Vacuum 100 (2014) 14e17

Contents lists available at ScienceDirect

Vacuum
journal homepage: www.elsevier.com/locate/vacuum

Rapid communication

A standard to test the dynamics of vacuum gauges in the millisecond range

Karl Jousten a, *, Sarantis Pantazis a, Joachim Buthig a, Regine Model a, Martin West b, Jaroslaw Iwicki c
a b c

Physikalisch-Technische Bundesanstalt, Abbestr. 2-12, 10587 Berlin, Germany INFICON AG, Alte Landstrasse 6, 9496 Balzers, Liechtenstein VACOM GmbH, Gabelsberger Str. 9, 07749 Jena, Germany

a r t i c l e i n f o
Article history: Received 28 May 2013 Received in revised form 19 July 2013 Accepted 23 July 2013 Keywords: Vacuum metrology Dynamic pressure Choked ow Response time Vacuum gauge

a b s t r a c t
Vacuum gauges that control fast processes in industrial applications, e.g. load locks, should immediately react to pressure changes. To study the response time of vacuum gauges to rapid pressure changes, a dynamic vacuum standard was developed where the pressure may change from 100 kPa to 100 Pa within 20 ms in a step-wise manner or within longer times up to 1 s in a predictable manner. This is accomplished by a very fast opening gate valve DN40 and exchangeable orices and ducts through which the mass ow rate can be calculated by gas ow simulation software. A simple physical model can be used to approximate the calculations. Experiments have been performed with capacitance diaphragm gauges with improved electronics to give a read-out every 0.7 ms. Preliminary results indicate that their response time is at most 1.7 ms, but may be signicantly less. 2013 Elsevier Ltd. All rights reserved.

In industry there are many fast processes that require rapid changes of pressures in the vacuum regime, e.g. leak testing, CD metallization and coating processes. Typically, vacuum load locks are used for such processes, where the pressure changes from atmospheric to medium vacuum within 1 s or less. The control of such systems is triggered by vacuum gauge readings. Vacuum gauge manufacturers are interested to know how fast their gauges can follow such rapid changes of pressure. To investigate the response time of vacuum gauges for rapid pressure changes, in the framework of the European project EMRP IND12 [1] the Physikalisch-Technische Bundesanstalt (PTB) has developed a dynamic pressure standard, where the pressure changes in a predictable manner from 100 kPa to 100 Pa within one second or less. This is accomplished by expansion of gas via a calculable conductance and a very fast opening DN40 gate valve. Also, a pressure step with a time constant of less than 2 ms (pressure reduction to 37% of the initial value) can be generated when the full opening is used. Fig. 1 shows the schematic of the system. The pressure reduction is realized by gas expansion from a very small volume V1

(0.09472 L) into a much larger volume V2 (185.4 L). During this expansion the valve to the pump system is closed. Either volume may be evacuated or lled independently with gas up to a desired pressure of less than 100 kPa. The expansion is released by a fast opening DN40 gate valve which opens the full diameter within 4.6 ms [2]. The gas within V1 is depleted via an exchangeable orice, 0 (0.03309 L) duct or nozzle with conductance C. The dead volume V1 between this duct and the valve plate contributes to an offset poffs in pressure in V2 after the valve opening, which can be calculated by

poffs p01

0 V1

0 V1 1; 785$104 p01 ; V2

(1)

where V2 was evacuated to a negligible pressure before and p01 is 0 . When the expansion is the initial pressure in both V1 and V1 completed and the gas recovers room temperature, the nal pressure is determined by the expansion ratio

f2

0 V1 V1 0 V1 V1 V2

(2)

and the nal pressure in both chambers will be

* Corresponding author. Tel.: 49 303481 7262. E-mail address: Karl.Jousten@ptb.de (K. Jousten). 0042-207X/$ e see front matter 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.vacuum.2013.07.037

pf f2 p01 6:889$104 p01 :

(3)

