NIST Sp250 63 PVTT GasFlowCal

NIST Measurement Services:
Gas Flowmeter Calibrations with the 34 L and 677 L PVTt

Standards

NIST Special Publication 250-63

John D. Wright, Aaron N. Johnson, Michael R. Moldover, and Gina M. Kline

U. S. Department of Commerce
Technology Administration
National Institute of Standards and Technology
Table of Contents

Gas Flowmeter Calibrations with the 34 L and 677 L PVTt Standards
Abstract ...................................................................................................................... 1
1 Introduction............................................................................................................ 2
2 Description of Measurement Services ................................................................... 4
3 Procedures for Submitting a Flowmeter for Calibration........................................ 7
4 Pressure, Volume, Temperature, and time (PVTt) Standards ................................ 7
5 Design and Operation of the PVTt Standard........................................................ 11
5.1 Average Temperature of the Collected Gas ...............................................11
5.2 Mass Change in the Inventory Volume.......................................................16
5.3 Measurement of the Tank and Inventory Volumes .....................................24
6 Uncertainty Analysis of the 34 L and 677 L PVTt Flow Standards..................... 27
6.1 Techniques for Uncertainty Analysis .........................................................28
6.2 Mass Flow Uncertainty ..............................................................................31
6.3 Pressure ......................................................................................................33
6.4 Temperature................................................................................................36
6.5 Compressibility, Molecular Weight, and Gas Constant .............................39
6.6 Density........................................................................................................42
6.7 Collection Time..........................................................................................43
6.8 Volume of the 677 Liter Collection Tank (Gravimetric Method)..............44
6.9 Volume of the 34 Liter Collection Tank (Volume Expansion Method) ....48
6.10 Inventory Volume.....................................................................................50
7 Experimental Verification of the Uncertainty of the 34 L and 677 L PVTt Gas
Flow Standards......................................................................................................... 55
7.1 Comparison of the 34 L and 677 L Flow Standards...................................56
7.2 Multiple Diversions in the 677 L Flow Standard.......................................59
8 Summary.............................................................................................................. 60
9 References............................................................................................................ 62
Appendix: Sample Calibration Report

i

Abstract

This document provides a description of the 34 L and 677 L pressure, volume,
temperature, and time (PVTt) primary gas flow standards operated by the National
Institute of Standards and Technology (NIST) Fluid Flow Group. These facilities are
used to provide gas flowmeter calibration services as reported in NIST Special
Publication 250 [Marshall 1998] for Test Numbers 18010C and 18050C. The PVTt
standard uses two collection tanks and two diverter valve systems to perform gas
flowmeter calibrations between 1 L/min and 2000 L/min (the reference temperature and
pressure conditions are 293.15 K and 101.325 kPa). The standard measures flow by
collecting gas in a tank of known volume during a measured time interval. The
uncertainty of the flow measurement is between 0.02 % and 0.05 % (k = 2 or
approximately 95 % confidence level), depending on the gas used and where in the flow
range the facilities are being used.

We provide an overview of the gas flow calibration service and the procedures for
customers to submit their flowmeters to NIST for calibration. We describe the significant
and novel features of the standard and analyze its uncertainty. The gas collection tanks
have a small diameter and are immersed in a uniform, stable, thermostatted water bath.
The collected gas achieves thermal equilibrium rapidly, and the uncertainty of the
average gas temperature is only 7 mK (22 10
-6
T). A novel operating method leads to
essentially zero mass change in and very low uncertainty contributions from the
inventory volume.

Gravimetric and volume expansion techniques were used to determine the tank and the
inventory volumes. Gravimetric determinations of collection tank volume made with
nitrogen and argon agree with a standard deviation of 16 10
-6
V
T
. The largest source of
uncertainty in the flow measurement is drift of the pressure sensor over time, which
contributes a relative standard uncertainty of 60 10
-6
to the determinations of the
volumes of the collection tanks and to the flow measurements. Throughout the range 3
L/min to 110 L/min, flows were measured independently using the 34 L and the 677 L
collection systems, and the two systems agreed within a relative difference of 150 10
-6
.
1

Double diversions were used to evaluate the 677 L system over a range of 300 L/min to
1600 L/min, and the relative differences between single and double diversions were less
than 75 10
-6
.

Key words: calibration, correlated uncertainty, flow, flowmeter, gas flow standard,
inventory volume, PVTt standard, mass cancellation, meter, sensor response, uncertainty.

1 Introduction

Calibrations of gas flowmeters are performed with primary standards [ISO 1993] that are
based on measurements of more fundamental quantities, such as length, mass, and time.
Primary flow calibrations are accomplished by collecting a measured mass or volume of
the flowing fluid over a measured time interval under approximately steady state
conditions of flow, pressure, and temperature at the meter under test. The flow measured
by the primary standard is computed along with the average of the flow indicated by the
meter under test during the collection interval. All of the quantities measured in
connection with the calibration standard (i.e., temperature, pressure, time, etc.) are
traceable to established national standards. A gas calibration facility consists of a fluid
source (e.g., an air compressor or compressed gas cylinders), a test section that provides
stable thermodynamic conditions and a fully developed flow profile, and a system for
diverting and timing the collection of a quantity of the fluid.

NIST offers calibrations of gas flowmeters in order to provide traceability to flowmeter
manufacturers, secondary flow calibration laboratories, and flowmeter users. For a
calibration fee, NIST calibrates a customers flowmeter and delivers a calibration report
that documents the calibration procedure, the calibration results, and their uncertainty.
The flowmeter and its calibration results may be used in different ways by the customer.
The flowmeter is often used as a transfer standard to perform a comparison of the
customers primary standards to the NIST primary standards so that the customer can
establish traceability, validate their uncertainty analysis, and demonstrate proficiency.
2

Customers with no primary standards use their NIST calibrated flowmeters as working
standards or reference standards in their laboratory to calibrate other flowmeters.

The Fluid Flow Group of the Process Measurements Division (part of the Chemical
Science and Technology Laboratory) at NIST provides gas flow calibration services over
a range of 1 L/min to 77600 L/min.
Table 1 presents the flow ranges covered by the

primary gas flow standards in the Fluid Flow Group. Flows from 900 L/min to 77600
L/min can be measured with a 26 m
3
PVTt standard that was built in the late 1960s and
has been upgraded several times [Olsen and Baumgarten 1971, Johnson et al. 2003].
Flows of 1 L/min or less can be calibrated by the NIST Pressure and Vacuum Group.

Table 1. Primary gas flow calibration capabilities within the NIST Fluid Flow Group.
Flow
Standard
Flow Range
(L/min)

Gas
Pressure Range
(kPa)
Uncertainty
(k = 2) (%)
1 - 100 N
2
100 7000 0.03 0.04
1 - 100 Air 100 - 1700 0.05
1 - 100 CO
2
100 - 4000 0.05
1 - 100 Ar 100 - 7000 0.05

34 L PVTt
1 - 100 He 100 - 7000 0.05
10 - 150 N
2
100 - 800 0.02 0.03 677 L PVTt
10 -2000 Air 100 - 1700 0.05
26 m
3
PVTt 860 - 77600 Air 100 - 800 0.13

This document describes the theory, methods of operation, and uncertainty of the 34 L
and 677 L PVTt primary standards that cover the 1 L/min to 2000 L/min flow range.
From the late 1960s to 2002, this flow range was covered by a set of three mercury-
sealed piston provers and two bell provers [Wright and Mattingly 1998]. The 34 L and
677 L PVTt standards were completed in 2002, and for a period of one year we used both

Reference conditions of 293.15 K and 101.325 kPa are used throughout this document for volumetric
flows.
3

the old and the new primary standards for customer calibrations. In 2003, the new
primary standards and their uncertainty statements were sufficiently validated [Wright
2003A], and we began using the new standards alone to calibrate customers flowmeters.

The PVTt standard is ideally suited for the calibration of critical flow venturis (CFVs)
since they provide pressure isolation and provide a well defined boundary for the
inventory volume. A set of NIST CFVs, calibrated by the PVTt standards are used as
working standards in another calibration facility called the Working Gas Flow Standard
(WGFS). The WGFS provides calibrations, particularly for laminar flowmeters, in which
the reference flow is measured with uncertainty of 0.1% or less. The methods and
uncertainty analysis for the WGFS are covered in a separate document.

2 Description of Measurement Services

Customers should consult the web address www.nist.gov/fluid_flow to find the most
current information regarding our calibration services, calibration fees, technical contacts,
and flowmeter submittal procedures.

NIST uses the PVTt primary standards described herein to provide gas flowmeter
calibrations for flows between 1 L/min and 2000 L/min. The facility has been used at
flows as low as 0.025 L/min, but calibrations below 1 L/min should be discussed with the
technical contacts before a flowmeter is submitted for such low flows.

The gases available for calibrations in the 34 L PVTt standard are dry air, nitrogen,
carbon dioxide, argon, and helium. The source of air, at pressures up to 1700 kPa, is an
oil-free reciprocating compressor and a refrigeration drier. The dew point temperature of
the dried air is 250 K so the mole fraction of water in the air is 0.08 %. Nitrogen (at
pressures up to 800 kPa and purity of 99.998%) is supplied by liquid nitrogen dewars.
Higher pressures of nitrogen as well as argon, carbon dioxide, and helium gas can be
supplied from compressed gas cylinders. Other non-toxic, non-corrosive gases can be
accommodated upon customer request. While other gases are certainly feasible in the 677
4

L PVTt standard, gases are practically limited to air from the compressor and nitrogen
from dewars since a very large number of gas cylinders would be necessary to provide
gas at 2000 L/min. The gas temperatures are nominally room temperature.

Readily available fittings for the installation of flowmeters in the 34 L and 677 L PVTt
standards are Swagelok
(1/8 to 1), A/N 37 degree flare (1/4 to 1), national pipe

thread or NPT (1/8 to 3), VCR (1/4 and 1/2), and VCO (1/2 and 1).

Meters can be tested if the flow range, gas, and piping connections are suitable, and if the
system to be tested has precision appropriate for calibration with the NIST flow
measurement uncertainty. The vast majority of flowmeters calibrated in the gas flow
calibration service are either critical flow venturis (critical nozzles) or laminar
flowmeters since these are presently regarded as the best candidates for transfer and
working standards by the gas flow metrology community [Wright 2003B]. Other meter
types that we have tested include positive displacement meters, roots meters, rotary gas
meters, thermal mass flowmeters, and turbine meters. Meter types with calibration
instability significantly larger than the primary standard uncertainty should not be
calibrated with the NIST standards for economic reasons. For example, a rotameter for
which the float position is read by the operators eye normally cannot be read with
precison any better than 1 %. It is not practical to pay thousands of dollars to obtain
0.05 % or less uncertainty flow data from NIST for such a flowmeter when 0.5% data is
perfectly adequate and available from other laboratories at significantly lower cost.

A normal flow calibration performed by the NIST Fluid Flow Group consists of five
flows spread over the range of the flowmeter. For a CFV, typical calibration set points
are at 200 kPa, 300 kPa, 400 kPa, 500 kPa, and 600 kPa. A laminar flowmeter is
normally calibrated at 10 %, 25 %, 50 %, 75 %, and 100 % of the meter full scale. At
each of these flow set points, three (or more) flow measurements are made with the PVTt

Certain commercial equipment, instruments, or materials are identified in this paper to foster
understanding. Such identification does not imply recommendation or endorsement by the National
Institute of Standards and Technology, nor does it imply that the materials or equipment identified are
necessarily the best available for the purpose.
5

standard. The same set point flows are tested on a second occasion, but the flows are
tested in decreasing order instead of the increasing order of the first set. Therefore, the
final data set consists of six (or more) primary flow measurements made at five flow set
points, i.e., 30 individual flow measurements. The sets of three measurements can be
used to assess repeatability, while the sets of six can be used to assess reproducibility. For
further explanation, see the sample calibration report that is included in this document as
an appendix. Variations on the number of flow set points, spacing of the set points, and
the number of repeated measurements can be discussed with the NIST technical contacts.
However, for data quality assurance reasons, we rarely will conduct calibrations
involving fewer than three flow set points and two sets of three flow measurements at
each set point.

