K thermocouple: the CJ ends of the thermocouple were in an ice pot and were within ±0.02°C of O°C (reasonable: Vi. 8). Since its Seebeck coefficient at (Joe is 40 jIVK-l this uncertainty component is equivalent to

a change in emf of 0.8 ltV.

1:3. Calibration of DVM: for an emf at 400'1(; of

calibration for the DVM is 1.81LV with A:i

1 G400 ltV, the uncertainty of

2 and Vi = 20 item 5).

14. Use of DVM: as for item 6, assume time (drift) was the only significant departure from calibration conditions for the DVM and, to cover this, assume a 'use' component of 25ppm, i.e., O.4p.V (v; = 8).

15. Type A uncertainty in mean reading: is 0.51 x 9 = 4.G ILV with Vi 4 and

k:i 2.78 (see item 7 above).

lG. Resolution of DVM: the resolution uncertainty for the 100mV range is

U O.C) /I.Y and is considered rectangular with k, = 1. 73 and Vi = IOO.

5

Other errors

17. Temperature difference: assume the tips of the two thermocouples differed in temperature by less than 0.02% (page 99 in ref. [3]), since they were wire-wrapped

together. This amounts to =, Cl.08"C at 400°C, considered 'reasonable' (Vi == 8).

18. Ac pick-up & thermals: the calibration furnace was switched off, briefly, at about lOOO°C, then on again--the effect was less than] pV--an indication of ac pick-up. Also, spurious thermoelectric signals (,thermals') in the circuit would have been less than 0.] ltV. I shall use the sensitivity coefficient for the type R thermocouple (the most sensitive of the two) and take the estimates as 'rough' (Vi== 2).

19. Fitting and interpolation: The cali brat ion data wa .. '3 plotted as a function of temperature (similar to Figure 4.4 in [2]) and the uncertainty in the choice of fit (±0.3°C) was gauged from the range of possible curves that were considered reasonable.

20. Rounding: the calibration corrections will be rounded to the nearest O.l°C when

reported == Cl.05()C and ki 1.7:3, being 'rectangular').

Correlation effects: In this example there were two sources with potential correlaLions: (1) the CJ region for the two thermocouples (items 4 and 12) and (2) the use of a common DVM (items 5, 6 and 13, 14). These components are small and, for convenience, were considered independent. (uncorrelated.), an assum ption that had no significant effect.

However, it is instructive to show how the effect of correlations may be handled. For example, if the CJ temperature had been low by O.02°C (value of uncertainty used above) and thought to affect both thermocouples equally, the emf's developed in both would be high. 'I'he emf of the standard would be high by 0.11 p.v, giving a measured temperature high by 0.01 °C (item 5, above), and that for the type K, high by 0.02°C (itern B). The net effect on the calibration, on 7;) T, would be only O.OI°C, equal to that chosen for item 4 alone. On the other hand, the effect of both, if assumed uncorrelated (independent) would be O.022"C ('root sum of squares').

Uncertainty calculation

the values of It; and Vi from Table 1, the calculation of expanded uncertainty proceeds as follows:

the combined standard uncertainty, Uc

the number of i.,ffective of freedom, Vei ! :36

the coverage factor, k: 2.03

and, thus, the expanded uncertainty, U Finally, after rounding: U

0.644°C. 0.7°C.

Notice that even though 20 uncertainty components were involved in the above calculation, less than half are significant. For example, items 9, 10 and Hl, which depend on the

behaviour of the type K thermocouple, if considered alone, U 0.5°C.

G

CALIBRATION OF A THER1HOCOUPLE

Table 1: An uncertainty for the calibration of a type K thermocouple. It lists the components of uncertainty at 400"(:, as both (raw) expanded uncertainties U, and standard uncertainties in the measurand Vi. Included are values of reducing factor hi, sensitivity coefficient

c, and number of of freedom V,.

Cornponent (i = ],2, ... 20)

Component values
k led LLj * (OC) vI
l
2.06pV 2.10 O.09G 0.094 18
1.5 IIV 2 0.09G (l.()72 8
O.S p.V 2.10 0.096 0.023 18
0.11 pV 2 O. Cl96 O.OOS 8
1.17/N 2 Cl .cJ96 (J.Of)6 20
O.09,N 2 0.096 0.004 8
0.87/N 2.78 0.096 0.030 4
O.OS ltV 1.7:3 0.09G 0.003 ]00
16.4/N 2 0.024 0.197 8
8.2/N 1.7:3 0.024 0.114 8
O.8IN 2 0.024 O.OIO 8
1.8 pV 2 0.024 0.022 20
OA /IV 2 0.cl24 O.OOS 8
4.G ltV 2.78 Cl.024 0.040 4
0.5 pV 1.73 0.024 0.007 100
O.08°C 2 1.0 0.040 8
] IN 2 0.096 0.048 2
0.:3 DC 2 1.0 O.IS0 8
0.05°C 1.7:3 1.0 0.029 100
(lid k,) x led Reference temperature

1. calibration of SI'D

2. USi, of S'ID

:3. extension lead for STD 4. C.J temperature for STD ,I. calibration of DVM

6. use of DVM

7. mean reading (type A)

8. resolution of DVM Type K thermocouple

9. inhomogeneity

10. drift during calibration

11. thermocouple 'head'

12. C.J temperature

1:3. calibration of DVM lit use of DVM

15. mean reading (type A)

16. resolution of DVM Other errors

17. tip temp. difference

18. ac pick-up & thermals

19. data fit and interpolation

20. rounding of final result

* standard uncertain ty in measurand .

Note 1 on page 1.

7

2 Calibration of a ternperature indicator for therrnocouples

Temperature indicators designed for use with thermocouple input may be calibrated using a vol tage source, in parallel with a DVM, connected in series with a thermocouple extension lead operating from O°C to the input terminals. They are described in ref. [:3, 4].

However, it is simpler (but potentially less accurate) to use a 'thermocouple calibrator', a device that can either indicate temperature when connected to a thermocouple or act as a thermocouple simulator for calibrating instruments. It has automatic cold junction compensation (ACJC:) that may be switched in (IN1') or out (EXT) of the circuit

1). Its use is convenient because the operator merely selects the appropriate thermocouple type and 'dials' up suitable values of temperature, Moreover, the extension lead that COlUH;Cts the two instruments need not be calibrated and introduces minimal error if the two sets of terminals an, similar in temperature,

TC calibrator

EXT

extension wires

indicator

T s

Figure 1: Circuit for calibrating an indicator with thermocouple input. 'I'he calibrator acts as a source V, with an option to add a negative AC.JC source (select 'INT') that corresponds t.o the temperature of its output terminals. Then, as shown, thermocouple extension wires of the correct type and polarity are used to compensate for any difference in temperature between the two pairs of terminals.