K. Jousten et al. / Vacuum 100 (2014) 14e17

15

estimate, choked ow can be expected in the period where 100 kPa > p1(t) > 500 Pa. In this period, the ow rate through the duct, expressed as throughput q* pV , can be estimated by [4]

q* pV Kc1 p1 ;

(4)

where c1 is the mean thermal speed of the gas in V1 and K is a constant

r K

! p* 2 AJ ; 4 p1

(5)

with A as the smallest effective cross sectional area of the duct and J the dimensionless ow function at choked ow conditionp* 2. In a rst approach we will assume isothermal conditions and set

K 0 Kc1 ;

(6)

which has the physical unit of conductance. The change of pressure in the upstream chamber V1 is then determined by the following differential equation

V1
Fig. 1. Diagram of the dynamic vacuum standard. Gas inlets are not shown.

dp 1 K 0 p1 ; dt
We denote the time constant

(7)

When no orice or duct is mounted and the full opening is used, the expansion is the fastest possible with this system. The nal pressure is again given by Eq. (3). The vacuum gauges to be investigated are mounted on the upper chamber V1. They will slightly change the value of the upper volume which has to be corrected. To calculate the pressure p1(t) in the small volume V1 after the valve has being opened, the mass ow rate qm through the conductance C is needed. In the viscous ow regime qm depends on p1(t) and the ratio p2(t)/p1(t), while in the molecular ow regime qm will depend on (p1(t) p2(t)) and the conductance C, which will be pressure independent. To decide which regime of ow is valid, the shape and the dimensions of the conductance element need to be known. One of the ducts used in our apparatus is shown in Fig. 2. At around 100 Pa and 293 K the Knudsen number in the small volume is Kn < 0.06 for all gases. This means that the molecular ow regime will not be reached in the area in front of the duct of Fig. 2; rarefaction effects, however, will play some role in the later stage of the expansion, while the viscous ow regime will dominate the rst part of the expansion starting at 100 kPa. The critical pressure ratio p2/p1 for choked ow for all gases and for short conductance elements (length to radius of about 1) may be estimated to be around 0.3 [3] so that choked ow can be expected for the rst phase of the expansion as long as p2(t)/p1(t) < 0.3. Since, for p01 100 kPa, pf 65 Pa (Eq. (3)) choked ow may prevail until p1(t) 220 Pa. At this low pressure, however, Kn > 0.01 and the ow may exhibit a different characteristic than viscous. As a rough

s1

V1 ; K0

(8)

and solve Eq. (7) with the initial boundary condition p1(t 0) p01 by

p1 t p01 expt =s1 :

(9)

The isothermal condition is no reasonable assumption, as it is well known for fast gas expansions [5] that the gas in the upstream chamber will cool down to quite low temperatures. This has the 0 consequence that c1 and K will decrease and s1 will increase with time. The exponential pressure decrease described by Eq. (9) will therefore be slowed down. To consider this, we assume c1 t to be time dependent replacing Eq. (6) with

K 0 Kc1 t :

(10)

Fig. 2. Orice between upper chamber and lower one.

Inserting this into Eq. (8), we obtain with Eq. (9) an implicit function to characterize the pressure decay in the upstream chamber until p1 reaches 500 Pa. We will see later that this function gives a good tting function for the condition of choked ow of both experimental results and results obtained by numerical simulations. 0 K for the case of the duct shown in Fig. 2 and nitrogen at 20  C has the value of 1.43 L/s which gives s1 69 ms. For comparison, the conductance at molecular ow conditions is 0.78 L/s, calculated from the minimum open area (diameter 3.006 mm) of the duct and a transmission probability of 0.938 obtained by Direct Simulation Monte Carlo (DSMC) method. Numerical simulations are needed to more accurately predict 0 p1(t) and K for the conductance element of Fig. 2 which will be called orice in the following. As a rst approach, a commercially available computational uid dynamic module, ANSYS CFX, was used. Since the package available for us did not include boundary conditions appropriate for the slip regime as well as due to numerical difculties for low downstream pressures <1 kPa, it was used to calculate ows into V2 with initial pressures that ensured that viscous ow prevails throughout the expansion. It could be expected that the results for p1(t) of these simulations do not