The Fluid Flow Group prefers to present flowmeter calibration results in a dimensionless
format that takes into account the physical model for the flowmeter type [Wright 1998].
The dimensionless approach facilitates accurate flow measurements by the flowmeter
user even when the conditions of usage (gas type, temperature, pressure) differ from the
conditions during calibration. Hence for a CFV calibration, the calibration report will
present Reynolds number and discharge coefficient and for a laminar flowmeter, a report
presents the viscosity coefficient and the flow coefficient. In order to calculate the
uncertainty of these flowmeter calibration factors, we must know the uncertainty of the
standard flow measurement as well as the uncertainty of the instrumentation associated
with the meter under test (normally absolute pressure, differential pressure, and
temperature instrumentation). We prefer to connect our own instrumentation
(temperature, pressure, etc.) to the meter under test since they have established
uncertainty values based on calibration records that we would not have for the customers
instrumentation. In some cases, it is impractical to install our own instrumentation on the
meter under test and the meter under test outputs flow. In these cases, we provide a table
of flow indicated by the meter under test, flow measured by the NIST standard, and the
uncertainty of the NIST flow value.

6

3 Procedures for Submitting a Flowmeter for Calibration

The Fluid Flow Group follows the policies and procedures described in Chapters 1, 2,
and 3 of the NIST Calibration Services Users Guide [Marshall 1998]. These chapters can
be found on the internet at the following addresses:
http://ts.nist.gov/ts/htdocs/230/233/calibrations/Policies/policy.htm,
http://ts.nist.gov/ts/htdocs/230/233/calibrations/Policies/domestic.htm, and
http://ts.nist.gov/ts/htdocs/230/233/calibrations/Policies/foreign.htm.

Chapter 2 gives instructions for ordering a calibration for domestic customers and has the
sub-headings: A.) Customer Inquiries, B.) Pre-arrangements and Scheduling, C.)
Purchase Orders, D.) Shipping, Insurance, and Risk of Loss, E.) Turnaround Time, and
F.) Customer Checklist. Chapter 3 gives special instructions for foreign customers. The
web address www.nist.gov/fluid_flow has information more specific to the gas flow
calibration service, including the technical contacts in the Fluid Flow Group, fee
estimates, and turnaround times.

4 Pressure, Volume, Temperature, and time (PVTt) Standards

PVTt systems have been used as primary gas flow standards by NIST and other
laboratories for more than 30 years [Olsen and Baumgarten 1971, Kegel 1995, Ishibashi
et al. 1985, Wright 2001]. The PVTt systems at NIST consist of a flow source, valves for
diverting the flow, a collection tank, a vacuum pump, pressure and temperature sensors,
and a critical flow venturi (CFV) which isolates the meter under test from the pressure
variations in the downstream piping and tank (see Fig. 1).

7

Figure 1. Arrangement of equipment in a PVTt system.

The process of making a PVTt flow measurement normally entails the following steps:
1. Close the tank valve, open the bypass valve, and establish a stable flow through the
CFV.
2. Evacuate the collection tank volume (V
T
) with the vacuum pump.
3. Wait for pressure and temperature conditions in the tank to stabilize and then acquire
initial values for the tank ( and
i
T
P
i
T
T
). These values will be used to calculate the
initial density and the initial mass of gas in the tank ( ).
i
T
m
4. Close the bypass valve and, during the dead-end time when both the bypass and
tank valves are fully closed, choose a start time ( ). At the same time, acquire the
initial pressure and temperature in the inventory volume ( and T ). These values
will be used along with the equation of state for the gas and the inventory volume (V
i
t
m
i
I
P
i
I
I
)
to obtain an initial mass in the inventory volume ( ). Shortly after the bypass valve
is fully closed, open the tank valve.
i
I
5. Wait for the tank to fill to a prescribed upper pressure and then close the tank valve
and choose the stop time ( ) during the dead-end time. At the same time, acquire
the pressure and temperature in the inventory volume ( and
T
) and hence the
final mass in the inventory. Open the bypass valve.
f
t
f
I
P
f
I
8

6. Wait for stability and then acquire and and hence .
f
T
P
f
T
T
f
T
m

By writing a mass balance for the control volume composed of the inventory and tank
volumes (see the volume defined by the dashed line in Fig. 1), one can derive an equation
for the average mass flow during the collection time:

i f
i f i f
) ( ) (
t t
m m m m
m
I I T T
+
= &
, (1)

or, neglecting the volume changes between the initial and final conditions:

( ) ( )
i f
i f i f
t t
V V
m
I I I T T T
+
=

& , (2)

where is the gas density determined via a real gas equation of state, ( ) ZRT PM = ,
where M is the gas molecular weight, R is the universal gas constant, and Z is the
compressibility factor.

The start and stop times can be chosen at any point during the dead-end time as long as
the inventory conditions are measured coincidentally. Why is this true? Implicit in the
PVTt basis equation (Eq. 2) are two requirements: 1) the measurement of the initial and
final densities must be coincident with the measurement of the start and stop times and 2)
there must not be any other sources or sinks of mass flow to the control volume. The
second condition is met for the entire time that the bypass valve is fully closed, including
the start and stop dead-end times. It is not necessary that the initial and final
determinations of the mass in the collection tank be done coincidentally with the start and
stop times because the tank is free of leaks and it is advantageous to measure these mass
values when the tank conditions have reached equilibrium. The freedom to choose the
start and stop times from within the dead-end time intervals allows one to choose times
where the initial and final inventory densities match, giving essentially zero mass change
9

in the inventory volume ( ) and extremely good cancellation of certain correlated
inventory uncertainties.
I
m

The PVTt measurement process can be performed in a blow down mode also, where
initial and final values of the mass in a tank that is the source of flow instead of the
collector of flow are utilized. Such a system has the advantage that a small compressor
can be used to charge a large pressure vessel over a long period of time allowing one to
achieve very large flows relatively inexpensively. The blow down method has the
disadvantage that it is more difficult to maintain stable pressure and temperature
conditions at the meter under test since the high-side pressure of the flow control
throttling process is changing continuously as the tank discharges.

The bypass and tank valves can be operated with valve overlap, i.e. where one valve
begins to open before the other is fully closed, or with zero overlap, where one is
completely closed before the other begins to open (as described above) [Harris 1980].
With zero overlap, there is no question about lost or extra mass occurring during the
diversion. For instance, if the tank is at an initial pressure less than atmospheric, when
both valves are partially open, flow can enter the tank from the room instead of through
the meter under test. Zero overlap avoids this possibility. For a zero overlap system it is
important that any valve design be fast acting. There is a short period of time during the
actuation of the diverter valves during which both valves are closed (the dead end time)
and the mass of gas that passed through the critical venturi accumulates in the inventory
volume. The mass accumulation leads to a pressure rise in the inventory volume that will
depend on the mass flow, the size of the inventory volume, and the dead end time of the
diverter valves. The pressure in the inventory must not be permitted to reach a high
enough level that the flow at the venturi is no longer critical, lest pressure perturbations
reach the meter under test and disrupt the steady state flow conditions at the meter.

10

5 Design and Operation of the PVTt Standard

Experience with a previous, 26 m
3
PVTt flow standard at NIST indicated that
improvements in temperature and pressure instrumentation as well as in the design and
operation of the new system would be necessary in order to achieve an uncertainty of
0.05 % or better [Johnson et al. 2003]. In the following three sections we will describe
aspects of the design and operation of the new flow standard that are important to achieve
this low uncertainty. The first of these sections describes the measurement of the average
temperature of the collected gas. A subsequent section describes the procedures that
minimize the uncertainty of the mass change in the inventory volume, and the last section
describes the determination of the tank and inventory volumes.

5.1 Average Temperature of the Collected Gas

One of the most important sources of uncertainty in a PVTt flow standard is the
measurement of the average temperature of the gas in the collection tank, particularly
after filling. The evacuation and filling processes lead to cooling and heating of the gas
within the volume due to flow work and kinetic energy phenomena [Wright and Johnson
2000]. The magnitude of the effect depends on the flow; however, the temperature rise in
an adiabatic tank can be 10 K or more. Hence, immediately after filling and evacuation,
significant thermal gradients exist within the collected gas. For a large tank, the
equilibration time for the gas temperature can be many hours. If the exterior of the tank
has non-isothermal or time varying temperature conditions, stratification and non-
uniform gas temperatures will persist even after many hours [Johnson et al. 2003].

11

Figure 2. A schematic of the PVTt collection tanks, water bath, duct, and temperature
control elements.

In this work, we avoided long equilibration times and the difficult problem of measuring
the average temperature of a non-uniform gas by designing the collection tanks for rapid
equilibration of the collected gas and by immersing the tanks in a well-mixed,
thermostatted, water bath (see Fig. 2 and Fig. 3). There are two control volumes, a 34 L
collection tank and a 677 L collection tank. Because the equilibration of the 677 L tank is
slower, we consider it here. The 677 L tank is composed of eight, cylindrical, 2.5 m
long, stainless steel shells connected in parallel by a manifold. Each shell has a wall
thickness of l = 0.6 cm and an internal radius of a = 10 cm. Because all of the collected
gas is within 10 cm of a nearly isothermal shell, the gas temperature quickly equilibrates
with that of the bath. After the collected gas equilibrates with the bath, the gas
temperature is determined by comparatively simple measurements of the temperature of
the recirculating water. Remarkably, the water temperature measurements made with 14
sensors had a standard deviation of only 0.4 mK during a typical, 20-minute-long,
equilibration interval. In the next section, we describe the bath; in the section after that,
we discuss the equilibration of the collected gas.
12

Large diverter valves
Small diverter valves
677 L collection tank
manifold
Heating, cooling, and
mixing elements
34 L collection tank

Figure 3. A photograph of the two PVTt collection tanks submerged in the temperature
controlled water bath.

5.1.1 The Water Bath

The water bath is a rectangular trough 3.3 m long, 1 m wide, and 1 m high. Metal frames
immersed in the tank support all the cylindrical shells and a long duct formed by four
polycarbonate sheets. The duct surrounds the top, bottom, and sides of the shells:
however, both ends of the duct are unobstructed. At the upstream end of the bath, the
water is vigorously stirred and its temperature is controlled near the temperature of the
room (296.5 K) using controlled electrical heaters and tubing cooled by externally
refrigerated, circulated water. A propeller pushes the vigorously stirred water through the
duct along the collection tanks. When the flowing water reaches the downstream
(unstirred) end of the trough, it flows to the outsides of the duct and returns to the stirred
13

volume through the unobstructed, 10 cm thick, water-filled spaces between the duct and
the sides, the top, and the bottom of the rectangular tank.
297.286
297.287
297.288
297.289
297.29
0 10
Time (min)
T

(
K
)
20

Figure 4. Temperature data for 14 thermistors distributed in the water bath.

The uniformity and stability of the water temperature was studied using 14 thermistors.
The thermistors were bundled together and zeroed at one location in the water bath.
Then, they were distributed throughout the water bath. Figure 4 plots data recorded at 5 s
intervals from these 14 thermistors. Nearly all of the data in Fig. 4 is within 1 mK of
their mean and the standard deviation of the data from their mean is only 0.4 mK. The
largest temperature transients occur where the mixed water enters the duct, indicating
incomplete mixing. The tank walls attenuate these thermal transients before reaching the
collected gas. Thus, after equilibration, the non-uniformity of the water bath and the
fluctuations of the average gas temperature are less than 1 mK (3 10
-6
T).

5.1.2 Equilibration of the Collected Gas

For design purposes, we estimated the time constant (
gas
) that characterizes the
equilibration of the gas within the collection tank after the filling process. The estimate
considers heat conduction in an infinitely long, isotropic, solid cylinder of radius a
[Carslaw and Jaeger 1946]. For the slowest, radially symmetric heat mode,
14

gas
= (a/2.405)
2
/D
T
, where D
T
is the thermal diffusivity of the gas. This estimate gives
gas
= 80 s for nitrogen in the 677 L tank. This estimate for
gas
is too large insofar as it
neglects convection, conduction through the ends of the tanks, and the faster thermal
modes, all of which hasten equilibration. The time constants for heat to flow from the gas
through the tank walls and the time constant for a hot or cold spot within a wall to decay
have been calculated and found to be less than a second. Therefore, we expect the
collected gas to equilibrate with a time constant of 80 s or less.

Figure 5. The equilibration of pressure and temperature immediately following a filling
of the 34 L tank at 25 L/min.

The equilibration of the collected gas was observed by using the tank as a constant-
volume gas thermometer. After the tank valve was closed, the pressure of the collected
gas was monitored, as shown in Fig. 5. Our analysis of data such as those in Fig. 5 leads
to the experimental values
gas
of less than 60 s for both the 677 L and 34 L tanks, in
reasonable agreement with the estimates. The measured time constant and Fig. 5 show
that a wait of 20 minutes guarantees that the collected gas is in equilibrium with the bath,
within the resolution of the measurements.