The measurement

For this example, assume a thermocouple calibrator had been properly calibrated as II 'source' with ACJC set for INT operation and that it was used to calibrate a type T indicator over the range Cl to 400°C (connected according to Figure 1).

The model

In this C2\,Sf:, the model is very simple-- the calibration correction. C. is:

C 7~) I'

where C is the 'rneasurand', TJ is the 'true' temperature set on the reference source (the 'calibrator"), and calibration corrected, and I' is the corresponding temperature indicated by the type 'I' instrument.

The sensitivity coefficient for 7:, is DC/DTo = 1 and for I' is DC/DT 1.

2 CALIBRATION OP A TE/viPERATUH.B INDICATOR

Uncertainty components

'The components of uncertainty at 200°C are as follows, and are listed in Table 2.

Reference signal

l. Calibration of calibrator: the type '1' range of the thermocouple calibrator (simulator mode) had been calibrated for IN'r AC.JC and at an ambient temperature of200(:. The expanded uncertainty for the calibration was stated as (0.20()C O.

of temperature) with a coverage factor of k 2.05, equivalent to v r-;» 27 (from table

of Student's values [2]). At 200°C this amounts to cc·0.24°C:.

2. Use of calibrator: assume that drift in calibration for the calibrator will need estimation and that other conditions of use (ambient temperature, AC;.JC, are the same as those that applied during calibration. Data in the manufacturer's specifications for the device were insufficient to estimate drift behaviour. So, a value equal to half the, calibration uncertainty was chosen to cover drift ( 0.12"C).

This is considered a rough (estimate ([I.; 2).

:3. Resolution: the resolution of the calibrator was O.05"C (output wa .. s adjustable to

0.1 and the uncertainty is considered rectangular with k, 1.7:3 and Vi .=c 100.

TC indicator

:1. Stability of indicator at QOC: repeated checks of the indicator during its cali bration gave a scatter of ±O.l°C. Thus, Ui= O.l°C and is considered a 'reasonable' estimate (Vi 8, from Note 1 on page 1).

'1. Change in ambient temperature: at NML, thermocouple indicators are checked in an oven at 40°C (20°C above the ambient temperature for the calibration) to observe and report on, their dependence on ambient. Assume this was done and that the indicator 'zero' had increased by 0.3°C when equilibrated at 40°C and that the indication for 200°C gave the same increase, indicating that the gain had been unaffected. Thus, a ±2°C variation in ambient during calibration (reasonable: [I; =cc 8) would have affected the calibration at 20°C by ±O.0:3°C, giving = 0.03°C.

6. Type A uncertainty in mean reading: for an input equivalent to 200°C, the indicator display flickered between two adjacent values (over a range of 0.1 °C, the least count) over a period equivalent to approximately 2CJ integrations. So, put

TlLpR CU:; x 0.1= 0.013°C with k, c= 2.1 and Vi. = 19.

7. Resolution of indicator: the least count was 0.1 °C and the resolution uncertainty 0.05°C) is considered rectangular with ki = 1. 7:3 and [Ii = 100.

Other errors

8. Extension lead: the type T extension lead was unlikely to be in error by more than about 1 % of the temperature difference it spanned, say <5°C, an over-estimate, thus, 'rough' ([Ii 2). So, assume an uncertainty of U, = 0.05°C.

9. Thermals: the thermals (spurious thermoelectric signals) in the circuit were assumed less than l/1V (reasonable, i.e., [Ii = 8). Since the Seebeck coefficient for type T thermocouple is 5:3 pV.K-1 at 200°C, this amounts to U, = 0.019°C.

10. Ac pick-up: ac effects in the measurement circuit were thought to be negligible.

11. Fitting and interpolation: the plotted data (calibration corrections versus tern perature) from 0 to 400°C scattered about a linear fit by up to ±O.] °C. So,

O.PC and Vi 8 (the estimate was treasonable").

12. Rounding: the calibration data will be rounded to the nearest O.OPC when reported

( 0.OCJ5()C and k, =" 1.73, since 'rectangular').

There were no correlation effects.

Table 2: An uncertainty for the calibration of a ternperature indicator for thermocou-

It lists the components of uncertainty at 2000C\ both (raw) expanded uncertainties [I, and

standard uncertainties in the measurand 1[,. Included are values of reducing factor k" sensitivity

coefficient and number of of freedom u,

Component (i= 1,2, ... 12)

Component values
( k led lhi * (Oe) v+ I
'1 L I
0.24 2.05 1 0.117 27
0.12 2 C1.06 2
Cl.05 1. 73 1 0.029 100
0.1 2 1 0.0,') 8
0.03 2 I 0.015 8
0.013 2.1 O. ClO6 19
(1.05 1.73 0.029 100
Cl.05 2 1 0.025 2
O.OH) 2 0.010 8
0.1 2 1 o.os 8
0.005 1.73 0.003 100 Reference signal

1. calibration of TC calibrator

2. use of calibrator :3. resolution

TC indicator

4. short-term instability at O°C

5. effect of ±2°C in ambient

6. mean reading (type A)

7. resolution Other errors

8. type T extension lead

9. thermals

10. ac pick-up

11. data fit and interpolation

12. rounding of final result

* standard uncertainty in measurand , u'i ~=; ([Ii) ki) X lei,] NClte 1 C!Il pa .. ge 1

Uncertainty calculation

the values of Ui and Vi from Table 2, the calculation of expanded uncertainty proceeds as follows:

the combined standard uncertainty. Ilc

the number of effectiVE, of freedom, vef f 40

the coverage factor, k: 2.Cl2

and, thus, the expanded uncertainty, U Finally, after rounding: U

0.:319°C. CJ.:32°C.

Notice that the use of the thermocouple calibrator had doubled the uncertainty. Without items 1 2 and :3. U = o.is-c.

10

.3 CALIBRATION OF A PT RESISTANCE TllERAtONIETER

3 Calibration of a Pt resistance thermometer

The rneasurernent

Consider the calibration of a 100 n (nominal value at -l-wire industrial Pt resistance thermometer (IPHT) that had been compared with a good quality calibrated digital thermometer. Trw digital thermometer consisted of a 100 n Pt resistance probe connected to a free-standing instrument (the combination was calibrated as a unit-s-the 't.hermorneter"). The resistance of the IPHT being calibrated was measured using the -l-terminal input of a calibrated digital multi meter (DMM) and the two probes were compared in stirred liquid baths (water and oil) over the range 0 to 250°C. Measurements at O°C were done only at the start and finish of the sequence.

The model

The aim in calibration is to obtain the correction:

where to is the 'true' temperature derived from the calibrated reference sensor (corrected for its calibration) and t is the value of temperature given for the IPRT by:

aiu: 1) b(w-l)2 +-c(w-l):>, «t«

and R; awl R" are the resistances of the IPET at temperature t and at O°C, respectively.