16
10 810 610 410 A 210 0 B

K. Jousten et al. / Vacuum 100 (2014) 14e17

0.050

0.100 t in s

0.150

0.200

0.250

Fig. 3. Results for p1(t) for a point in the centre axis of V1, 41.5 mm upstream of the lower orice plane, obtained from ANSYS CFX with isothermal boundary condition (Curve A), the same curve tted by Eq. (11) and adjusted for the actual volume in the experiment (Curve B), and the experimental curve (Curve C).

depend on the downstream pressure as long as choked ow conditions exist. In addition to p1(t) we were interested in the upstream temperature T1(t) in volume V1. Both quantities can also depend on the specic point within V1. The code was used with two boundary conditions on the walls: no heat exchange of the uid with the wall (adiabatic case) and full accommodation of the near surface layer with the wall (isothermal case). For the simulation we assumed an instantaneous opening of the valve by starting the simulation at t 0 with p01 100 kPa and p02 1 kPa; 2.5 kPa; 5 kPa nitrogen. This modelling simplication is justied by the fact that critical ow through the orice is achieved very early and dictates the ow evolution. A minor side effect of this choice is a time shift of about 5 ms compared to experiments. We neglected pf since this is much smaller than p02. The simulations indicate that the pressure eld near the orice and in the dead volume is initially characterized by fast changes after the valve opening stage. Large gradients are observed inside the orice and around it, while the rest of the eld in V1 remains almost homogeneous. The pressure evolution on the axis of symmetry and at the height of the pressure gauges for a downstream pressure of 1 kPa is shown in Fig. 3 for isothermal boundary conditions (Curve A). As expected, the results for p1(t) did not depend

on the downstream pressure for p1(t) > 10 kPa, the differences being less than 2.5%. Comparing the simulation results with adiabatic and isothermal boundary conditions, the pressure decrease p1(t) is only moderately faster with the adiabatic condition, the value being at most a factor 0.75 different from the value for the isothermal condition. In both cases, the pressure eld is quite homogeneous in V1 with relative differences in the 105 range. However, the boundary condition choice is vital for the temperature eld (Fig. 4). In the adiabatic case the temperature eld is very homogeneous in V1, while strong gradients can be observed in the isothermal case. The temperature minimum is about 160 K after 150 ms for isothermal boundary conditions. Since the simulations are quite time consuming, it is helpful to approximate the results by the analytical model described above. This enables one to consider minor changes in the set-up (pressure not exactly 100 kPa, volumes slightly changed due to attached gauges or temperature sensor etc.) to make predictions for the pressure versus time curve for the changed set-up. The analytical model described by Eq. (9) has to be slightly modied to take into account the non-vanishing downstream pressure p02:

p in Pa

p1 t p01 p02 expt =s1 p02 :

(11)

This equation should be valid for p1(t) > 3.3 p02 for our 0 conductance element [3]. Fitting this function with K (Eq. (10)) as t parameter we obtained a curve that could be approximated by a 0 second order polynomial. In the initial phase K was 1.80 L/s approaching 1.46 L/s after about 200 ms. Comparing these values with the simple analytical choked ow model, where we expected 1.43 L/s one can see that the simple physical model is reasonably accurate. Therefore, this model enables us to rescale the CFX results for a different volume V1 according to Eq. (11) to consider the extra volumes of the gauges attached (Curve B). Experiments were performed with capacitance diaphragm gauges (CDGs) of INFICON, for which new electronics were developed by this company. With these, a measurement signal is obtained every 0.7 ms with an output resolution of 21 bit (corresponds to 0.05 Pa for a CDG with fullscale 133 kPa). This electronic was provided for CDGs of 133 kPa and 1.33 kPa fullscale to cover the full range of expansion pressures with good accuracy.

Fig. 4. Cross sectional view of the temperature contours in V1 (left from small orice) and part of the downstream side (right) as received by ANSYS CFX in adiabatic (lower part) and isothermal boundary condition (upper part) at 200 ms after the beginning of expansion. In the adiabatic case the temperature in V1 is 110 K, while in the isothermal case the temperature in front of the orice is 165 K.