15

The manifold linking the 8 cylindrical shells is completely immersed in the water bath.
Thus, the gas in the manifold quickly equilibrates to the bath temperature as well.
However, each collection system has small, unthermostatted, gas filled volumes in the
tubes that lead from the collection tanks to the diverter valves, the pressure transducers,
etc. In Section 6.4.1, we show the possible temperature variations of these small,
unthermostatted volumes make very small contributions to the uncertainty of the gas
temperature and the flow measurements.

5.2 Mass Change in the Inventory Volume

5.2.1 Overview and Strategy

As outlined in Section 4, the start time t
i
and the stop time t
f
used in Eqs. 1 and 2 are
chosen to occur during the brief dead-end times (< 100 ms) when both the tank valve
and the bypass valve are closed, i.e., we use a zero overlap diversion [Harris and
Johnson 1990]. This choice has the advantage of clear mass balance accountability for
all the gas flowing through the critical flow venturi during both diversions and the tank
filling. Unfortunately, it is difficult to determine either m or m and hence the change in
mass within the inventory volume accurately (especially at high flows) because both the
pressure and the temperature in the inventory volume rise rapidly as the flow through the
critical venturi accumulates in the inventory volume (see Fig. 6).
f
I
i
I

Our strategy for dealing with the inventory mass change has two elements. First, by
design, the inventory volume V
I
is much smaller than the collection tank volume V
T
. (For
the 34 L system, V
T
/V
I
= 500; for the 677 L system, V
T
/V
I
= 700.) Thus, the uncertainty
of mass flow is relatively insensitive to uncertainty in and m since both are small
compared with the total mass of collected gas. Second, we choose t
f
I
m
i
I
i
near the end of the
dead-end time and we chose t
f
such that P(t
i
) = P(t
f
). These choices define a mass
cancellation method: since the initial and final inventory densities are essentially equal,
m
I
is nearly zero. In fact, we will assume that m
I
is zero and consider the quantity only
in terms of flow measurement uncertainty, not as part of the flow calculation. Symmetry
16

of the inventory transients (see Fig. 7) and the mass cancellation method also give
uncertainty benefits due to high correlation in the uncertainties of pressure and
temperature measurements for m
I
.

We tested our strategy for choosing t
i
and t
f
for both the 34 L and the 677 L flow
standards. (See Section 7 for details of these tests.) To test the 34 L system, we collected
identical flows spanning the range 3 L/min < < 100 L/min in both the small and the
large tanks, using the large tank as a reference for the small tank since its inventory
uncertainties are quite small in this flow range. To test the 677 L collection system, we
collected identical flows in the 677 L tank following two different protocols. In the first
protocol, the inventory volume was dead-ended at the beginning and end of the collection
interval in the usual manner. In the second protocol the collection interval was divided
into two subintervals, which doubled
m&
I
m and allowed assessment of its uncertainty
contribution.

These tests indicate uncertainties due to the inventory volume that are proportional to
flow as would be expected based on a thermodynamic model of the inventory pressure
and temperature transients. If the inventory uncertainties are considered to arise from
uncertainty in the collection time, the inventory mass change uncertainty found
experimentally for the 34 L system was u m
I
m
& =
ms 4 (200 10
-6
for its maximum
flow). For the 677 L system, single and double diversions changed the flow measurement
by 75 10
m&
-6
or less. m&

In the remainder of this section, we describe conditions within the inventory volume
during the dead-end times using both a model and measurements. The measurements
show that T(t) and P(t) are nearly the same during the start and stop dead-end times.
Finally, we show that is insensitive to the exact choice of t
I
m
i
, provided that the
condition P(t
i
) = P(t
f
) is applied near the end of the dead-end time.

17

Figure 6. Experimentally measured data (25 L/min in the 34 L tank) and thermodynamic
model predictions for zero and non-zero sensor time constants. The model outputs
demonstrate that neglect of sensor response causes significant error in the measurement
of inventory conditions.

5.2.2 Conditions within the Inventory Volume

Figure 6 displays the time dependent temperature T(t) and pressure P(t) in the inventory
volume of the smaller collection system at a typical collection rate ( m = 25 L/min;
collection time = 82 s). The time t = 0 in Fig. 6 is defined by the signal indicating that
the valve is fully closed. The triangles ( = 0) in Fig. 6 were calculated from the lumped-
parameter, thermodynamic model developed by Wright and Johnson [2000]. The model
assumes a constant mass flow at the entrance to the inventory volume. The model
&
m&
18

neglects heat transport from the gas to the surrounding structure and non-uniform
conditions, such as the jet entering the volume. For Fig. 6, T(t) and P(t) were calculated
on the assumption that the diverter valve reduced the flow linearly (in time) to zero
during the interval 0.02 s < t < 0. Experimentally measured values of T(t) and P(t)
recorded at 3000 Hz (smooth curves) are also shown in Fig. 6. Most of the differences
between the measured curve and the ( = 0) calculated triangles result from the time
constants of the sensors used to measure T(t) and P(t). This is demonstrated by the
agreement between the experimental curve and the model results when time constants are
incorporated (circles).

In Fig. 6, the calculated curves do not display features that mark either the onset or the
completion of the diverter valve closing. Thus, even T(t) and P(t) data from perfect
sensors cannot be used to mark these events. For this reason, the times t
i
and t
f
were
chosen at times that were clearly within the dead-end time intervals.

Figure 6 shows that the measured values of T(t) and P(t) are consistent with the Wright-
Johnson model for the inventory volume after allowance is made for the response times
of the sensors. The consistency shows that the behavior of the inventory volume is
understood semi-quantitatively. However, this is not sufficient to accurately calculate the
density (t) from measurements of T(t) and P(t) because the fraction of the flow collected
as the valves are closing cannot be deduced from the measurements. Instead, we relied
on the pressure sensor to choose t
i
. The pressure sensor is preferred to the temperature
sensor because it responds more quickly and also because it responds to the average
conditions throughout the inventory volume rather than the conditions at only one
location. We choose t
i
near the end of the dead-end time, where the P(t) measurements
are nearly parallel to the = 0 model. In this regime, the derivative dP/dt is large and its
dependence on precisely how the valve closed is small. Because the dependence on how
the valve closed has decayed, we expect that P(t) will be the same during the start and the
stop dead-end times, improving the mass cancellation as well as the correlation of initial
and final inventory density uncertainties.

19

5.2.3 Near Symmetry of Start and Stop Behavior of P(t)

Figure 7 shows records of T(t) and P(t) taken during the dead-end time intervals at the
start and the stop of a single flow measurement. As before, the data were recorded at
3000 Hz for 500 ms and the plots were displaced along the horizontal axis until they
nearly overlapped. The pressure and the temperature at the beginning of the start dead-
end time were slightly lower than those at the stop dead end time; however, the two
records match closely during the dead-end time. This implies that the time-dependent
densities (t) also nearly match.

Figure 7. Superimposed inventory data traces for a start diversion and a stop diversion in
the 34 L tank at 25 L/min demonstrating symmetric diverter valve behavior. The stop
dead-end time was approximately 15 ms longer.

At both diversions shown in Fig. 7, valve trigger signals were gathered along with the
temperature and pressure measurements using a commercially manufactured data
20

acquisition card (see Fig. 8). The trigger signals originate from an LED/photodiode pair
and a flag on the valve actuator positioned so that the circuit output rises to a positive
voltage when the valve is closed. These valve signals are used to trigger timers that give
the approximate collection time.

As represented in Fig. 8, the inventory record is post-processed by the controlling
program to obtain both the initial and final measurements of pressure and temperature in
the inventory volume as well as the final collection time. A match pressure P(t
i
) is
chosen that falls late within the start dead-end interval. The stop time is then found in the
stop dead-end interval by choosing P(t
f
) = P(t
i
). Time corrections between the match
pressure measurement and the start and stop trigger signals (t
i
and t
f
) are determined
from the data record. The appropriate time corrections are added to the approximate
collection time from the timers. Thus, by adjusting the collection time using the inventory
data records, the initial and final inventory pressures and temperatures are nearly
matched, leading to nearly equal initial and final inventory densities and inventory mass
cancellation.

Figure 8. Data records of inventory sensors and valve trigger signals are used to adjust
the collection time and improve cancellation of the initial and final inventory mass as
well as inventory uncertainties.

21

5.2.4 Insensitivity of to the Match Pressure
I
m

Figure 9 shows the total correction time as a function of the match pressure for two flows
in the 34 L system. The 100 L/min flow is very high for the 34 L tank, having only an
18 s collection time. Match pressure is shown as a percentage of the range of pressures
measured during the diversion transient. For a perfectly fast system (valves and sensors),
these plots would be horizontal lines, i.e. any chosen match pressure would result in the
same time correction. However, for the real system with its inevitable limitations, the
match pressure does matter. Exploring the possible reasons for this is valuable for
improving the system and for obtaining an accurate uncertainty analysis.

First recall that the inventory sensors have non-zero time constants and therefore the
measurements they provide are damped versions of the real conditions and further, the
values they report at any given instant are subject to the recent history of the pressure or
temperature value. Second, realize that perfect symmetry of conditions before and during
diversion is unobtainable and that these imperfections and the significance of the sensor
damping increase with the flow. For example, at high flows, the rate of change of
pressure during the tank filling process is large and it becomes more difficult to make the
pressure at which the stop diversion begins closely match the pressure at which the start
diversion began (due to sensor response and valve control delays). This trigger pressure
difference will be considered again in Section 7. As another example, the bypass and
tank valves may not close at the same speed.

22

Figure 9. The collection time correction versus the match pressure used in the inventory
mass cancellation algorithm.

Analysis of the thermodynamic model of the inventory and its sensors shows that times
later in the dead-end time give better mass cancellation under these circumstances since
the sensor output enters a period with nearly constant slope that is equal to the real
pressure slope. The experimental results given in Fig. 9 support this assertion: match
pressures between 50 % and 90 % result in nearly constant correction times, while low
match pressures (early in the dead-end time) give much larger corrections. Based on this
analysis, a match pressure of 80 % has been selected for use in the flow standard. Figure
9 demonstrates the insensitivity of
I
m to a wide range of match pressure values.

Figure 9 also illustrates the concept that uncertainties related to the inventory volume can
be treated not only as mass measurement uncertainties, but as time measurement
uncertainties as well. One can consider the uncertainty in the measurement of time
between conditions of perfect mass cancellation, or one can consider the uncertainty in
the measurement of inventory mass differences between the start and stop times. Both
perspectives offer insight and verification of the uncertainties of the inventory volume
and flow diversion process.

23

5.3 Measurement of the Tank and Inventory Volumes

5.3.1 Gas Gravimetric Method

The volume of the 677 L tank was determined by the gas gravimetric method. In this
method, the mass of an aluminum high pressure cylinder was measured before and after
discharging its gas into the evacuated collection tank. The change in mass of the high-
pressure cylinder and the change in density of the gas in the collection tank were used to
calculate the collection tank volume. Nominally,

extra
T T
c c
grav
V
m m
V
=
i f
f i

, (3)

where the m
c
indicates the mass of the high-pressure cylinder and V
extra
represents the
extra volume temporarily connected to the tank for the purpose of introducing the gas
from the cylinder to the tank (usually a small volume of tubing and a valve body). The
extra volume is calculated from dimensional measurements or measured by liquid
volume transfer methods.

In practice, a more complex formula than Eq. 3 was used to account for a small amount
of gas that enters the control volume from the room when the cylinder is disconnected
from the collection tank since the final tank pressure was less than atmospheric. For the
volume determinations performed for the 677 L tank, the effect amounts to only
5 10
-6
V
T
.

The volume determination was conducted with both nitrogen and argon gas. In both cases
high purity gas was used (99.999 %) and care was taken to evacuate and purge the system
to minimize composition uncertainties. When nitrogen was used, the aluminum cylinder
weighed approximately 4200 g when filled at 12.5 MPa, and approximately 3800 g after
it was emptied to 55 kPa. When argon was used, the initial and final masses were 4440 g
24

and 3820 g respectively. The standard deviation of the six volume measurements (4 with
nitrogen, 2 with argon) was 16 10
-6
V
T
.

