Вы находитесь на странице: 1из 7

VAC=1902=Susan=Venkatachala=BG

Vacuum 60 (2001) 153}159

Some considerations on handling the calibration results


of capacitance membrane gauges
Mercede Bergoglio*, Anita Calcatelli
Istituto di Metrologia Gustavo, Colonnetti del Consiglio Nazionale delle Ricerche (IMGC-CNR), strada delle Cacce 73,
10135 Torino, Italy

Abstract

Capacitance diaphragm gauges (CDG) are widely used also as transfer instruments; consequently, they need to be
referred frequently to primary systems mostly in the (1000}0.1) Pa range where no other accurate gauges are available.
They must be characterized individually since a general rule on their metrological behaviour is not possible. Various
CDGs are calibrated at IMGC for the Italian Accreditation Service. The aim of the present paper is to demonstrate that,
for each gauge and after the "rst calibrations, it is possible to arrive at an equation which represents the calibration
coe$cient as a function of pressure and is based on a reduced number of experimental points; as a consequence the time
for a complete calibration procedure is cut down. One of the four gauges we examined, for which many calibration curves
were available, was considered for a complete evaluation; then the evaluation model was applied to other
gauges.  2001 Elsevier Science Ltd. All rights reserved.

1. Introduction connected with a growing demand for well-de"ned


and well-characterized transfer standards.
As a general rule, a calibration certi"cate pro- Since systems which are primary references for
vides the means for correcting gauge readings to get vacuum measurements cannot be moved to per-
the best information on the quantity being mea- form in-situ calibrations, there is a general need for
sured. In addition, a certi"cate should make the gauges which are easier to handle but still as accu-
users con"dent about the gauge-calibration system rate as possible. For vacuum measurements, the
reliability. This makes the laboratory issuing the following secondary instruments have been used or
certi"cate considerably responsibile for: are considered as transfer gauges:
E calibration procedure E capacitance membrane gauges (CDG) [1,2]
E uncertainty evaluation method E spinning rotor gauges (SRG) [3,4]
E calibration data treatment. E ionization gauges (IG) [5,6].
These three items are becoming more and more In our experience, the more precise and more fre-
important with the increasing need for transfer quently required measurements are in the pressure
gauges by accredited or non-accredited laborator- range of CDGs, which are commercially available
ies. The reason for the present paper is, in fact, with various full scales (from 10 to 10 Pa) and
with di!erent accuracies. The most accurate CDGs
are considered as transfer gauges and are the sub-
* Corresponding author. ject of the present paper. For the best of these

0042-207X/01/$ - see front matter  2001 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 4 2 - 2 0 7 X ( 0 0 ) 0 0 3 6 9 - 9
Vac=1902=SK=VVC

154 M. Bergoglio, A. Calcatelli / Vacuum 60 (2001) 153}159

gauges the manufacturers state an accuracy value considered with the aim of:
in the range of 5;10\ [7] and even readings as
low as 2;10\ for some of them. However, it is not E trying to "nd out a general behaviour at least on
clear how the uncertainty had been evaluated [8]; the same gauge, if not subjected to inappropriate
probably long-term instability was not considered, handling;
since the experience acquired in our own [9] and E "nding the "tting curve ensuring a continuity of
other primary laboratories [1] has shown that cal- gauge reading corrections and therefore gauge
ibration data may change as much as several parts reliability;
in 10 and that it is really di$cult to establish E trying to reduce the number of calibration points
a common rule not only for gauges in general but without signi"cantly decreasing accuracy.
also for the same gauge, metrological behaviour is
dependent on the individual gauge history.
Most of the papers on CDGs deals with the 2. Experimental procedure
thermal transpiration e!ect introducing factors re-
lated to the mass of the gas used [9,10] and con- Three gauges (B,C,D) of the four CDGs con-
sider the non-linear response of sensors which are sidered belong to the same manufacturer, while
based on the deformation of a thin membrane un- only one (A) was produced by another factory.
der pressure, so that a linear interpolation of their They have the general characteristics shown in
output over the full pressure range does not repres- Table 1.
ent their real behaviour. On this basis and in con- All the gauges were operated with their own
nection with the cost of the gauges and of their control and measuring units and were continuously
calibration, we have focused our attention on the temperature-controlled by their heating system.
treatment of the calibration data that may intro- The normally adopted procedure consists of cali-
duce additional uncertainty or may even be sources brating gauges in nitrogen atmosphere against the
of errors. IMGC static expansion system [11], whose work-
Calibrations of CDGs are time consuming irre- ing ranges and related uncertainty values are
spective of whether they are under temperature quoted in Table 2.
control or not. Several pressure values must be Calibrations are usually repeated on three di!er-
generated, to enable us to give to a gauge a calib- ent days and every other year, or more frequently.
ration factor which is a function of the pressure.
At IMGC, several gauges from di!erent manu- Table 2
facturers are frequently calibrated; many calib- Pressure ranges and standard uncertainty for the IMGC static
system
ration data concerning four of them have been
collected over the years and the data treated by Pressure range (Pa) Standard uncertainty
"tting the experimental values to enable users to
easily correct the readings of their gauges. 0.1}1 2.0;10\;p#1.5;10\
In the present paper four gauges (10 and 10 Pa 1}10 6.8;10\;p#1.4;10\
10}1000 8.3;10\;p#7.0;10\
full scale) from two di!erent manufacturers are