The first step in the procedure would be to compare the [PET with the reference sensors at five or more values of t and to evaluate the coefficients o., band c by fitting the above cubic to the data. The two or more extra data allow a check on the fit. So, in this example, assume measurements were made at six temperatures. (At NML, an automatic calibration facility is used to generate 15 data for the range 0 to 250°C.)

One way of examining the fit and of estimating the uncertainty of fit is to convert the raw calibration data, of R; into values of t using the coefficients determined from them, so values of C at the calibration temperatures. The values of C, which would scatter about zero, may be plotted as a function of temperature to display the goodness of fit.

The rneasurand is C and the modelling expression is equation in combination with

The 'input quantities', the measured parameters required to get C, are R; and Ro.

So, we need three sensitivity coefficients, and the eCksiest is that for to. It is aCjfJto 1.

The sensitivity coefficients for Rt and R; are, respectively, Ct and co:

Ct

DC URI ac

(G)

and. similarly.

Dt w

where

Dt Dw

a +- 2b(w - 1) + 3c(w

from eqn (5)

(8)

11

Let us assume for the IPRT typical values for the coefficients o., band c. They are, for a sensor with Ro 10CH2: a 251 b = 9.7"C and c ec_c 0.6°C. Consequently, as a guide to values for 11) and atl Dw and, thus, for sufficiently accurate estimates for Ct and

Co, we have from equations and (8):

U'

1.007 -!- 0.00:381 251 + 008 t

(9) (10)

and

Notice that equations (9) and (10) are given here for the readers convenience and to allow the calculation to proceed. In practice, values ofw and Dt/ Dw would be known by the user from experience. Alternatively, the USE)r may use a numerical approach in evaluating Ct and Co, as explained in 121, an approach not requiring knowledge of at/D·w.

At 200°C withRo

100 n, we have UJ 1.77 E:ft/aw = 267 and:

and

-2.G7 1.7:3

mK/mil mK/mn.

(11)

In this example, it is particularly important that the correct signs be assigned to the sensitivity coefficients because the coefficients will be used when handling any correlated errors that affect both Rt. and Ro·

Uncertainty components

For such a calibration I suggest the following components of uncertainty (evaluated at 200()(: and listed in Table 3).

Reference ternperature

l. Calibration of digital ther-mometer: the calibration uncertainty was ±20 rnK

20 mK) over the range (J to 250"C and the; calibration was done at an ambient temperature of 212°C. The stated coverage factor wa .. s 2.10 (corresponding to Vi 18: table of Student's values [21). The calibration correction would have included the resistive-heating effect (see item fl, below) of the measuring current (typically 1 mA)---tlw effect would have been small for a probe in direct contact with stirred water or oil.

2. Use of thormomctor: the conditions of use for the digital thermometer (ambient temperature, measuring current, being in a liquid, etc) were the same as those that had applied during its calibration, so drift in calibration was the only contribution that needed consideration. One guide would be any change observed at the ice point (O°C), measurements that should be done regularly and after any significant temperature; excursion (>200"C). Drift at (Joe has three causes: (1) drift in the Pt sensor, which would affect mea .. surernent.s at higher temperatures to a lesser extent, (2) drift in the 'zero' of the electronics (A/D converter, DVM module, which would affect all temperature values equally, and Ul) drift in gain, which affects temperature proportionally. In this example I have chosen an uncertainty of ±G mK, based on manufacturer's specifications and on experience. Moreover, drift is thought more likely to be linear (rectangular distribution) than Gaussian, so, U, was taken as the half-width 6 mK, half the extent of possible drift for the period, and the reducing factor k, as 1.73. Being a reasonable estimate, Vi = 8 (from Note 1 on page 1).

12

3 CALIBRATION OF A PI' RESIS'I~4NCE THERMOMETER

3. Type A uncertainty in mean reading: at 200"C the observed range (R) of data displayed on the digi tal thermometer was 8 mK over (nominally) 20 readings (instrument integrations). The mean reading was estimated 'by and the component expanded uncertainty calculated a .. S n~pR, with Trip 0.13, i.e., U;== 1.04 mK (for

n ~ 20, ki 2.1 and 1/; = 19).

l. Resolution of thermometer: the IE:ast count of the digital thermometer was

1 InK and the resolution uncertainty O.5rnK) is considered rectangular with

k:i 1.73 and Vi 100.

The IPRT

5. Stability of IPRT at O°C:: ice point measurements of the IPltT' at the start and finish of the calibration sequence differed by 2 mn. Consider this value as a reflection of short-term instability in the sensor and possibly the DMM~see item 8, belowand aSSUIIW it was due to linear drift. As for item 2, above, its effect was assumed

rectangular with == 1 mn, k; 1.73 and Vi 8 ('reasonablE:').

G. Heating effect in IPRT: the measuring current for the 1 kfl range of the DMM was 1 rnA and would have caused some resistive heating of the probe. The effect for a probe in direct contact with either ice/water (Ooe:) or stirred oil at 200CJC would be up to ~lO mK depending on the sensor construction and temperature. ThE) electrical power dissipated in a PRT sensor is proportional to its resistance Rt· If the thermal resistance between the element and its surroundings is assumed a constant, the resultant temperature rise will also bE) proportional to Rt. Moreover the resultant increase in sensor tern perature will be proportional to Rt since cit/ dsn is roughly constant (page 10). Hence, the effect of power dissipation on R; and R) cancel on forming the ratio, u: Rtf Ro. That is, the heating effect in a PKT' will have no effect on the measurement, given the above assumptions.

In practice, though, the measuring current of the DMM (nominally 1 rnA) will not be the same for the two measurements, of Ro and Rt, because of medium-term instability. Consequently, the heating effects won't cancel exactly, and I have allowed 10% of the potential heating effect of about 10 mK to cover the instability, i.e, 1 mK. Moreover, being equivalent to drift in current its effect was assumed rectangular with 0.5rnK, k, = 1. 7:3 and, being a rough estimate, Vi = 2. Notice that it lOrnA current would generate 100 times the effect.

7. Calibration of DMM: the 1 kO range of the DM!'v1 had been calibrated for an

integration time of 10 power line and at an ambient temperature of 21 ± 2°C.

The calibration uncertainty was Inn + 0.002% of reading), based on a coverage

factor of 2.10. The DMM was used for trw measurements of both Rt and R; and the effects of its calibration would have been correlated.

At 200"C, R, = 177 nand Ro =c~ 1000, and the corresponding expanded uncertainties for the calibration are 5.54 mrl and 4.0 mn. When converted to effects on the measurand t+es values of ciUd ki---they become -7.04 mK and 9.01 mK

-2.G7mK/mn, Co = 4.73mK/mn and k; = 2.10). Thus, if the calibration uncertainties had been fully correlated. their effects would add and their net effect would have been -7.04+9.01 ~~ 1.97mK. On the other hand, if the components had been independent their net effect would have been /(7.042 + 9.012) = 11.43rnK.