K. Jousten et al. / Vacuum 100 (2014) 14e17

1E+05

17

8E+04

6E+04 p in Pa

4E+04

2E+04

0E+00

6 t in ms

10

12

Fig. 5. Measured pressure points versus time of two CDGs, with fullscale of 133 kPa down to 2 kPa and with fullscale of 1.33 kPa for lower pressures, when the full opening of the expansion system is used (no duct mounted). The CDG data are changed near 6 ms. The CDGs deliver data every 0.7 ms. The solid line is an exponential t function to the data and exhibits a time constant of 1.7 ms.

Curve C in Fig. 3 shows the result of the experimental expansion of nitrogen gas with p01 100 kPa and p02 1 kPa. As expected, also the measured Curve C drops with a rate slower than exponential, if the time constant s1 derived from the initial phase is used. The agreement between predicted Curve B and the measured Curve C is fair, the maximum deviation is near the end of the approximation range and 50% with the value from Curve C as the base. This deviation will need more clarication in the future. For temperature measurement in V1 thermocouples with 13 mm and 25 mm wire diameter were used. Investigations on their design and their response time will be reported elsewhere [6]. Preliminary measured temperatures showed a somewhat higher (210 K) and delayed minimum (by 80 ms) than the ones estimated by the simulation with isothermal boundary condition. This point will be further investigated by simulating the temperature response of the thermocouples to the gas temperature changes around it. To check the response time of the developed fast CDGs, an experiment with the full opening (orice removed) was performed. In this case, the time constant of the system is less than 2 ms. Fig. 5 shows the measured curve p1(t) of two fast CDGs, with fullscale of

133 kPa down to 2 kPa and with fullscale of 1.33 kPa for lower pressures. Although the exponential t curve with a time constant of 1.7 ms gives a reasonable approximation, it must be mentioned that on this time scale the opening of the valve which happens in the rst 5 ms plays a signicant role. During the opening phase the open area and therefore the conductance is a rather linear function of time [2]. Thus, it seems that the response time of the gauges may be signicantly less than 1.7 ms. To extend the range of predictable pressure p1(t) to 100 Pa and further improve the accuracy of predicting p1(t) by considering rareed gas ow effects, a hybrid continuum-particle simulation method [7] is under way. Further validation of these simulations will be achieved by comparing calculated and measured temperature evolution. In conclusion, by the standard developed it will be possible to generate predictable pressures changing in time with an expected uncertainty of less than 50% at any time and compare the predicted pressure with measured ones. Due to the homogeneous pressure eld in the upstream volume, attached gauges will be exposed to the same pressure as calculated in the centre. It will be possible to investigate the response time of vacuum gauges in the ms range. The availability of dynamically calibrated vacuum gauges may give new opportunities to manufacturers of vacuum plants to improve time efciency of vacuum systems. Acknowledgement This work was supported by EMRP IND12 project. The EMRP is jointly funded by the EMRP participating countries within EURAMET and the European Union. ANSYS CFX is a registered trademark of ANSYS Incorporation. References
[1] http://www.emrponline.eu/ and http://www.ptb.de/emrp/vacuum.html. [2] Sonderegger K, Dr M, Buthig J, Pantazis S, Jousten K. J Vac Sci Technol A 2013;31:060601. http://dx.doi.org/10.1116/1.4813836. [3] Varoutis S, Valougeorgis D, Sharipov F. J Vac Sci Technol A 2009;27(6):1377e91. [4] Jousten K, editor. Handbook of vacuum technology. Weinheim: Wiley-VCH; 2008. p. 93e7. [5] Jousten K. Vacuum 1994;45:1205e8. [6] Pantazis S, Buthig J, Jousten K. 2013, to be published. [7] Pantazis S. AIP Conference Proceedings 2012;1501:451e6. http://dx.doi.org/ 10.1063/1.4769567.