The initial and final masses of the gas cylinder were measured using a substitution
process with reference masses and a mass comparator enclosed in a wind screening box.
The comparator has a full scale of 10 kg and a resolution of 1 mg. The cylinder and a set
of reference masses of nearly the same weight were alternated on the scale 5 times. The
zero corrected scale readings were then calibrated to the reference masses and buoyancy
corrected via the following formula:

ext air
ref
air
ref
ref
c
c
V m
S
S
m
+
|
|
.
|
\
|
= 1 , (4)

where S represents the scale reading, ref indicates the reference masses,
air
is the
ambient air density where the measurements were conducted, and V
ext
is the external
volume of the high pressure cylinder and its valve and fittings. The density of the ambient
air was calculated from the barometric pressure, the temperature and humidity inside the
wind screen, and an air density formula that includes humidity [Jaeger and Davis 1984].

The external volume of the high-pressure cylinder appears in Eq. 4 due to air buoyancy
corrections. The external volume of the cylinder was measured by Archimedes principle,
i.e. by measuring the change in apparent mass of the object in two media with differing
and known densities. One of the media was distilled water, and the cylinder apparent
mass in the water was measured as follows. Liquid was added to the cylinder interior
until it was nearly neutrally buoyant in the tank of distilled water. The addition of liquid
inside the cylinder has no effect on its external volume. The temperature of the distilled
water was raised or lowered (thereby changing the density of the distilled water) until the
cylinder was essentially neutrally buoyant. At this point, the apparent mass in the distilled
water is zero. The temperature of the distilled water was recorded and its density was
calculated via an equation from the literature [Patterson and Morris 1994]. Hence, the
temperature of the distilled water was used in place of a weigh scale to measure the
25

apparent mass in water. The apparent mass of the cylinder in air (with the liquid still
inside) was measured using the comparator described above. The density of air with
humidity was calculated as previously described. The external volume of the cylinder
was calculated for the nominal room temperature (T
ref
) of 296.5 K with the following
formula:

( )
( ) | | ( | )|
ref air air ref water water
A
water
A
air
ref ext
T T T T
m m
T V
+ +

=
3 1 3 1
, (5)

where the superscript A indicates apparent mass and is the coefficient of linear thermal
expansion for the aluminum tank. The terms containing correct for changes in the
cylinder volume due to differences between the water temperature, the air temperature,
and the reference temperature. However, for this particular case, these thermal expansion
issues could have been neglected since both the water temperature and air temperature
never differed from T
ref
by more than 1.5 K. The thermal expansion corrections to the
external volume were less than 0.5 cm
3
or 100 10
-6
V
ext
and the external volume has a
small sensitivity coefficient in the collection tank volume determination process.

The expansion of the external cylinder volume as a function of its internal pressure was
not negligible. The Archimedes principle measurements showed a volume increase from
4697.5 cm
3
to 4709 cm
3
between the 100 kPa and 12.5 Mpa pressures. This change
agreed well with predictions based on material properties, and the appropriate
experimental values for external volume were used in the cylinder mass calculations (Eq.
4), depending on whether the cylinder was empty or full. If this issue were neglected, it
would lead to relative errors in the mass change measurements of about 35 10
-6
.

26

5.3.2 Volume Expansion Method

The 34 L collection tank volume, the inventory volume for the large collection tank, and
the small inventory volume were all determined with a volume expansion method. In this
method, a known volume is pressurized, the unknown volume is evacuated, a valve is
opened between the two volumes, and the density changes within the two volumes are
used to calculate the unknown volume. Applying conservation of mass to the system of
the two tanks yields:

( )
( )
extra
V
V
V
=
f
2
i
2
1
i
1
f
1
2

, (6)

where the subscripts 1 and 2 refer to the known and unknown volumes respectively. As
before, the density values are based on pressure and temperature measurements of the gas
within the volumes and gas purity issues must be considered. Note that in many cases the
final densities can be considered the same in both volumes 1 and 2, but for the
determination of the 34 L tank volume, elevation differences between the two tanks
required a head correction to the pressure measurements and therefore the two densities
were not strictly equal. The difference in elevation resulted in a relative difference in gas
density of 20 10
-6
even though the two tanks were connected.

6 Uncertainty Analysis of the 34 L and 677 L PVTt Flow Standards

In this section, we will analyze the uncertainty of the 34 L and 677 L PVTt standards. We
will begin by giving an overview of the subject of uncertainty analysis including the issue
of correlated uncertainties. Next we will give the results of the uncertainty analysis for
mass flow. In following sections, we will give uncertainties of the sub-components that
were combined to obtain the mass flow uncertainty. The largest source of uncertainty in
the flow measurement is drift of the pressure sensor over time, which contributes a
27

relative standard uncertainty of 60 10
-6
to the determinations of the volumes of the
collection tanks and to the flow measurements.

6.1 Techniques for Uncertainty Analysis

The uncertainty of a mass flow measurement made with the PVTt standard use the
propagation of uncertainties techniques described in the ISO Guide to the Expression of
Uncertainty in Measurement [ISO 1996]. The process identifies the equations involved
in the flow measurement so that the sensitivity of the final result to uncertainties in the
input quantities can be evaluated. The uncertainty of each of the input quantities is
determined, weighted by its sensitivity, and combined with the other uncertainty
components to arrive at a combined uncertainty.

As described in the references [ISO 1996, Coleman and Steele 1999], consider a process
that has an output, y, based on N input quantities, x
i
. For the generic basis equation:

) , , , (
2 1 N
x x x y y K = , (7)

if all the uncertainty components are uncorrelated, the standard uncertainties are
combined by root-sum-square (RSS):

( ) ( )
=
|
|
.
|
\
|
=
N
i
i
i
c
x u
x
y
y u
1
2
2
, (8)
where u(x
i
) is the standard uncertainty for each of the inputs, and u
c
(y) is the combined
standard uncertainty of the measurand. The partial derivatives in Eq. 8 represent the
sensitivity of the measurand to the uncertainty of each input quantity.

In cases where correlated uncertainties are significant (as in the following analysis), the
following expression should be used instead of Eq. 8:

28

( ) ( ) ( ) ( ) ( )

=
= + =

+
|
|
.
|
\
|
=
N
i
N
i
N
i j
j i j i
j i
i
i
c
x x r x u x u
x
y
x
y
x u
x
y
y u
1
1
1 1
2
2
, 2 , (9)

where r(x
i
, x
j
) is the correlation coefficient, ranging from 1 to 1, and equaling zero if the
two components are uncorrelated. As will be seen in the following analysis, some
uncertainty components in the present system are correlated and this leads to a significant
improvement in the uncertainty of the measurand.

A simple example of a correlated uncertainty is illustrative. Suppose that a thermometer
was used to measure a temperature difference. Also suppose that the only uncertainty in
the thermometer measurement was an unknown offset in its calibration. When the
difference between two temperatures was calculated from measurements made with this
thermometer, the offset would cancel and would not contribute to the uncertainty of the
temperature difference. In this case, the subtraction process used to calculate the
difference leads to sensitivity coefficients of opposite sign for the two temperature
measurements. Since the sensor always has the same offset, the uncertainties are perfectly
correlated (r(x
i
, x
j
) = 1). When this hypothetical scenario is processed through Eq. 9, the
uncertainty of the temperature difference is zero. Of course in a real case, there would be
other, uncorrelated uncertainties that would make the uncertainty of the temperature
difference non-zero. Nonetheless, the example demonstrates that under certain
circumstances, correlated uncertainties will reduce the uncertainty of a measured
quantity.

In the following uncertainty analysis, correlated and uncorrelated uncertainties will be
treated as separate components, even if they are related to the same physical quantity. For
instance, there will be a correlated as well as an uncorrelated inventory pressure
component. In this manner, the correlated components can be considered as having a
correlation coefficient of 1, while uncorrelated components have correlation coefficients
of 0. This approach simplifies the process to deciding which uncertainty sources are
correlated versus uncorrelated and checking that the assumption of perfect correlation is
reasonable.
29

Most of the equations utilized to calculate flow, mass, volume, and other necessary
intermediate quantities for the PVTt standard have been discussed in prior sections. In
Fig. 10 the information is summarized in a diagram that shows the measurement chain
used to calculate flow. At the top of the diagram is the output, mass flow. At the second
level are the inputs to Eq. 2, the quantities needed to calculate mass flow: density,
volumes, and collection time. To calculate density, the inputs to the equation of state are
necessary: pressure, temperature, compressibility, the universal gas constant, and
molecular weight. The other necessary quantities and their basis equations are shown as
well. Figure 10 will serve as a guide for the PVTt uncertainty analysis.

Eq. 6
Eq. 5
Eq. 2

Eq. 4
Eq. 3
m&
I
V
x
T
V
t
P
T
Z
m
ref
m
ext
V
air
water
R
x
M
air
m

Figure 10. The chain of measurements and equations used for the PVTt flow standard.

30

The discharge coefficient resulting from a flowmeter calibration will have additional
uncertainties not considered herein due to measurements associated with the meter under
test. For instance, if the meter under test is a critical flow venturi, uncertainties related to
the temperature and pressure measurements at the meter must be included in the
uncertainty of the discharge coefficient.

The uncertainties tabulated herein are k = 1, standard, or 68 % confidence level
uncertainties. At the conclusion of the uncertainty analysis, a coverage factor of 2 will be
applied to give an expanded uncertainty for mass flow measurements with an
approximate 95 % confidence level. In the remainder of this section, we will give the
uncertainty of mass flow for both the 34 L and 677 L PVTt systems, and then the
uncertainty components that contribute to the PVTt mass flow measurement will be
traced to their fundamental sources.

6.2 Mass Flow Uncertainty

Table 2. Uncertainty of nitrogen flow measurement with the 677 L standard.

Uncertainty Category Standard Uncertainty
(k = 1)
Contrib Comments
Flow (677 L, N
2
) Relative (10
6
) (%)
Tank volume 71 48.44 cm
3
50 to 23 see Table 15
Tank initial density 10 2.27 10
-12
g/cm
3
1 to 0
Tank final density 68 7.77 10
-8
g/cm
3
Inventory mass change 0 to 109 0.084 g 0 to 53 see Table 21
Collection time 15 0.287 ms 0 to 1 see Table 11
Std deviation of repeated meas. 20 0.001 g/sec 4 to 2
RSS (combined uncertainty) 102 to 150
Expanded uncertainty (k = 2) 204 to 300

The uncertainty for flows between 20 L/min and 2000 L/min of nitrogen or argon in the
677 L tank is given in Table 2. The standard uncertainty of each sub-component is given
31

in both relative ( 10
-6
) and dimensional forms. The units of the dimensional values are
given in the third column. The relative contribution of each sub-component to the
combined uncertainty is listed in the fourth column. This contribution is the percentage of
the squared individual component relative to the sum of the squares of all sub-
components. The uncertainty from the inventory volume, the combined uncertainty, the
expanded uncertainty, and the uncertainty contributions are given as a range covering the
minimum to maximum flow. To calculate their relative uncertainty in Table 2, the tank
initial density was normalized by the tank final density and the inventory mass change
was normalized by the total mass collected.

At the highest flow, uncertainty contributions are principally divided between the tank
volume, the final gas density, and the inventory uncertainty. The k = 2 uncertainty falls to
204 10
-6
for the smallest flows as the uncertainty contributions of the inventory
volume become negligible. For an air flow measurement, the uncertainty of the 677 L
system is less than 500 10
m&
-6
over the entire flow range and the uncertainty is driven
by the tank final density measurement (80 % contributor).
m&

Table 3. Uncertainty of nitrogen flow measurement with the 34 L standard.