In this application the DMM components were thought to be partially correlated and their net effect taken as the average magnitude of the effects of full and no

by treating each as positive) [2]: ii; = (1.97 11.1:3)/2 6.7mK.

and the approach was considered reasonable (Vi 8).

correlation

8. Use of DMM: the DIvIM was used under the same conditions (ambient temperature, integration tirne, that applied during its calibration, so drift was the only relevant component. Such drift would already have been included in item 5, above.

9. Type A uncertainty in mean DMM readings: adopting the same approach taken in item 3, above, a range of 4 rnn was noted on the DMM for H; and :3 mD for R,. These translate to the values 0.52 and (Ul9mn, respectively, a reducing

factor of k, 2.1 and Vi 19.

10. Resolution of DMM: the resolution uncertainty is ± half the least count of 1 rnn.

It would affect both resistance measurements, in the sense that it represents trw lower limit for uncertainty in their measurement. So, apply it once and use the largest sensitivity coefficient, that for R). The uncertainty would have a rectangular

distribution with k; .=c 1. 7:3 and Vi 100.

Other errors

ll. Temperature difference: assume the spatial variation in the oil bath at 200°C was assessed as 5 mK over the region in which PRT sensors are generally placed for calibration, being the maximum difference between any two temperatures mea ... sured in the region. It was assumed from this that there was a 95% chance that the temperature difference between any two probes was within the range 0±5 rnK, i. e., [I; 5 mK. The value of was considered only reasonable (Zli 8) because the data do not refer specifically to the probes used. The temporal variation in the bath due to control action was typically 4 mK, but would have little effect on the probe comparison since they have similar response times.

12. Ac pick-up: ac pick-up was considered insignificant.

13. Fitting and interpolation: calibrations were done at six temperatures, from which six values of C were obtained, scattering about zero. The range of values was 20 mK, and with Inp 0.1 (Vi = 5), we have U, 8 rnK. This reflects the goodness of fit for the chosen cubic.

14. Rounding: the measurand will be rounded to 1 m K in the calibration report (U; O.5mK with k, 1.73 and Vi = 1(0).

Correlations were noted for the heating effect (item 6, above) and for the DMM calibration (item 7) and were dealt with when evaluating their contributions to uncertainty.

Uncertainty calculation

the values of u., and Vi from Table 3, the calculation of expanded uncertainty proceeds as follows:

the combined standard uncertainty, Ue the number of effective degrees of freedom, Vel 1 the coverage fad or , k and, thus, the expanded uncertainty, U Finally, after rounding: [I

BG mK 40

2.02

27.5 mK. 28mK

11

3 CALIBILA110N OF A. PT RESlSTj·tNCB TILER1\;101\1ETER

Table 3: An uncertainty for the calibration of an IPRT. It lists the components of

uncertainty at 2000e. as both expanded uncertainties Ut and standard uncertainties in the

mcasurand u.. Included are values of reducing factor sensitivity coefficient Ct and number of

of freedom II,.

Component (i 1,2, ... 14)

Reference temperature

1. calibration of ST'D thermometer

2. use of STD thermometer :3. mean reading (type A)

4. resolution (half lea .. st count)

IPRT

5. repeatability at ice point G. heating effect (Rt &. Ro)

7. calibration of DMM .n, &. R))

8. USE; of DMM

9. mean reading (type A) for R;

mean wading (type A) [orRo

10. resolution (half least count) Other errors

11. bath temperature uniformity

12. ac pick-Up

L:l. data fit and interpolation 14. rounding of final result

Component values
k; leil (mK) v+
1
- -------------.--,--,- 1-._._._-_.
20rnK 2.Hl 1 9.5 18
GrnK I 7:3 1 :3.5 8
J mK 2.1 0.5 19
0.5mK 1.7:3 0.:3 100
lmn 1.7:3 4.7:3 2.7 8
O.5mK 1.73 o.a 2
na 2.10 na G.7 8
O.52mS2 2.1 2.67 D.7 19
0.39mS2 2.1 4.73 0.9 19
O.5mn 1.73 4.73 1.4 100
f)mI( 2 1 2.5 8
8rnK 2 4.0 5
O.5mK 1.73 0.:3 100 '* standard uncertainty in measurand . 'Ui (Ui./ki,) X Ie;,:· I

t see N cite 1 on page 1.

15

4 Calibration of a digital therrnometer (therrnocouple pro be)

A digital thermometer senses temperature, displays its value in temperature units and does so digitally. The sensor may be a thermocouple, a PRT or a thermistor.

I ts calibration and the assessment. of uncertainty are much the same whatever the probe

with one exception. A thermocouple is not a localised sensor, lib" the others. Its emf is distributed along it.s wires from the tip to the instrument and, thus, any inhomogeneity in Seebeck coefficient will cause error (an undesired dependence on temperature profile). Moreover, if the probe is in two sections, one designed as 'the probe', often of an integrated design with metal sheath. and the other, a flexible section for connection to the instrument. then the difference in mean Seebeck coefficient between the two sections would be a source of error. Thus, two additional measurements need to be made:

• the tip of the thermocouple probe and the instrument end of its flexible lead should be placed at 0°(: with their connection (referred to as the 'head') at ambient temperature. The net developed emf is a direct indication of the error that occurs when the head is 20°C above the temperature of the instrument terminals (item 7, below), and

• the probe should be thermoelectrically scanned at the temperature furthest from ambient (item 8, below). The measurement should cover as much of the probe length as possible without causing a significant change in head temperature (depending on the results of the above ice-point check).

For this example I have chosen a thermometer based OIl a thermocouple probe

so as to demonstrate how to deal with the above difficulties. For other probe types, the two additional components of uncertainty (items 7 and 8, below) would not apply.

The rneasurernent

Consider the combination: a type-K thermocouple probe plugged int.o an instrument designed for type K thermocouples. The probe had a 500 mrn long metal-sheathed (1\111\1S) section welded to a 1 In length of flexible lead, which in turn was attached to a miniature type-K thermocouple plug. The digital thermometer was calibrated in stirred liquid baths from :30 to 200°C by comparison with a calibrated IPHI', whose resistance was measured with a DMM, also calibrated. The resistance of the IPET WW3 also measured at (J°C. at the start of the calibration sequence.

The model

The calibration correction for the digital thermometer (the measurand}, C, is:

C

1;) t

f (~:)-1'

(12)

where J~) is the 'true' temperature derived from the calibrated IPHT, its calibration being in the form of equation ,represented here as f(), and T is the temperature indicated by the digital thermometer. 1;, is determined from the ratio of the measured resistances for the If'R'I', Rt, obtained at 1', and Ro, at (r'c. Both resistances were corrected for the DMM calibration.