Table 1
Types of the considered gauges

Gauge Full scale (Pa) Type

A 1000 Di!erential used as absolute


B 100 Di!erential used as absolute
C 100 Di!erential used as absolute
D 1000 Absolute
VAC=1902=SK=VVC

M. Bergoglio, A. Calcatelli / Vacuum 60 (2001) 153}159 155

Fig. 1. Calibration results for the four gauges over three years at various pressure values: calibration factor vs. pressure: (a) gauges
A and D , (b) gauges B * and C .

At least "ve pressure values are generated in each model is included in the "nal uncertainty evalu-
pressure decade and the corresponding gauge read- ation.
ings are collected. At pressures of 0.1, 1 Pa and at
full scale the calibrations are repeated ten times to
evaluate short-term experimental repeatability. To 3. Results and data treatment
represent the correction to be applied to the gauge
by users, the calibration factor is calculated Fig. 1 shows all the experimental data for the
(F"gauge reading/generated pressure), then four considered gauges calibrated on di!erent days
F values are "tted as a function of the pressure; the and years; in addition, the best curves of F versus
uncertainty component due to the mathematical pressure for each gauge are plotted.
Vac=1902=SK=VVC

156 M. Bergoglio, A. Calcatelli / Vacuum 60 (2001) 153}159

Fig. 2. Transducer A: "tting curves compared with the experimental data: (a) over each pressure decade (polynomials),
(b) over total range from Eq. (1), (c) over total range from Eq. (2). at the beginning, * one year later, two years
later.
VAC=1902=SK=VVC