(k = 1)
Contrib Comments
Flow (34 L, N
2
) Relative (10
6
) (%)
Tank volume 116 3.955 cm
3
Tank initial density 10 2.27 10
-12
g/cm
3
1 to 0
Tank final density 68 7.77 10
-8
g/cm
3
Inventory mass change 0 to 170 0.007 g 0 to 61 see Table 22
Collection time 15 0.287 ms 0 to 0 see Table 11
Std deviation of repeated meas. 20 4 10
-5
g/sec 2 to 1
RSS (combined uncertainty) 137 to 219
Expanded uncertainty (k = 2) 274 to 438

32

Table 3 presents the uncertainty of flow measurements from the 34 L system for flows
between 1 L/min and 100 L/min. The expanded uncertainty varies between 270 10
-6

and 440 10
m&
-6
. At high flows, the significant uncertainty sources are the tank volume,
the tank final density, and the uncorrelated inventory uncertainties. For low flows, the
major contributors are tank volume and final gas density. For air flow measurements, the
34 L system has a nearly constant uncertainty over its entire flow range of about
500 10
m&
-6
and it is driven by the uncertainty of the final gas density. m&

6.3 Pressure

A Ruska Model 2465-754 gas lubricated piston pressure gauge is used as the primary
pressure standard to calibrate pressure transducers within the Fluid Flow Group. The
uncertainties in a single pressure measurement made with this device are listed in Table
4. Uncertainties in the pressure standard can be traced to the effective area of the piston,
piston thermal expansion, the masses, local gravity, and the measurement of the pressure
under the bell jar covering the piston and masses (necessary for absolute pressure
measurements). The uncertainty shown in Table 4 is for a pressure value of 100 kPa. No
buoyancy corrections are made to the masses since the reference pressure, and hence the
density under the bell, are small enough that the buoyancy corrections (and uncertainties)
are negligible (<<1 10
-6
P). The uncertainty of the piston pressure gauge is 17 10
-6
P
at 100 kPa.

33

Table 4. Uncertainties for a 100 kPa pressure measurement made with the piston pressure
gage used as the standard for pressure calibrations.

(k = 1)
Contrib Comments
Relative (10
6
) (kPa) (%)
Piston Pressure Gage
Pressure value 100
piston effective area 12 0.0012 59 from NIST Pressure and Vacuum Group cal.
thermal expansion 6 0.0006 15 assume T unc of 0.2K
masses 1 0.0001 0 Mass Group calibration
local gravity 0.2 0.00002 0 9.801011+/- 0.000002
air density for buoyancy 0 0 0 neg. air density relative to mass density
ref. P for absolute 10 0.001 41 based on calibration data of the vac gauge
RSS 17 0.0017

6.3.1 Collection Tank Pressure

Measurements of the collection tank initial pressure are made with a pair of thermocouple
vacuum gauges (Varian Convectorr P-type) that have been calibrated by comparison to a
reference standard in the NIST Pressure and Vacuum Group. The manufacturers
uncertainty specification for this gauge is 10 % of reading. Based on the NIST calibration
results, the consistent agreement between the redundant sensors, and the repeatable
readings of the gauges at the vacuum pump ultimate pressure, a standard uncertainty of
5 % of reading will be used. As will be seen when the components are combined to give
the flow measurement uncertainty, this large value has little impact due to the low initial
pressure in the collection tank (20 Pa).

Pressure measurements of the full collection tank are made with a Paroscientific Model
740 with a full scale of 200 kPa. The manufacturers uncertainty specification for this
transducer is 0.01 % of full scale, but under the conditions of the present usage, the
uncertainty is less. The uncertainties in the collection tank pressure measurement are
34

listed in Table 5. They include the uncertainties from the piston pressure gage, the long
term drift of the Paroscientific transducer which has been quantified by periodic re-
calibrations, as well as the residuals from the best fit calibration equation (including
hysteresis), and thermal effects. Uncertainties due to spatial non-uniformity of pressure
within the tank and time response of the sensor are negligible since the calibration
procedure is to wait as much as 20 minutes for equilibration before the measurements are
made.

Figure 11 is a control chart that shows the changes in pressure calibration versus time at a
pressure of 100 kPa for one of the pressure transducers used to measure collection tank
pressure. Also shown are the k = 1 uncertainty tolerance bounds (64 10
-6
P from Table
5) and error bars that represent the k = 1 uncertainty of the piston pressure gauge used to
calibrate the sensor (17 10
-6
P from Table 4).

-150.0
-100.0
-50.0
0.0
50.0
100.0
150.0
1/1/99 1/1/00 12/31/00 12/31/01 12/31/02 12/31/03
Date
(
P
-
P
l
a
s
t
)

/

P

*

1
0
6
New
sensor
Figure 11. A calibration control chart for a 200 kPa pressure transducer used to measure
the collection tank pressure.

Temperature effects as large as 40 10
-6
P have been observed in the tank pressure
sensor, and care was taken to minimize their influence. When the tank is quickly filled
from a pressurized cylinder during the volume determination process, cold gas enters the
sensor, cooling it. The pressure readings asymptotically approach a final value as the
35

sensor returns to room temperature (with a time constant of approximately 1 hr). The
temperature dependence of the sensor was confirmed by testing with an environmental
chamber. Temperature effects also result in a hysteresis loop for the sensor calibration
data that enlarges the calibration fit residuals. The calibration process entails increasing
and decreasing the pressure in steps. The pressure steps result in heating and cooling of
the pressure sensing elements and a hysteresis loop. Therefore the residuals of the
pressure calibration fit include contributions due to thermal effects. We noticed that the
thermal effects due to pressure changes in the transducers can be larger than the values
given in Table 5. During volume determinations, we allowed sufficient time for the
sensor to return to room temperature so that the remaining temperature effects were much
smaller than the allowance for calibration drift.

Table 5. Uncertainties in the collection tank pressure measurement at 100 kPa.

(k=1)
Contrib Comments
Relative (10
6
) (kPa) (%)
Pressure Measurement
Pressure value 100
piston pressure gage 17 0.0017 7 from Table 4
drift 60 0.0060 88 < 0.01 % in 6 mos, assume rect.
residuals, hysteresis, thermal effects 14 0.0014 5 from cal. records, experiments
RSS 64 0.0064

6.4 Temperature

The temperature sensors used in the flow standard are traceable to the NIST
Thermometry Group through calibrations made with a four-wire thermister transfer
standard (Thermometrics Model TS8901) and a recirculating water constant temperature
bath. The uncertainty of the transfer standard thermister is 1.2 mK (see Table 6). The drift
of the transfer standard is considered negligible based on 7 annual calibrations that have
always differed from each other by less than the calibration uncertainty.
36

Table 6. Uncertainties for the Fluid Flow Group temperature transfer standard.

(k =1)
Contrib Comments
Relative (10
6
) (mK) (%)
Temperature Transfer Standard
Temperature value 3 10
5

Thermometry Group cal 4 1.2 94 unc. for 274 K to 368 K
fit residuals 1 0.3 6 some years, 1/6 this size
drift 0 0 0 less than discernable given cal unc.
radiation, self-heating, etc. 0 0 0 deeply immersed in water bath
RSS 4 1.2

6.4.1 Collection Tank Temperature

The measurement of the temperature of the gas in the collection tank has additional
uncertainties that are listed and quantified in Table 7. Temperature is measured with YSI
Model 46000 thermisters in 3 mm diameter stainless steel sheaths, a Keithley model 224
current source, a Keithley model 7001 switch system, and a Keithley model 2002
multimeter. Uncertainty sources include the temperature transfer standard covered by
Table 6, the uniformity and stability of the water bath used to calibrate the thermisters,
and the residuals of the best-fit equation to the calibration data. The largest uncertainty
component is the calibration drift between periodic calibrations. Radiation and stem
conduction are negligible since the thermisters are immersed at least 15 cm in room
temperature water. Tests were conducted to measure the significance of self-heating by
varying the thermister current while the sensor was held in a stable water bath and
watching the resulting change in sensor reading. Based on this experiment, the current
through the 5000 ohm thermisters was set to 10 A which leads to self-heating of less
than 1 mK. The PVTt bath stability and uniformity (1 mK) were discussed earlier, as was
the issue of thermal equilibrium between the water bath and the gas in the collection tank.
The sensors are calibrated over their entire range of usage, so there is no uncertainty
37

related to extrapolation of their calibration data. Uncertainty related to the time response
of the thermisters is negligible since the time constant for the sensor is on the order of
10 s and the wait for thermal equilibrium is 30 or more times longer.

Small portions of the gas collection tank are not immersed in the water bath. They
include the tubing connecting the outlet of the diverter valve to the collection tank, the
tubing that connects the tank to pressure and vacuum transducers, and the internal volume
of these transducers. Because we assumed that the bath temperature represents the gas
temperature, and the room temperature may differ from that of the bath, the small portion
of the tank not immersed leads to a gas temperature uncertainty. The room temperature is
maintained at 23.5 1C. The fractional error in mass contained in the collection tank
due to a 1 K difference between the room and the water bath is:

T
T
V
V
m
m
T
out

, (10)
where V
out
is the volume of the portion of the tank that is at room temperature and T is
the difference between the room and water bath temperatures. For the large tank, V
out
is
200 cm
3
and the total tank volume is 677 L. For T of 1 K, the relative mass uncertainty
is 1 10
-6
. This uncertainty will be treated as an uncertainty in the average gas
temperature of 1 10
-6
T or 0.3 mK in the following analysis. For the small tank this
temperature uncertainty is 1.2 mK or 4 10
-6
T.

By RSS of the components listed in Table 7, the combined uncertainty for the average
temperature of the gas in the collection tank is 7 mK when using the thermisters
dedicated to the flow standard.

38

Table 7. Uncertainty of average gas temperature in the collection tank with the dedicated
temperature sensors.

(k =1)
Contrib Comments
Relative (10
6
) (mK) (%)
Tank Gas Average Temperature
Temperature value 297000
temperature transfer standard 4 1.2 4 from Table 7
cal. bath uniformity and stability 3 1 2 based on notes made during cal
fit residuals 7 2 9
drift (I, R, DMM, thermistors) 20 5.8 77 YSI spec is <10mK/10 mo., assume rect.
radiation, stem cond., self-heating 3 1 2 expt. varied current, 5 mK, assume rect.
PVTt bath uniformity and stability 3 1 2 based on experimental measurements
diff. between water and room T's 4 1.2 3 see Eq. 10
extrapolation 0 0 0 cal covers whole operating range
sensor time constant 0 0 0 wait time is >10 sensor time constant
RSS 22 7

6.5 Compressibility, Molecular Weight, and Gas Constant

The remaining contributors to gas density uncertainty are the compressibility, the
molecular weight, and the gas constant. Four gas cases will be considered. Tank volume
determinations were conducted with ultra high purity nitrogen (99.999 %) and ultra high
purity argon (99.999 %). During normal calibrations, industrial liquid nitrogen vaporized
from dewars (99.998 %) and compressed, dried air are used.

6.5.1 Gas Constant

The universal gas constant is known with relative standard uncertainty better than
2 10
-6
[Moldover et al. 1988].

39

6.5.2 Compressibility

The compressibility factor can be calculated from the following expression:

2
1
n n
C B Z + + = , (11)

where
n
is the molar density (mol/ cm
3
) and B and C are the second and third virial
coefficients respectively. For nitrogen, the virial coefficients can be calculated from the
correlations:

2 4
2
10 636 . 7 65125 . 0 21 . 131 T T B
N

+ + = , (12)

2
2
015 . 0 35 . 11 2 . 3454 T T C
N
+ = , (13)

where temperature is in K and B and C are in cm
3
/mol and cm
6
/mol
2
respectively and the
units of the constants have been surpressed. For dry air the correlations for the virial
coefficients are:

2 4
10 833 . 7 66785 . 0 06 . 137 T T B
Air

+ + = , (14)

2
016 . 0 40 . 12 9 . 3528 T T C
Air
+ = . (15)

These expressions are the result of least squares best fitting to outputs from a database of
the property measurements by numerous experimenters [Lemmon et al. 2002]. Equations
12 through 15 were fitted to data over the range of 270 K to 330 K. This range allows
application of the correlations at the conditions found in the test section of the flow
standard. However, it should be noted that the conditions in the collection tank are much
narrower, with pressures ranging from 0 kPa to 101 kPa at a nearly constant temperature
of 296.5 K. For these narrower conditions, the third virial coefficient could be ignored
with negligible impact on the density uncertainty.
40

Uncertainty estimates for experimental studies of compressibility are often unavailable,
especially for older publications. Comparison of previously compiled compressibility
values obtained by various researchers [Dymond and Smith 1980, Hilsenrath et al. 1955,
Span et al. 2000, Lemmon et al. 2000] shows agreement to within 10 10
-6
in the 270 K
to 330 K temperature range at 100 kPa. Perhaps more valuable is that this level of
agreement is achieved between compressibility measurements made by the traditional
PVT method and by the more recent speed of sound techniques [Trusler et al. 1997].
Based on this information, a relative standard uncertainty of 10 10
-6
will be used for the
experimental measurements of compressibility. This uncertainty is for a pressure of
100 kPa and it scales with density. The residuals of the equation fitting process to the
experimental data are negligibly small (<1 10
-6
Z). The uncertainty components of the
compressibility factor are listed in Table 8.