Hi

4 Calibration of a digital thermometer (thermocouple probe)

A digital thermometer senses temperature, displays its value in temperature units and does so digitally. 'I'he sensor may be a thermocouple" a PRT or a thermistor.

Its calibration and the assessment of uncertainty are much the same whatever the probe type, with one exception. A thermocouple is not a localised sensor, like the others. Its emf is distributed along its wires from the tip to the instrument and, thus, any inhomogeneity in Seebeck coefficient will cause error (an undesired dependence on temperature profile). Moreover, if the probe is in two sections, one designed as 'the probe', often of an integrated design with metal sheath, and the other, a flexible section for connection to the instrument, then the difference in mean Seebeck coefficient between the two sections would be a source of error. Thus, two additional measurements need to be made:

• the tip of the thermocouple probe and the instrument end of its flexible lead should be placed at O°C with their connection (referred to as the 'head') at ambient temperature. The net developed emf is a direct indication of the error that occurs when the head is 20"C above the temperature of the instrument terminals (item 7, below), and

• the probe should be thermoelectrically scanned at the temperature furthest from ambient. (item 8, below): The measurement should cover as much of the probe length as possible without causing a significant change in head temperature (depending on the results of the above ice-point check).

For this example I have chosen a digital thermometer based on a thermocouple probe so as to demonstrate how to deal with the above difficulties. For other probe types, the two additional components of uncertainty (items 7 and 8, below) would not apply.

The rneasurernent

Consider the combination: a type-K thermocouple probe plugged into an instrument designed for type K thermocouples. The probe had a 500 mm long metal-sheathed (MIMS) section welded to aLrn length of flexible lead, which in turn was attached to a miniature typo-K thermocouple plug. The digital thermometer was calibrated in stirred liquid baths from :30 to 200°C by comparison with a calibrated IPRT, whose resistance was measured with a DMi'vL also calibrated. The resistance of the IPRT was also measured at (PC. at the start of the calibration sequence.

The rnodel

The calibration correction for the digital thermometer (the rneasurand}, C, is:

C To T

f (17t) - T

Ro

(12)

where 1~) is the 'true' temperature derived from the calibrated IPRT, its calibration being in the form of equation (5), represented hew as fO, and T is the temperature indicated by the digital thermometer. To is determined from the ratio of the measured resistances for the IPRT, Rt, obtained at 1', and Ro, at O°C. Both resistances were corrected for the DMM calibration.

4 CALIBRATION OF A DIGIL'lL THERlv1011;fETEH

For convenience, let's assume the IPHI was the one calibrated in section 3 and,in particular, that its characteristics were those given on page 1 L Thus, we can use the sensitivity coefficients for Ht and R; (with a change in sign, because the IPHI was used for T;) rather than T) evaluated at 2ClOoC, the temperature of this analysis:

and

2,67 -Ltn

mKjrnft

mE/Inn at 20(('C ,

(13)

Notice from equation (12) that the sensitivity coefficient for '1;, is it is I ,

and for T

Uncertainty components

The components evaluated at 200°C, and listed in Table 4, are as follows,

Reference tcmperat.ure

L Calibration of standard IPRT: the IPF(f had been calibrated with an expanded uncertainty equivalent to O'c)200(; and the coverage factor was not given, So, let's assume k'i 2 and use the conservative value, Vi 20 (suggested in [2]), Moreover, the cali brat.ion relates directly to '1:) (sensitivity coefficient is 1),

2, Use of the standard: the ice-point resistance Ro, measured for the IPHI' during the calibration of the digital thermometer, was ~2 mn different from that given in the NML report Since the NML calibration uncertainty covered an ice point drift of up to 10 mf2 and the conditions of use (ambient temperature, measuring current, etc') matched those of the NML calibration, the (additional) uncertainty of 'use' for the IPHT is nominally zero,

~3, Calibration of Dl\1M: assume the same, DMM as that described on page 12 (it.ern 7) with kic= 2, lCL And, for the reasons given before, the net effect on I;, of calibration

errors in Ht and H; was taken aSLLi. G,7 ml< (partial correlation) and Vi 8,

4, Use of DM'M: the DlVIlVI was used under the same conditions that applied during its calibration, so drift was the only contribution relevant in a 'use' component, In this case, drift since calibration was considered insignificant.

5, Type A uncertainty in mean reading for IPRT: while measuring R; and Ro, the range (R) of data was in both cases about 4 rnn over roughly 20 'readings' (integrations of the DMM), Each represents an expanded uncertainty, calculated as mpR (page 1) of 0,52 mn and the reducing factor k, was 2, L They independently affect the resistance ratio by the same amount, so, both were represented in the analysis table by the single value = CL52 rnn with an effective sensitivity coefficient of = O,00543°Cjmll Instead of using a single entry for both they could have been represented separately, with their individual sensitivity coefficients, Cl,00267 and -0.00473°Cjrnn (equation (13)), The result would have been the same,

6, Resolution of DMM: the resolution component is O,5mn and I shall use the sensitivity coefficient for Ro, -0,0047~rCjmn, being the largest of the two. The uncertainty would be rectangular with k, L 73 and Vi ='c 100,

Digital (thermocouple) thermometer

17

7. Head temper-ature: because of possible differences in Seebeck coefficient between that of the probe and its extension lead, variations in head temperature will affect the net emf developed. So, the thermocouple was unplugged from its instrument and the pins of the plug were attached to copper wires and immersed sufficiently in an ice pot to reach O°C. The probe tip Wit .. S also placed at O°C and the emf developed between the two copper wires was measured on the DMM. The emf, 8 jJ,V, arising from the head being 20°C above that of the tip and the plug, is equivalent to 0.2°C (for type 1(, the Seebeck coefficient is <10{NK1).

The conversion of 8/tV into O. the corresponding chango in the input quantity T of equation (12) is equivalent to using a 'local' sensitivity coefficient, as described in l2J (another example is in item 10, below).

Allowing for a possible variation in ambient temperature of ±10°C (to be stated in the calibration report}, relative to the temperature of the instrument terminals, include the component U; 0.1 ClC. This was considered a 'good' estimate (Vi = 50, from page 1).

8. Inhomogeneity: the calibration at 200T: was repeated for various immersion depths of the probe, beginning at 350 mm, at which point the head temperature rose 5°C abOVE; ambient, and decreased in 50 mrn increments. For a depth range of 100 to :350 mm the calibration corrections fluctuated over a range of 0.:3°C, and at less than 100 mm immersion the probe temperature fell rapidly, because of thermal conduction losses. Therefore, the inhomogeneity component for the probe is ±0.15°C (and the cali brat.ion report would specify UlE: immersion range covered). It is a good estimate (Vi ,= 50).