M. Bergoglio, A. Calcatelli / Vacuum 60 (2001) 153}159 157

As regards gauge (A), calibrated at di!erent times on the various calibration cycles performed in dif-
since 1996, various "tting procedures were applied ferent years are shown in Fig. 2 and are compared
starting from our usual one, which was based on with experimental data. The curves show that "t-
treating the data over each pressure decade separ- ting over separate decades (Fig. 2a) is not appropri-
ately by a simple polynomial curve (F"a#bx# ate to correctly represent all the experimental data
cx). This procedure proved to be incorrect due to unless the number of calibration points is consider-
discontinuities at the `interfacea between decades, ably increased. For type (1) "tting, ratios of various
which may not be negligible and are as high as polynomials were considered from the second to
a few percent, as will be shown later. In addition, the "fth or sixth order. The low-pressure limits of
the number of the calibration pressure values ought the ratios of the lower-order polynomials do not
to be considerably high, if reliable "tting is to be tend to the previously de"ned constant values.
obtained. Two other "tting methods were con- Hence a high-degree polynomial ratio must be
sidered by starting from the experimental values chosen, the number of the exponents depending on
F as functions of the pressure: the overall pressure range considered.
(1) a polynomial ratio such as Fig. 2b refers to the curves obtained from Eq. (1)
where up to the sixth degree even polynomials were
A#Bx#Cx#Dx#2 chosen as the number of available experimental
F" , (1) values was su$cient high; the polynomial ratio
1#Ax#Bx#Cx#2
represents the best "ttings of this type, on the basis
of the best standard deviation values. The two types
(2) a continuous function such as
of "tting curves (Figs. 2b and 2c) apparently show
a comparable behaviour, other considerations must
a!d be added to decide which of them better represents
F" #d. (2)
1#(x/c)@ the experimental situation. The polynomial ratio
curves (Fig. 2b) do not always approach the ideal
In both equations x denotes the gauge reading. behaviour represented by the curves in Fig. 1a,
Under ideal conditions, at very low pressure whereas curves from Eq. (2) follow (Fig. 2c) con-
(molecular regime) F tends to be equal to the tinuously the experimental ideal curve.
square root of the ratio of the gauge temperature to Since high-precision mechanical sensors are fre-
that of the calibration chamber and, at full scale, it quently employed as transfer gauges, the "tting
tends to one. These two limits should coincide with curves given to the users (i.e. accredited laborator-
the limits of the "tting functions which are, respec- ies) must be the best representation of a gauge
tively, the constants A (x"0) and the ratio of the behaviour, and consequently, any pressure value
coe$cients having highest power exponents should be evaluated with acceptable uncertainty;
(x"R) for curve (1) and a (x"0) and d (x"R) on the other hand, a primary laboratory needs to
for curve (2). A possible di!erence between the reduce the number of calibration points to decrease
values of the two sets of constants and their ideal the calibration time. In addition, a good choice of
condition values is due to the initial calibration the "tting curve o!ers the possibility of controlling
when the gauge scale was made. the overall reliability of experimental data; in fact,
Fitting and the related controls were (and are) some calibration cycles show points which are evi-
performed by means of commercially available dently outside the regular curves, such points can-
software; obviously, the "tting procedure continues not be refused a priori without checking the whole
until the best value of the correlation coe$cient is curve and also considering the previous history of
obtained. Each calibration cycle was treated separ- the gauge.
ately with the aim of "nding out possible discrepan- A reduction of the number of experimental
cies among the various cycles. This method was points was considered only for type (2) curves; as
also a check of the experimental data. All the "tting shown in Fig. 3, a reduction of the experimental
curves concerning the same gauge which are based points by about 50% gives an approximately
Vac=1902=SK=VVC

158 M. Bergoglio, A. Calcatelli / Vacuum 60 (2001) 153}159

Fig. 3. In#uence of the number of experimental points on the "tting curves, Eq. (2): (*) all experimental points, (- - - - -) reduction of
experimental points by about 50%, (***) further reduction by 45%.

Table 3
Main characteristics of the considered CDGs

Gauge Pressure Short-term repeatability Long-term repeatability Fitting Combined uncertainty


range (Pa) model

0.1}1 2.0;10\}2.2;10\ 4.0;10\}3.6;10\ 4.6;10\}4.2;10\


A 1}10 2.2;10\}2.0;10\ 3.6;10\}2.0;10\ 1.0;10\ 4.2;10\}2.4;10\
10}1000 2.0;10\}4.5;10\ 2.0;10\}1.8;10\ 2.4;10\}2.2;10\
B 0.1}1 2.0;10\}2.0;10\ 1.5;10\}1.0;10\ 6.0;10\ 1.5;10\}1.0;10\
1}100 2.0;10\}2.0;10\ 1.0;10\}2.0;10\ 1.0;10\}2.0;10\
C 0.1}1 5.0;10\}2.8;10\ 6.0;10\}6.0;10\ 1.0;10\ 6.4;10\}6.4;10\
1}100 2.8;10\}7.6;10\ 6.0;10\}1.2;10\ 6.4;10\}1.7;10\
0.1}1 5.7;10\}2.0;10\ 6.6;10\}2.7;10\ 7.0;10\}4.0;10\
D 1}10 2.0;10\}2.0;10\ 2.7;10\}2.5;10\ 2.0;10\ 4.0;10\}3.0;10\
10}1000 2.0;10\}8.0;10\ 2.5;10\}1.0;10\ 3.0;10\}2.0;10\