Table 8. Uncertainty of the compressibility for nitrogen, argon, and dry air.

(k =1)
Contrib Comments
Compressibility (Z) Relative (10
6
) (-) (%)
Compressibility value 1
experimental data 10 1 10
-5
99 based on analysis of literature
impurity effects on Z 1 1 10
-6
1
residuals of best fit 0 0 0 <1 10
-6

RSS 10 1 10
-5

The source of the dry air used in the flow standard is a Joy oil-free two stage
reciprocating compressor and a Zurn refrigeration dryer. The mole fraction of water in
this air source is 0.0010 0.0005. This uncertainty in composition leads to uncertainty in
the compressibility of only 1 10
-6
Z. The uncertainty in compressibility due to
impurities is smaller for the other gases.

41

6.5.3 Molecular Weight

The departure of the molecular weight of ultra high purity nitrogen, industrial liquid
nitrogen, and ultra high purity argon from the molecular weight of the pure substance was
examined using the impurity specifications of the gas manufacturer. This analysis results
in molecular weight relative standard uncertainties less than 1 10
-6
, but 1 10
-6
will be
assumed for the molecular weight of nitrogen (28.01348 g/mol) and argon
(39.94800 g/mole). For the dry air described above, the molecular weight is
28.9532 g/mol and its relative standard uncertainty is 190 10
-6
due to the variability of
water content.

6.6 Density

Now that the sub-components have been quantified, the uncertainty of density
measurements made in the collection tank with nitrogen and argon (for volume
determinations and for flow measurements) and with dry air (for flow measurements) can
be calculated and they are presented in Tables 9 and 10. For the pure gases, the relative
standard uncertainty is 68 10
-6
and the primary contributor is the pressure measurement.
For air, the relative standard uncertainty is 208 10
-6
and the largest contribution is due
to water content variations. The uncertainties of the density of ambient air (needed for
buoyancy corrections) and of distilled water are considered in the section pertaining to
tank volume determinations.

42

Table 9. The uncertainty of collection tank gas density for nitrogen and argon.
(k=1)
Contrib Comments
Collection Tank Density (N
2
& Ar) Relative (10
6
) (%)
pressure 64 6.45 10
-3
kPa 88 from Table 5
temperature 22 6.47 mK 10 from Table 7
compressibility 10 1 10
-5
2 from Table 8
molecular weight (purity) 1 2.80 10
-5
g/mol 0
gas constant 2 1.41 10
-2
(cm
2
)/(s
2
K) 0
RSS 68 7.82 10
-8
g/cm
3

Table 10. The uncertainty of collection tank gas density for dry air from the NIST Fluid
Flow Group small compressor and dryer system.

(k =1)
Contrib Comments
Collection Tank Density (Air) Relative (10
6
) (%)
pressure 64 6.45 10
-3
kPa 9 from Table 5
temperature 22 6.47 mK 1 from Table 7
compressibility 10 1 10
-5
0 from Table 8
molecular weight (purity) 190 5.5 10
-3
g/mol 89
gas constant 2 1.41 10
-2
(cm
2
)/(s
2
K) 0
RSS 208 2.38 10
-7
g/cm
3

6.7 Collection Time

As explained in section 5.2.3, the collection time is an approximate time measured by
timers that is then corrected via analysis of records of the inventory pressure data and
trigger voltages to minimize inventory mass and improve uncertainty cancellation. The
base time is measured redundantly with two Hewlett-Packard Model 53131A counters.
The counter calibration and usage leads to less than 0.01 ms uncertainty. Due to the
3000 Hz recording frequency of the inventory pressure and trigger data, the actual rise of
43

the trigger voltage can be any time within a 0.33 ms window. Assuming a rectangular
distribution, the post-processing corrections will have a standard uncertainty of 0.19 ms.
This uncertainty applies to both the start and stop times. The time uncertainties of the two
pressure measurements used in the mass cancellation procedure are negligible since the
times are found by interpolation of the data records and are much smaller than 0.33 ms.
The combined collection time uncertainty is 0.3 ms (see Table 11).

Table 11. Collection time uncertainties.

(k =1)
Contrib Comments
Relative (10
6
) (ms) (%)
Collection Time
Time value 100000
timer cal and usage 0 0.01 0 base time uncertainty
inventory correction (start) 2 0.19 50 3000 Hz, rectangular distribution
inventory correction (stop) 2 0.19 50 3000 Hz, rectangular distribution
RSS 3 0.3

6.8 Volume of the 677 Liter Collection Tank (Gravimetric Method)

The uncertainty of the determination of the collection tank volume by the previously
described gravimetric method (section 5.3.1) is traceable to the uncertainty of the mass
and density measurements made during the process, which are in turn dependent on the
quantities shown in Fig. 10. The uncertainty of the density measurements for the pure
nitrogen and argon gases used for the volume measurements was given in Table 9. The
scale used was a Mettler-Toledo model PR10003, which has a 10 kg capacity. The
uncertainty of the mass measurements of the high-pressure cylinder before and after
discharge is dependent on the buoyancy corrections (ambient air density and cylinder
external volume), the reference masses used with the mass comparator, and the
performance of the mass comparator. The measurement of the external volume of the
high-pressure cylinder via the Archimedes principle was described in an earlier section.
44

The uncertainty of this measurement is traceable to the density of distilled water, ambient
air, reference masses, and the performance of the mass measuring systems.

The uncertainty of the density of ambient air during the course of the various weighing
procedures is given in Table 12. The pressure, temperature, and relative humidity
uncertainties account for instrument calibration uncertainties as well as variations in the
room conditions during the time needed to make a mass measurement. The relative
combined uncertainty of the ambient air density is about 500 10
-6
, with the largest
contribution being from temperature.

Table 12. Uncertainty of the density of ambient air.

(k =1)
Contrib Comments
Ambient air density Relative (10
6
) (%)
pressure 129 0.013 kPa 6 cals and change during meas
temperature 336 100 mK 45 cals and change during meas
relative humidity 189394 2.5 % 15 cals and change during meas
equation of state (Z, M, R) 200 0.0002 33 based on literature references
RSS 513 6.02 10
-7
g/cm
3

Table 13. Uncertainty of the external volume of the high-pressure cylinder.

(k =1)
Contrib Comments
External tank volume Relative (10
6
) (%)
apparent mass in air 1 4.61 10
-3
g 0
apparent mass in water - 0.14 g 34 based on water T uncertainties
density of air 513 6.01 10
-7
g/cm
3
0 correlation and meas unc
density of water 20 2 10
-5
g/cm
3
15 correlation and meas unc
expansion 20 9.38 10
-2
cm
3
15 T and P effects
std deviation of repeated meas. 30 0.14 cm
3
35
RSS 51 0.24 cm
3

45

The uncertainties related to the determination of the external volume of the high-pressure
cylinder are listed in Table 13. The 1 10
-6
relative standard uncertainty for the apparent
mass in air is based on uncertainty calculations for the true mass of the weighed cylinder
(discussed later). The uncertainty of the apparent mass in water is based on the recorded
temperatures of water that made the cylinder barely sink or barely float. Uncertainty of
the water temperature for neutral buoyancy of 0.2 K leads to a value for the uncertainty
of the apparent mass in water. This uncertainty is not reported in a relative form since the
apparent mass in water is zero and the relative uncertainty is therefore undefined. The
RSS is calculated using the results in the third column and their sensitivity coefficient.
The water density uncertainty includes thermometer uncertainties, estimates of non-
uniformity of the water temperature, and uncertainty of the water density correlation
obtained from the literature. As previously stated, different values of external volume
were used for the full and empty cylinder due to pressure dilation, but an uncertainty
related to pressure and temperature effects is included here. The collection tank volume is
not very sensitive to this external tank volume, hence larger uncertainty values than
51 10
-6
V
ext
could certainly be tolerated.

Table 14. Uncertainty of the high-pressure cylinder mass measurement.

(k =1)
Contrib Comments
Final mass of high P cylinder Relative (10
6
) (%)
reference masses 0.5 1.91 10
-3
g 26 from NIST Mass Group cal report
room air density 513 6.01 10
-7
g/cm
3
45
reference mass density 0 0 g/cm
3
0 same value for cal. and usage
external cylinder volume 51 0.24 cm
3
1 from unc. of Archimedes method
std deviation of repeated meas. 1 0.002 g 28
RSS 1 0.00376 g

46

The relative standard uncertainties of the cylinder mass measurement (Table 14) are quite
small (1 10
-6
) with the major components being the reference masses, the room air
density (for buoyancy corrections) and the performance of the comparator (repeatability
of the 5 measurements made with the comparator, a type A uncertainty). The complete
mass measurement process was repeated several times for the full and empty cylinder
conditions to assess the repeatability of the process, and these repeated mass values never
differed by more than 1 10
-6
m
c
. This repetition was undertaken since a previously used
cylinder showed changes over time, probably due to absorption of water from room
humidity variations. The uncertainty of the density of the reference masses (7.8 g/cm
3
) is
negligible since the sensitivity of mass measurements to this component is extremely
small. While some of the mass measurement uncertainties for the full and empty cylinder
are correlated and this could be used to reduce the uncertainty of the mass delivered to
the collection tank, this benefit was not utilized since the improvement was not
significant. A table of uncertainty components of the empty cylinder mass measurement
is not presented because it is so similar to Table 14.

Finally, the uncertainty of the collection tank volume is presented in Table 15. The
relative value of the initial collection tank density uncertainty is quite large, but the
sensitivity of the result to this quantity is very small. Recall that uncertainty due to the
effects of room temperature variations on the portion of the collection tank not
submerged has already been incorporated as a temperature uncertainty. The effects of
pressure changes from vacuum to 100 kPa on the tank volume have been considered
analytically and found to be negligible. A small volume (1 cm
3
) of connecting tubing and
valve was necessary to introduce gas from the pressurized cylinder to the collection tank.
This volume was measured with alcohol and a graduated syringe and the uncertainty of
this volume correction was estimated to be 0.5 cm
3
. By far the largest contribution is
from the final gas density, and this in turn is nearly completely traceable to the pressure
measurement.

47

Table 15. Uncertainty of the 677 liter collection tank volume.

(k =1)
Contrib Comments
Collection tank volume Relative (10
6
) (%)
Initial mass of high P cylinder 1 0.00385 g 2 see Table 14
Final mass of high P cylinder 1 0.00376 g 2 see Table 14
initial collection tank gas density 100000 1.14 10
-9
g/cm
3
0
final collection tank gas density 68 4.25 10
-8
g/cm
3
92 see Table 9
expansion due to P and T 0 0 cm
3
0 < 1 ppm
extra volume uncertainty 1 0.50 cm
3
0 related to liq transfer meas.
3
5 6 measurements, 2 gases
RSS 71 48.44 cm
3

Each gravimetric volume determination required one day to complete due to the time
required to achieve ultimate vacuum in the tank (1 Pa), the time for the pressure
transducer to reach thermal equilibrium after filling, and the time to take multiple
cylinder mass measurements separated by an hour each time. The volume measurements
were done 6 times, 4 times with nitrogen and twice with argon. The standard deviation of
these six determinations was 16 10
-6
V
T
, quite good when one considers that two
different gases were used and the uncertainty of their equations of state. The combined
uncertainty of the 677 L collection tank volume is 71 10
-6
V
T
. The 677 L volume was
used as a reference to determine the remaining three unknown volumes via the volume
expansion method.

6.9 Volume of the 34 Liter Collection Tank (Volume Expansion Method)

As explained in section 5.3.2, the volume expansion method is performed by pressurizing
a known volume with pure gas and opening a valve to expand the gas into a previously
evacuated unknown volume. The density changes and known volume are used to
calculate the unknown volume via Eq. 6. In this case the 677 L volume was used to
determine the large inventory volume and the 34 L collection tank volume. Subsequently,
the 34 L collection tank volume was used to determine the small inventory volume.
48

Significant uncertainties of the volume expansion method are related to density and the
measurement of pressure. Many components of the previously given pressure uncertainty
are correlated for a pressure change measurement made with the same pressure
transducer over a short time period (about 1 h). Uncertainties of the piston pressure gauge
and the sensor drift can be considered correlated for the short time period and small
pressure changes involved (5 kPa to 8 kPa out of 200 kPa full scale).