9. Stability of digital thermometer at (Y'C; during calibration the digital thermorneter was checked at O°C a number of times and a range of 0.1 °C was noted. This was taken as a reasonable (Vi ,- 8) measure of short-term instability (U; 0.05°C) for the thermometer and assumed Gaussian (centralised and unlikely to be truncated at ±0.1 GC).

10. Change in ambient temperature: the 'zero' of the indicator increased by 0.5 °C when equilibrated at 4CfC in an oven. 'Thus, a ±2°C fluctuation in ambient during

calibration (reasonable: Vi. 8) would have affected the calibration by ±0.05°C.

11. Type A uncertainty in mean reading: the approach suggested in Note 1 on page 1 was adopted to give, for roughly 20 readings, an observed range of O.PC. This amounts to an expanded uncertainty of O.01:3°C (ki c.= 2.1 and Vi = 19).

12. Resolution of thermometer: the resolution uncertainty is ±0.05°C (least count

0.1 and is rectangular with k, = 1. 7:3 and 1/; 100.

Other errors

13. Temperature difference: the spatial variation in the oil bath at 200°C was assessed as 5 rnK over the region in which PRT sensors are generally placed for calibration, being the maximum difference between any two temperatures measured in the region. It was assumed from this that there was a chance that the temperature difference between any two probes was within the range 0±5 m1(, i.e., [I; == CJ.005°C. TIw value of was considered only reasonable (Vi = 8) because the data do not refer specifically to the probes used. The temporal variation in the bath due to control action was typically <1 mK, but would have little effect on the probe comparison since they have similar response times.

l~

"1 CALIBRATION OF A DIGITAL THERMOMETER

14. Ac pick-up: ac pick-up was considered insignificant.

15. Fitting and interpolation: the largest deviation of the calibration data, values of T~) - T versus T, from the smooth fit chosen to represent the calibration was 0.1 °C.

'Iaking" 0.1 "C was considered reasonable (Vi 8).

16. Rounding: the calibration results will be rounded to O.c)] °C when finally presented = 0.OD5°C and k, 1.7:3, being 'rectangular').

There were no correlation effects.

Table 4: An uncertainty for the calibration of a digital thermometer. It lists the components of uncertainty at 200"·C;. as both expanded uncertainties U, and standard uncertainties in the measurandu.. Included are values of reducing factor k, sensitivity coefficient

c, and number of of freedom II,.

Component (i= 1,2, ... 16)

ti,

Component values

k, [cd

0.020°C

2 1

(GCl

Reference temperature

1. calibration of 3'1'D PHI'

2. use of 3TD PHT

:3. calibration ofDMM

4. use of DMM

5. mean PRJ' readings (type A)

6. resolution (half least count)

Digital thermometer

7. effect of 10°(: in head

8. probe inhomogeneity

9. short-term instability at (('C

10. ±2°C in ambient (instrum.)

11. mean readings (type A)

12. resolution (half least count) Other errors

1:3. sensor temperature difference 0.005°C 2 1

14. ac pick-up

15. data fit and interpolation (Ll 0 °C 2

Hi. rounding of final result 0.005 °C 1. 7:3 1

na

2.10 na

O.52mn

2.1 O.00.,}4:3

0.5 mn

1. 7:3 0.0047:3

0.10 °C 0.15 °C 0.05 °C

2 1

2 1

2 1

2 1

0.05 °C O.OI3°C D.05 °C

2.1 1

1.7:3 1

0.010 20
0.007 ~
o.uot 19
D.OOI JOO
0.05 50
0.075 50
0.025 s
Cl.025 8
0.006 19
0.029 100
0.00:3 8
o.os 8
0.00:3 100 standard uricert.ai nty in mea .. surand , t see Not« 1 on page 1.

Uncertainty calculation

Using the values of u ; and Vi from Table 5, the calculation of expanded uncertainty proceeds as follows:

the, combined standard uncertainty,

the number of effective of freedom,

Veil the coverage factor, k

and. thus, the expanded uncertainty, U Finally, after rounding: U

0.114 CiC 101

1.98 O.226°C. 0.2:3°C.

19

5 Calibration of a mercury in glass therrnometer

In my opinion, the calibration of a mercury in glass thermometer at 'partial immersion', at significantly less than full immersion, is messy because of the difficulty in establishing the temperature of the mercury within that section of the bore not enclosed by the bath liquid. 'rhus, I shall consider a thermometer designed for full immersion-s-a 'total-immersion' t.ype graduated in O.I°C steps from 100 to l50GC. The thermometer had a subsidiary scale around (J°C to allow stability checks at trw 'ice point' (OC'C).

The rneasurernent

The thermometer wa .. S initially conditioned [6J to minimise instability in bulb volume. It was then calibrated at full immersion in a stirred-oil bath and an ice-point reading was obtained before and after calibration as a stability check. The calibration was done relative to the IPH'T/DMM combination used in section 4 (page 15) at the temperatures corresponding to its pointing marks (every lOOC) and midway between these marks. In other words, the calibration was done in 5°(: steps. The resistance of the IPRT at (J°C, R" was measured at the start of the calibration sequence, required because its calibration is given as a function of resistance ratios (page 10).

'The bath temperature wa .. s slowly ramped through each calibration temperature (for a stable meniscus) and readings were taken in equal time steps, alternating between the two sensors in the sequence: IPRT, thermometer, IPRT', thermometer, IPRT. In all, four thermometer readings were obtained in each sequence and, for each reading, a correction was determined using the average of the IPHT values obtained immediately before and after. To minimise parallax, the thermometer was mounted vertically and viewed horizontally with a telescope.

The model

The calibration correction for the thermometer (the rneasurand ), C, is:

C

T~) T

f (~~:) - T

(14)

where T~) is the 'true' temperature derived from the calibrated IPRT, its calibration being

in the form of equation ,represented here as f(), and T is the temperature indicated

by the mercury in thermometer. T;) is determined from the ratio of the measured

resistances for the IPHI', Rt, obtained at T, and Ro, at O°C. Both resistances were corrected for the DlVU\!! calibration.

As mentioned above, the IPRT was the one used in section 4 (calibrated in section :3) and, in particular, that its characteristics were those given on page 11. Thus, we can use the sensitivity coefficients for R; and Ro given by equations (G) and (7) (with a change in sign, because the IPRr was used for T~ rather than T). They were evaluated at 100GC. the temperature at which the following analysis was done.

Ct

2.59 -3.Gl

mK/mn

mK/mn at lOO°C .

(15)

and

Notice from equation (14) that the sensitivity coefficient for 7;) is 8C / (Y1~) it is -1.

1 and for T

20

5 CALIBRATION OF A MERCURY IN GLASS THERMOJ'viETER

Uncertainty components

The components evaluated at 100Ge, and listed in Table 5, are as follows.