3;10\ shift of the curve (2) from the experimental On the basis of the foregoing considerations,
data. With a further 45% reduction the shift is still of function (2) seems, at present, to be one of the best
magnitude a few parts in 10. In addition, if the ex- of our experimental data "tting ways and able to
perimental pressure levels are appropriately chosen, guarantee the continuity of the calculated function;
the shift of the curves from Eq. (2) can signi"cantly in addition, it does not need sophisticated calcu-
be reduced without departing too much from the lation for the users. This type of curve "ts accurate-
ideal behaviour. The reduction of the calibration ly experimental data when both pascal and millibar
points may be considered valid only for gauges pressure units are used.
which were never subjected to any kind of shock and Our experience on gauge A enabled us to treat
whose "rst calibration curves are well known. all the other gauges in the same way; a complete
VAC=1902=SK=VVC

M. Bergoglio, A. Calcatelli / Vacuum 60 (2001) 153}159 159

metrological characterization was thus per- not more than 1;10\ by a carefully chosen inter-
formed. polation curve. A good choice of the interpolation
curve will increase the reliability of the pressure
values that are calculated for readings not corre-
4. Metrological characteristics sponding exactly to the experimental points even if
the number of pressure points is not reduced.
Once the gauges are calibrated and their calib- To conclude, on the one hand, the use of a curve
ration factors are de"ned, the best curves "tting the makes it easier for the users to treat the data and on
experimental data and their metrological charac- the other hand, by its use, it is possible to evaluate
teristics may be de"ned, including the uncertainty transducer repeatability which could be di$cult to
values [12] which take all the factors into account: calculate only from the experimental points, as it is
resolution, short- and long-term repeatability, stan- not possible to generate several times the same
dard deviation of the "tting curve and uncertainty pressure in primary devices such as static expan-
of a primary device. ders.
The main characteristics are summarized in
Table 3.
Hysteresis was not evaluated since, in the present References
case, the gauges were calibrated only with ascend- [1] Hyland RW, Sha!er RL. J Vac Sci Technol 1991;A9(6):2843.
ing pressure which corresponds to the working [2] Bergoglio M, Calcatelli A, Rumiano G. Automazione
conditions of a static expander. The hysteresis com- e Strumentazione 1997:129.
ponent is therefore not included in the combined [3] Messer G, Jitschin W, Rubet L, Calcatelli A, Redgrave FJ,
uncertainties. Keprt A, Weinan F, Sharma JKN, Dittmann S, Ono M.
Metrologia 1989;26:183.
[4] Hirata M, Bergoglio M, Calcatelli A. Metrologia
1997;34:421.
5. Conclusions [5] Poulter KF, Calcatelli A, Choumo! PS, Iapte! B, Messere
G, Grosse G. J Vac Sci Technol 1980;17:679.
At present several data are available on CDGs [6] Jousten K, Filippelli AR, Tilford CR, Redgrave FJ. J Vac
and their behaviour is known even if a general rule Sci Technol 1997;A15(4):2395.
[7] Sullivan JJ. J Vac Sci Technol 1985;A3:1721.
cannot be de"ned. Their stability is considerably [8] Jitschin WJ. Vac Sci Technol 1990;A8(2):948.
far from that stated by the manufacturers. Calib- [9] Poulter KF, Rodgers MJ, Thompson TJ, Perkin MP.
ration curves are needed to employ such instru- Vacuum 1983;33:6311.
ments as transfer gauges over the full range in [10] Siu MCI. J Vac Sci Technol 1973;A3:1721.
which they are calibrated without discontinuities. [11] Bergoglio M, Calcatelli A. Proceedings of the Imeko,
vol. IXA, 1997. p. 140.
The number of pressure calibration levels in a pri- [12] Guide to the expression of uncertainty in measurement.
mary laboratory may be considerably reduced International Organization for Standardization, Gene`ve,
(even 50%) by increasing the uncertainty value by 1993, ISBN 92-67-10188-9.

Вам также может понравиться