The uncorrelated uncertainties in the measurement of pressure change are due to sensor
non-linearity, resolution, hysteresis, and thermal effects. In order to quantify these
uncertainties for the volume expansion method, the piston pressure gauge was used to
repeatedly provide a reference step change of 6 kPa to the pressure transducer. The
pressure change measured by the pressure transducer was compared to the pressure
change calculated from the pressure standard. Based on these experiments, the
uncertainty of the pressure difference measurement during the 34 L tank volume
determination is 0.3 Pa

Table 16. Uncertainty of the 34 L tank volume determined by the volume expansion
method.

(k =1)
Contrib Comments
6
) (%)
677 L volume 71 48.4 cm
3
38 from gravimetric determination
large inventory volume 1788 1.298 cm
3
0 from volume expansion method
677 L density change 40 3.53 10
-9
g/cm
3
12 dominated by pressure
34 L density change 65 -1.12 10
-7
g/cm
3
32 dominated by pressure
3
2 measured by liq transfer
3
16 13 volume determinations
RSS 116 4.0 cm
3

49

The 34 L tank volume uncertainty is summarized in Table 16. The largest uncertainty
contributions are from the 677 L volume uncertainty, the density change in the 34 L
volume, and the standard deviation of the repeated volume measurements. The
uncertainty of the density change in the 677 L volume is dominated by uncertainty in the
measurement of the pressure change.

6.10 Inventory Volume

The mass change in the inventory volume is negligible since we used the mass
cancellation procedure. However, since there are imperfections in the procedure,
uncertainty components related to the inventory mass change cannot be neglected.
Fortunately, the most significant of these uncertainty components (due to sensor time
constants) are correlated between the start and stop diversions for the methods of
operation used in this flow standard. Other correlated inventory volume uncertainties
include the pressure and temperature sensor calibrations and the differences between
sensed and stagnation values of pressure and temperature. For instance, the inventory
temperature is measured incorrectly low by the same amount at both the start and stop
conditions due to slow sensor response time, so cancellation of temperature uncertainty
occurs. More precisely, the uncertainty of the mass change within the inventory volume
caused by the correlated pressure and temperature uncertainties can be expressed as:

( ) ( )
2
1
2
2 2
2
) ( ) ( ) (
1
) (
1
) (
(
(
|
|
.
|
\
|
+
|
|
.
|
\
|
=
i
I
i
I
i
I f
I
f
I
f
I i
I
i
I
f
I
f
I
I
I
T u
T
P
T u
T
P
P u
T
P u
T ZR
M V
m u (16)

where in this equation, u(P
I
), u(T
I
), and u( m
I
) are the uncertainties of the inventory
pressure, the inventory temperature, and the inventory mass change during the collection,
respectively. Note that if the uncertainties and the initial and final conditions are equal
(i.e. , , T , and ), then the terms within
parentheses cancel, and the flow uncertainty related to the inventory volume is zero.
Equation 16 demonstrates the benefit of matching the initial and final inventory
conditions to optimize the cancellation of correlated uncertainties.
) ( ) (
f
I
i
I
T u T u = ) ( ) (
f
I
i
I
P u P u =
f i
I I
T =
f
I
i
I
P P =
50

Not all of the measurement uncertainties of the inventory volume are correlated. For
perfect inventory mass cancellation, the pressure and temperature measured at specific
locations in the inventory volume must exhibit perfect correlation with the pressures and
temperatures throughout the inventory volume. In this way, the sensor readings
represent the conditions throughout the volume and when the readings at the specific
locations match, the conditions throughout the inventory volume are matched.
Unfortunately, this representative relationship may not exist and there may be
inconsistencies between the pressure and temperature fields between the start and stop
diversions. These inconsistencies may originate from a change in the inventory wall
temperature or from differences in the flow paths between the start and stop diversions.
The spatial inconsistencies are uncorrelated and their magnitude is likely a function of the
mass flow.

Another source of inventory uncertainty, alluded to previously, is due to imperfection in
matching the stop diversion pressure to the start diversion pressure. Recall that the
inventory pressure is recorded while the bypass valve is open (nominally the barometric
pressure), and that the tank filling is stopped when the inventory pressure regains this
same pressure. In this way, the pressure at the beginning of the dead-end time transients
is nearly equal for both diversions and the symmetry of the transients is improved. At
high flows, it becomes more difficult to match these initial pressures in the high speed
data records and a trigger pressure difference that increases with increasing flow
occurs. The size of the trigger pressure difference can be reduced by using faster sensors
and a fast data acquisition and diverter valve control system. An example of the trigger
pressure difference can be seen in Fig. 7, where at times less than zero, the pressure
traces differ by about 1.6 kPa. The trigger pressure difference, coupled with the
historical nature of the inventory pressure measurements due to the sensor time
constant, is another reason that matching of the pressure sensor readings does not
necessarily lead to matching of the actual conditions in the inventory volume. Hence the
trigger pressure difference is another source of inventory uncertainty that scales with the
flow.
51

It is difficult to assess the magnitude of the uncorrelated uncertainties of pressure and
temperature between the start and stop diversions in the inventory volume. Our strategy is
to estimate the uncorrelated inventory uncertainties and then perform experiments (see
Section 7) to confirm that the estimates are reasonable. We assumed that the uncorrelated
inventory uncertainties scale with the flow and are essentially zero at the minimum flow
of each tank. We will assume values of 3 kPa and 9 K (about 3 % of the nominal values)
for the maximum flow of each tank.

Uncertainties related to the fast measurement of pressure with a 700 kPa full scale Heise
Model HPO sensor in the inventory volume are listed in Table 17, separated into
correlated and uncorrelated components.

Table 17. Uncertainties in the inventory pressure measurement, correlated and
uncorrelated, for maximum flow conditions.

(k =1)
Contrib Comments
Relative (10
6
) (kPa) (%)
Inventory Pressure Measurement
Pressure value 100
sensor calibration 3000 0.30 0
sensor time response 1350000 135 100 from inv. model at max flow
sensed vs. stagnation 710 0.071 0 for max flow
RSS (correlated) 1350004 135

sensor repeatability 3.00E+02 0.03 0 from calibration data
spatial inconsistency 3.00E+04 3 100 estimated
RSS (uncorrelated) 30001 3.0

Table 18. Inventory temperature uncertainties, correlated and uncorrelated, for the
maximum flow for each tank.
52

(k =1)
Contrib Comments
Relative (10
6
) (mK) (%)
Inventory Thermocouple
Temperature value 297000
sensor calibration 337 100 0
sensor time response 168350 50000 97 from inventory model
sensed vs. stagnation 50 15 0 for max flow
RSS (correlated) 168351 50000

sensor repeatability 286 85 0 from calibration data
spatial inconsistency 30303 9000 100 estimated
RSS (uncorrelated) 30304 9000

The uncertainties of the measurement of temperature in the inventory volume are listed in
Table 18, again divided into correlated and uncorrelated components. The uncertainties
include calibration, time response, sensed versus stagnation issues, sensor repeatability,
and spatial non-uniformity or inconsistency between start and stop diversions.

Table 19. Uncertainty of the 1000 cm
3
(large) inventory volume.

(k =1)
Contrib Comments
6
) (%)
677 L volume 71 48.4 cm
3
0
-9
g/cm
3
54
inv. vol. density change 120 2.81 10
-7
g/cm
3
1
3
14
std. deviation of repeated meas. 799 0.76 cm
3
32
RSS 1419 1.35 cm
3

The uncertainties for the large and small inventory volumes determined by the volume
expansion method are presented in Tables 19 and 20. Although the relative uncertainties
53

of these volumes are large, the sensitivity of the flow measurement to these volumes is
small due to the inventory mass cancellation scheme and the relatively small size of V
I
.
For the very small pressure changes used for the inventory volume determinations, the
ability of the pressure transducers to measure the pressure change becomes a large
uncertainty issue. The standard deviation of repeated measurements is large for these
volume determinations due to the uncertainty of the small pressure change measurements.
The ratio of the known tank volume to the unknown inventory volume is about 500 to 1,
hence the pressure change is only about 0.3 kPa for an initial pressure of 150 kPa. The
uncertainty of the pressure change measurement (0.3 Pa) is based on the previously
described experiments using the pressure standard. The instrument resolution (1.3 Pa) is
also a concern for such small pressure changes measured with a 200 kPa full-scale
transducer. Although the noise in the pressure data is greater than 1.3 Pa, averaging the
pressure measurements over 30 seconds reduces the resolution to an acceptable level. For
the inventory volumes, the extra volume includes tubing and valves used to introduce
gas to the system and volume changes caused by the actuation of the diverter valves. The
extra volume corrections are based on liquid transfer into or out of the volume by pipette,
and /or by dimensional calculations.

Table 20. Uncertainty of the 75 cm
3
(small) inventory volume.

(k =1)
Contrib Comments
6
) (%)
34 L volume 116 4.0 cm
3
0
-9
g/cm
3
24
inv. vol. density change 120 2.31 10
-7
g/cm
3
1
3
66
std. deviation of repeated meas. 511 0.038 cm
3
9
RSS 1661 0.12 cm
3

Tables 21 and 22 show the uncertainty in the mass change in the inventory volume for the
largest flows in the 34 L and 677 L systems. The relative inventory uncertainty is
54

extremely large since the mass change is nearly zero. The mass change was calculated
using the same 3 kPa and 9 K uncorrelated uncertainty values previously discussed.

Table 21. Uncertainty of the inventory mass change for the 677 L system.

(k =1)
Contrib Comments
Inventory mass change Relative (10
6
) (%)
initial density 42643 6.29 10
-5
g/cm
3
50 3 kPa and 9 K uncorr. inv. unc.
final density 42643 6.24 10
-5
g/cm
3
volume 1419 1.35 cm
3
0 see Table 19
RSS - 8.42 10
-2
g

Table 22. Uncertainty of the inventory mass change for the 34 L system.

(k =1)
Contrib Comments
Inventory mass change Relative (10
6
) (%)
initial density 42643 6.29 10
-5
g/cm
3
final density 42643 6.24 10
-5
g/cm
3
volume 1661 0.123 cm
3
0 see Table 20
RSS - 6.58 10
-3
g

The sub-component uncertainties of gas density, volume, collection time, and inventory
mass change above have been combined in Section 6.2 to give the uncertainty in mass
flow.

7 Experimental Verification of the Uncertainty of the 34 L and 677 L PVTt Gas
Flow Standards

55

Throughout the range 3 L/min to 110 L/min, flows were measured independently using
the 34 L and the 677 L collection systems, using a set of critical flow venturis (CFV) as
transfer standards, and the two systems agreed within a relative difference of 150 10
-6
.
Double diversions were used to evaluate the 677 L system over a range of 300 L/min to
1600 L/min, and the relative differences between single and double diversions were less
than 75 10
-6
. These differences are within our expectations based on the uncertainty
analysis presented in Section 6.

7.1 Comparison of the 34 L and 677 L Flow Standards

To test both systems, we performed comparisons between the two flow standards over the
range of flows where they could both be used: 3 L/min < m < 110 L/min (3.6 g/min to
138 g/min of nitrogen). The collections ranged from as short as 18 s to more than 4 h.
&

Figure 11 shows the difference in the discharge coefficients of several critical flow
venturis as measured by both the 34 L and 677 L systems, plotted versus flow. The
agreement between the two flow standards is 150 10
-6
over the entire range tested.
The throat diameters of the venturis used for the comparisons ranged between 0.2921 mm
and 1.7 mm. The comparisons were done with the same pressure and temperature sensors
associated with the venturi during the testing on both flow standards in order to reduce
some possible sources of discharge coefficient differences. Numerous collections were
made for each tank at each flow to confirm stability of the conditions at the meter under
test.
m&

How well should the two systems agree? The difference between the discharge
coefficients measured by the two PVTt systems should be less than the root sum square
(RSS) of the uncertainties of the two standards, especially when one considers that the
uncertainties due to pressure and temperature measurements are correlated between the
two standards. For the lowest flows of the comparison range, the uncertainties originating
from the inventory volume are quite small for both systems and the observed differences
between them are dominated by tank volume uncertainties. From Fig. 11 it can be seen
56

that the two systems differ by about 100 10
-6
for flows less than 20 L/min. The RSS
of the two relative standard volume uncertainties is 137 10
m&
-6
(k = 1).

Figure 11. Difference in the discharge coefficient of critical flow venturis calibrated on
both the 34 L and 677 L flow standards versus flow and the inverse of the collection time
for the 34 L tank. Also plotted is a linear best fit of the data.