Reference ternper at ure

I. Calibration of standard IPRT: the IPHT had been calibrated with an expanded uncertainty equivalent to 0.020"(' and the coverage factor was not given. So, let's assume k, 2 and use the conservative value, Vi. ~~ 20 (suggested in [2]). Moreover. the calibration relates directly to 1:, (sensitivity coefficient is 1).

2. Use of the standard: the ice-point resistance Ro, measured for the IPRT during the calibration of the thermometer, was ~2 mn different from that given in the NML report. Since the Nlvll, calibration uncertainty covered an ice point drift of up to 10 mn and the conditions of use (ambient temperature, measuring current, etc.) matched those of the NML calibration, the (additional) uncertainty of 'use' for the IPH.T is nominally zero.

3. Calibration of DT\,'1M: assume the same DMM as that described on page 12 (item 7), with a calibration uncertainty (expanded) of (2 mn + 0.002% of reading) and ki 2.10. As before, the mea .. surcments of R; and R; were considered partially correlated and the net effect of the DMM calibration was calculated as the average of the magnitudes of the effects determined for full and no correlation, as follows. At lOOGe .. Rt 1:39.2 fl, R; == 100 n and the corresponding expanded uncertainties due to calibration are ,1.48rnn and 4.0mn. The respective values of Ci Ujki are 5.90mK and -6.88mK and. thus, the effect of partial correlation is:

u, (0.98 9.(6)/2 5.02rnK

and the estimate is reasonable (v.; 8).

4. Use of DMM: the DMM was used under the same conditions that applied during its calibration, so drift was the only contribution relevant in a 'use' component. In this case, drift since calibration was considered insignificant.

a. Type A uncertainty for the IPRT /DMM: the random error contributed by the IPHT and D1\1M is covered in item 9, below, based on the scatter evident in repeated values of thermometer correction. If, instead, the correction had been obtained as the difference between two means, those for the two sensors, the type A components for both would have been required.

b. Resolution of Dl\1M: the resolution component is 0.5mn and I shall use the sensitivity coefficient for Ro, -O.00:361oC/mn, being the largest of the two. The

uncertainty would be rectangular with k;. = 1. 73 and Vi 100.

The thermometer

7. Stability at DOC: primarily as a check that the thermometer had recovered from 't.empor ary depression' [G]' the thermometer was re-checked at ooe, after the calibration. The measured difference from the value obtained before calibration was O.OI°e. I shall put CUll consider the estimate rough (V'i 2) and assume that the error was drifting linearly throughout the calibration and, thus, 'rectangular'

(k., 1.7:3).

8. Bore/scale non-linearity: even though the request for calibration hac! been at specific temperatures, the pointing marks, measurements at intermediate values were

21

done to a measure of the non-linearity in the bore and/or seale, a significant contribution to error in the USE, of a thermometer. The results are shown in figure 2 of ref. [1]. Because temperature was measured midway between each pair of pointing marks, U, was taken as the departure from linearity, 0.0:35°C, and considered reasonable (Vi 8). A rectangular distribution was assumed (k; = 1.73) because non-linearity errors tend to be proportional to the difference from a pointing mark.

9. Type A uncertainty in mean correction: the scatter in the four values of correction obtained at HJe)"'C was expressed as a standard deviation (O.014°C) and, therefore the standard uncertainty in the mean correction is 0.014/ J4 == 0.007°e. So, U, O.OO7'"'C, and being already a standard uncertainty, the reducing factor is k; 1. Alternatively, could be expressed as a 95% value, to be consistent with the other entries, although, it would then require an extra step. In which case, since there aren 1:3 offreedoIl1,ki :3.18 and wouldbe:3.18xO.007 0.cl22"C. The scatter in correction is clue mainly to the difficulty of 'reading' the meniscus at the chosen times.

10. R.esolution of thermometer: the scale was read to of a scale division

(0.01 Therefore, the resolution uncertainty is ±CJ.(105°C and is rectangular with

k, 1. 7:3 and Vi •... 100.

Other errors

11. Tmnperature difference: the spatial variation in the oil bath at 1000e was assessed as 7mK over the region in which thetrnometers are generally placed for calibration, being the maximum difference between any two temperatures measured in the region. It was assumed from this that there was a 95% chance that the temperature difference between any two probes was within the range O±7 mK, i.e., U, C1.007"C. The value of was considered only reasonable (V"= 8) because the data do not refer specifically to the probes used. The temporal variation in the bath due t.o control action was typically 4 rnK, at a fixed temperature of 100°C. This would have had little effect on the calibration because the fluctuation was rapid relative to the temperature-ramp rate, the probes have similar response times and the error would have been, at least partially, included in the scatter observed for item 9, above.

12. Fitting and interpolation: a smooth, continuous curve should not be fitted to calibration corrections for a Liquid-in-glass thermometer since they tend to have discontinuities at the pointing marks (Figure 2 of ref. [1]). On using a calibration report the user would linearly interpolate to derive corrections at temperatures other than those reported, and interpolation errors would result. Since an effort was made to estimate non-linearity (item 8) such errors are already covered.

1:3. R.ounding: the calibration results will be rounded to o.cn "C when finally presented 0.005°C and ki = 1.7:3, being 'rectangular').

There were no correlation effects.

22

s CALIBRAT']ON OF A MERCURY IN GLASS TI-IERMOl'vIETER

Table 5: An uncertainty for the calibration of a merc ury in glass t hcrrnomctcr. It lists the components of uncertainty at as both (raw) expanded uncertainties and standard uncertainties .in the rneasurand IL,. Included are values of reducing factor sensitivity coefficient c, and number of degrees of freedom u..

Component (ic 1,2, ... Hi)

Reference ternperature

1. calibration of STD PHI'

2. use of STD PH]'

:3. calibration of D1\1lVI

4. use ofD1\11\1

5. mean PRI' readings (type A) G. resolution (half least count)

The thermometer

7. short-term instability at O°C

8. non-linearity

9. mean correction (type A)

10. resolution (half least count) Other errors

11. sensor temperature difference

12. data fit and interpolation 1:3. rounding of final result

* st and ard un cert.ai nty in measur and .

:j: Note I on page 1.

(U;j ki) X leil

Uncertainty calculation

Using the values of Lli and Vi from 'DIble 5, the calculation of expanded uncertainty proceeds as follows:

the combined standard uncertainty, 1tc the number of effective degrees of freedom, Veil the coverage factor, k and, thus. the expanded uncertainty, U Finally, after rounding: U

0.0253°(: 18

2.10 0.0529"C. 0.05°C.

23

6 Calibration of a block calibrator

Briefly, a block (dry-well) calibrator is a temperature-controlled metallic block having at least one vertical well of depth 100] 50 mm and having a control set point that may be adjusted to the nearest 0.] or even closer. It is normally used for calibrating temperature sensors. When calibrating such a device, it is important that a metallic insert standard item supplied with the block) be chosen to gin; minimal clearance from the reference sensor used for its calibration [7]. Otherwise, errors of ~l"C may arise and the calibration uncertainty will be increased an order of magnitude.

The measurement

For this example, I shall assume the use of a calibrated digital thermometer based on an IPHT and fully immersed within the block calibrator. 'I'he probe had a stainless-steel sheath of 6.0mm diameter and, to optimise thermal contact, the block was fitted with an insert of internal diameter 6.5 mm.

The rnodel

The model is very simple--vthe calibration correction (the measurand ). C, is:

where 1~) is the 'true' temperature determined using the digital thermometer, after applying its calibration correction (item 1, below}, and T is the temperature indicated on the block calibrator.

The sensitivity coefficient for

is ec / m;)= 1 and for T is ()C/ DT -1.

Uncertainty components

The following items detail the errors thought to be present in the measurement of temperature at the bottom of the with the block controller displaying a stable value of 200.0°C during the measurement. The components an, listed in Table 6.

Reference ternperature

1. Calibration of digital t.her momet.er: the calibration correction for the thermomcter was -Cl.07 ± O.05°C: at. 200°C, where the value 0.05°C is an expanded

uncertainty and the coverage factor was 2.08, equivalent to lief f 20.

2. Use of thermorneter: regular ice-point checks on the digital thermometer showed a scatter of ±0.02°C during the period of calibration, and U, = 0.02°C is considered a reasonable estimate (IIi == 8, from Note 1 on page 1).

3. Type A uncertainty in mean thermometer reading: the range (R) of scatter in readings was O.03°C over roughly 20 readings (integrations), so the type A

uncertainty for the mean is = mpR = O.OOcFC (page 1).

4. Resolution: the resolution uncertainty of the instrument is U, 0.005°C (a least

count of Cl.Cll and is considered rectangular with ki 1. 7:3 and IIi = 10Cl.

Block calibrator

24

e CtLlBRATION OF A BLOCK CALIBRATOR

Ii Tmnperature difference: trw possible difference in temperature between that at the bottom of the well and that of the 1PHT sensor was estimated by progressively raising the sensor 5 mm at a time to explore the temperature uniformity, and U, was taken as the variation (range) observed over the bottom 20mrn, 0.18°C. Choosing the observed range, rather than the semi-range, is to over-estimate, and is intended to include possi ble differences in later use, due to variations in design of IPRT sensors. This was done because the calibration would be of little use if the block calibrator cannot be used for the calibration/checking of another IPRI (see Note, belowj-v-the mid-point of its sensor being typically 20 mm, but up to :30 mm, from the bottom of its sheath. The estimate is considered reasonable (Vi = 8).

6. Type A uncertainty in block calibrator: during the measurements, the display of the block calibrator remained stable on 200.(JOC (Uice, 0.0).

7. Resolution of calibrator: the resolution uncertainty is Ui.

0.1 and is considered rectangular with k; == 1. 7:3 and Vi.

0.05 "C (least count 100.

Other errors

8. Ac pick-up: the effects of ac pick-up were considered negligible.

9. Fitting and interpolation: when the calibration corrections (1:) - T versus T) were plotted over the range of calibration, a linear interpolation curve was fitted to the data with a scatter of ±0.12"C (U;e= 0.12°C). The estimate is considered reasonable (1/;== 8).

10. Rounding: the results of measurement will be rounded to 0.01 °C when reported

O.005"C and k; 1.73, being 'rectangular').

There were no correlation effects.

NOTE: the calibration result should not be used for sensors whose construction differs in basic design from that of the reference IPRT. For example, it cannot be assumed to apply to thermocouples or to liquid-in-glass thermometers.

Uncertainty calculation

Using the values of IL; and u, from Table 6, the calculation of expanded uncertainty proceeds as follows:

the combined standard uncertainty,uc the number of effective degrees of freedom, Ve] f the coverage factor, k and. thus, the expanded uncertainty, U Finally, after rounding: U

0.115°C: 17

2.1 ] 0.24:3"C. 0.24()C.

REFERENCES

25

Table 6: An uncertainty for the calibration of a block calibrator. It lists the components of uncertainty at 200"C, as both expanded uncert.ainties U, and standard uncertainties in the measurand 11,. Included are values of reducing factor sensitivity coefficient Ci and number

of of freedom 1/, ..

Component (i 1,2, ... 10)

Reference temperature

1. calibration of S'rD PR],

2. use of PHT

:3. mean PHT readings (type A) 4. resolution

Block calibrator

5. temperature diff. in well

6. mean readings (type A)

7. resolution Other errors

8. ac pick-up

9. data fit and interpolation

10. rounding of final result

Component values

k * ("e')

i 1_Li.l

0.05 2.08 0.024 20
0.02 2 0.01 8
0.004 2.1 0.002 19
(J.()05 1.7:3 1 0.003 100
0.18 2 0.09 8
0.0 1 0.0
0.05 1.7:3 1 0.029 100 0.12 0.005

2 1.73

0.06 0.00:3

8 100

1

* standard uncertainty in measur and. 1.1.1: (Ui..! k·1..) X le'i.1

t Not.e 1 p<:lge 1.

References

[1] R E. Bentley. Applying the ISO Gu/ide to the calculation of uncertainty: icmpcroiurc.

Publication TIP PB58 CSIRO, Sydney, Supplement to Monogr. 1: NML Tc,chnology Transfer Series. 2001.

[2] H. E. Bentley. Uncertainisj in measurement: the ISO guide. Monogr. 1: NML Technology Transfer Series, Fifth Edition, 90pp, TIP P1:3:37, CSIRO, Sydney, 2002.

[:3] R. E. Bentley. Thermocouples in temperature measurement. Monogr. 5: NML

Technology Transfer Series, 165pp, CSIHO, Sydney, 2002.

[4] H. E. Bentley. Theory and Practice of Thermoelectric Thermometry, volume :3 of Handbook of Temperature Measurement. Springer-Verlag, Singapore, 1998.

R. E. Bentley. Short-term instabilities in thermocouples containing nickel-based alloys. High Temp.-High PTlSSnrC8, 15:599 GIl, 1 98:i.

[e] R .. E. Bentley, editor.

Handbook of

Resistance E1 Liquid-in- Glass Thermometru, volume 2 of Measurement. Springer-Verlag, Singapore, 1998.

N. E. Kaiser. Temperature calibration in block calibrators. In Fmc. of TEl'vIPJ1;IEKO .9.9: 7th International on Temperature and Thermal Measurements 'in

Industr-) and pages 320-25, Delft, Netherlands, June 1999.

You might also like