At the higher flows of the comparison range, the uncertainties associated with the
transient conditions in the inventory volume should be negligible in the 677 L system, but
growing with increasing flow for the 34 L system. Because the collection times were
1/20
th
as long when using the smaller tank, any timing error (or, equivalently any
imperfection of the mass cancellation technique) was 20 times more important when
using the smaller tank. Figure 11 suggests that the inventory uncertainties cause the flow
standard to read too high as the flow is increased, changing the difference by about
200 10
-6
over the range of flows compared. This value is comparable to the relative
uncertainty of 135 10
m&
-6
(k = 1) contributed by the inventory volume at the highest
flows in the 34 L system (61% of 219 10
-6
from Table 3). Therefore, the differences
observed in the comparison are in reasonable accordance with the uncertainty analysis.

57

Figure 11 also shows the tank comparison results from the perspective of time
measurement uncertainty rather than the mass. We interpreted the comparison results
using the simplified model , where m is the mass collected and t is the collection
time using the 34 L tank. Making the assumption that there exists a constant error in the
mass measurement, m, for all flows (say due to a tank volume error) and a constant error
in time measurement, t, for all flows (say due to the time constant of the inventory
pressure sensor), we can derive a simple linear model for the error in mass flow
measurement, i.e.,
t m m / = &

t
t
m
m
m
m
=
&
&
. (17)

Since the mass collected and the mass error are both essentially constant, this model
suggests that a linear function of the inverse of the collection time should fit the tank
comparison data. Such a linear function fits reasonably well, as shown in Fig. 11, and its
slope implies a constant timing error of 4 ms. A timing error of this order is not surprising
because the timing is based on a pressure sensor with a time constant of approximately
20 ms. The standard deviation of a measurement from the best fit line is 24 10
-6
. m&

The comparison results could be the basis for corrections to the 34 L system. The
intercept of Fig. 11 could be used to change the volume of the 34 L tank and improve the
agreement between the two systems at low flows. The slope could be used to improve
agreement at higher flows. This approach offers the possibility of reducing the
comparison differences to zero with a standard deviation of 24 10
-6
. However, these
corrections have not been made at the present time for several reasons. The comparison
results are consistent with the uncertainty analysis. Also, despite the success of Eq. 17,
we feel that the inventory uncertainties are more related to pressure and temperature than
time, so it is more appropriate to make improvements in those measurements to reduce
the slope in the comparison data. The volume (or offset) differences can be improved by
repeating and refining the volume expansion process used to determine the 34 L tank
m&
58

size. This approach adheres to the definition of primary standard for both systems since
neither one has been calibrated by a flow measurement against some other flow standard.

During some of the comparison flows, we noticed that the pressure downstream of the
critical flow venturi was significantly higher in the 34 L system than in the 677 L system
(108 kPa vs. 100 kPa) due to the smaller tube size and resultant higher pressure drop. For
some of our venturis (with relatively short diffusers) this pressure difference caused slight
changes in the upstream pressure (and the discharge coefficient), even at conditions well
above the critical pressure ratio. Therefore, our assumption that for the same throat
Reynolds number, the discharge coefficient of the venturi is independent of the
downstream pressure may not be perfectly valid. We suspect that some of the differences
observed between the tanks in Fig. 11 are due to the venturis (even though long diffusers
were used).

In one series of experiments, the trigger pressure difference was purposely varied over a
range from 2 kPa to 27 kPa at a constant flow of 82 L/min in the 34 L system. The
purpose of the test was to measure the dependence of the venturi discharge coefficient on
the trigger pressure difference and hence assess its influence on the inventory volume
uncertainties. The tests showed a relative change of 10 10
-6
in discharge coefficient for
each 1 kPa change in the trigger pressure difference. Since the largest trigger pressure
difference is less than 3 kPa in the present system, this effect is expected to contribute
only 30 10
-6
to the flow uncertainty. Therefore the major contributor to the
inventory uncertainty appears to be spatial inconsistency of the pressure and temperature
fields between the start and stop diversions or some other, unknown flow dependent
uncertainty source.
m&

7.2 Multiple Diversions in the 677 L Flow Standard

To confirm that the uncertainty analysis for the inventory volume of the 677 L collection
system was reasonable, we performed CFV calibrations at identical flows following two
different protocols. In the first protocol, the inventory volume was dead-ended at the
59

beginning and end of the collection interval in the usual manner. In the second protocol,
the collection interval was divided into two subintervals, i.e. each flow measurement had
two start and stop diversions. The intermediate dead-end times were set up so that the
pressure transients in the inventory volume still permitted the mass cancellation
procedure. Breaking the collection into two subintervals has the effect of doubling the
uncertainty contribution from the inventory volume. The CFV discharge coefficients
from the two protocols were compared to assess the magnitude of the uncertainties
introduced by the inventory volume and the flow diversion process. Three flows between
300 L/min and 1600 L/min were tested and the differences in discharge coefficient were
all less than 75 10
-6
as shown in Table 23. m&

Table 23. Differences in CFV discharge coefficients (C
d
) for two and one diversion in the
677 L flow standard.
Flow (L/min) [C
d
(2 diversions) C
d
]/C
d
10
6

300 53 25
700 -27 31
1600 75 122

8 Summary

We have given a description of the gas flowmeter calibration services that use the 34 L
and 677 L PVTt primary standards within the NIST Fluid Flow Group. These PVTt
standards were designed to calibrate flowmeters between 1 L/min to 2000 L/min, with
uncertainties between 0.02 % and 0.05 %. See Table 1 for more details about the test
capabilities of the calibration service such as available gases and pressure at the test
section. The 34 L PVTt standard has been used at flows as low as 0.025 L/min.

The intended audience for this document is potential flowmeter calibration customers,
hence we have attempted to address their likely questions. We have described the method
of operation of the primary standard and have given the uncertainty analysis for a mass
flow measurement (see Tables 2 and 3). We have given a description of the standard gas
60

flow calibration, including a sample calibration report, as well as sources for instructions
for submitting a flowmeter for calibration.

The 34 L and 677 L PVTt flow standards have several novel features. The collection
tanks are immersed in a water bath that matches the nominal room temperature and is
stable and uniform to 1 mK. The collection tanks are divided into sections of small
enough diameter that the gas inside them achieves thermal equilibrium with the
surrounding water bath in 20 minutes or less. This reduces the contribution of
temperature to the flow measurement uncertainty to a low level.

Uncertainties related to the inventory volume and the diversion of gas into the collection
tank at the start and stop of a flow measurement have been studied in great detail. A
thermodynamic model of the inventory volume during diversion has been utilized to
understand the large pressure and temperature transients and the importance of sensor
time constants on the flow measurement uncertainty. The flow standard is operated to
achieve mass cancellation in the inventory volume, thereby taking advantage of
correlated sensor uncertainties to minimize uncertainty contributions from the inventory
volume. The uncertainty contributions of the inventory volume have been considered
from both the time and mass perspectives.

The volumes of the collection tanks were measured by two methods, a gas gravimetric
method and a volume expansion method. Six gravimetric determinations of the 677 L
collection tank volume made with nitrogen and argon agree with a standard deviation of
16 10
-6
V
T
.

A detailed uncertainty analysis for the gas flow standard has been presented The analysis
started with the basis equation utilized to calculate flow in a PVTt system and followed
the propagation of uncertainties method suggested by international standards. The
uncertainties of the sub-components have been examined at a fundamental level.

61

The uncertainty analysis shows that the 677 L system measures mass flow with an
uncertainty between 200 10
-6
and 300 10 m&
-6
for a pure gas like nitrogen or
argon. The higher uncertainty applies to higher flows as the inventory transients and the
related uncertainties grow larger. For the 34 L tank and pure gases, the uncertainties
range from 440 10
m&
-6
to 270 10 m&
-6
m. The uncertainties are larger for the 34 L tank
because the tank volume uncertainty is relatively greater and the inventory volume is
relatively larger for the small system. For pure gas measurements, the largest sources of
uncertainty can be traced to pressure measurement (about 70 10
&
-6
P) which is the major
contributor to gas density and tank volume uncertainties. For air flow measurements
using gas from the existing compressor and drier, mass flow uncertainties are about 500
10
-6
for both standards and the major contribution is the uncertainty in the moisture
content of the air.
m&

Comparisons between the 34 L and 677 L standards from 3 L/min to 100 L/min show
agreement within 150 10
-6
mor better. Experiments using single diversions (normal
operation) and double diversions to the collection tank were used to validate the
uncertainty estimates of the 677 L inventory volume and the differences between these
two methods were less than 75 10
&
-6
m. The evaluation results along with comparisons
to previously existing gas flow standards support the uncertainty statements for the new
standards.
&

9 References

Carslaw, H. S. and Jaeger, J. C., Conduction of Heat in Solids, 2
nd
Edition, Clarendon
Press, Oxford, pp 198-201, 1946.

Coleman, H. W. and W. G. Steele, Experimentation and Uncertainty Analysis for
Engineers, John Wiley and Sons, 2
nd
edition, 1999.

Dymond, J. H. and Smith, E. B., The Virial Coefficients of Pure Gases and Mixtures: A
Critical Compilation, Clarendon Press, Oxford, 1980.

Harris, R. E. and Johnson, J. E., Primary Calibration of Gas Flows with a Weight / Time
Method, Proceedings 2
nd
International Symposium on Fluid Flow Measurement, Calgary,
Canada, pp. 347358, June, 1990.
62

Hilsenrath, J., Beckett, C. W., Benedict, W. S., Fano, L., Hoge, H. J., Masi, J. F., Nuttall,
R. L., Touloukian, Y. S., and Woolley, H. W., Tables of Thermal Properties of Gases,
NBS Circular 564, 1955.

International Organization for Standardization, International Vocabulary of Basic and
General Terms in Metrology, 2
nd
edition, 1993.

International Organization for Standardization, Guide to the Expression of Uncertainty in
Measurement, Switzerland, 1996.

Ishibashi, M., Takamoto, M., and Watanabe, N., New System for the Pressurized Gas
Flow Standard in Japan, Proceedings of International Symposium on Fluid Flow
Measurements, AGA, 1985.

Jaeger, J. B. and Davis, R. S., A Primer for Mass Metrology, NBS Special Publication
700-1, National Bureau of Standards, Gaithersburg, Maryland, Eq. 20a, pp. 22, 1984.

Johnson, A. N., Wright, J. D., Moldover, M. R., and Espina, P. I., Temperature
Characterization in the Collection Tank of the NIST 26 m3 PVTt Gas Flow Standard,
Metrologia, 40, 211216, 2003.

Kegel, T., Uncertainty Analysis of a Volumetric Primary Standard for Compressible
Flow Measurement, 3
rd
International Symposium of Fluid Flow Measurement, San
Antonio, TX, USA, 1995.

Lemmon, E. W., Jacobsen, R. T., Penoncello, S. G., and Friend, D. G., Thermodynamic
Properties of Air and Mixtures of Nitrogen, Argon, and Oxygen from 60 to 2000 K at
Pressures to 2000 MPa, J. Phys. Chem. Ref. Data, 29, (3), 331-362, 2000.

Lemmon, E. W., McLinden, M. O., and Huber, M. L., Refprop 23: Reference Fluid
Thermodynamic and Transport Properties, NIST Standard Reference Database 23,
Version 7, National Institute of Standards and Technology, Boulder, Colorado, 2002.

Marshall, J. L., NIST Calibration Services Users Guide 1998, NIST Special Publication
250, January, 1998.

Moldover, M. R., Trusler, J. P. M., Edwards, T. J., Mehl, J. B., and Davis, R. S.,
Measurement of the Universal Gas Constant R Using a Spherical Acoustic Resonator,
NIST J. of Res., 93, (2), 85143, 1988.

Olsen, L. and Baumgarten, G., Gas Flow Measurement by Collection Time and Density
in a Constant Volume, Flow: Its Measurement and Control in Science and Industry,
Instrument Society of America, Pittsburgh, PA, USA, pp. 12871295, 1971.

63

Patterson, J. B. and Morris, E. C., Measurement of Absolute Water Density, 1 C to
40 C, Metrologia, 31, pp. 277-288, 1994.

Trusler, J. P. M., Wakeham, W. A., and Zarari, M. P., Model Intermolecular Potentials
and Virial Coefficients Determined from the Speed of Sound, Molecular Physics, 90, (5),
pp. 695703,1997.

Wright, J. D., The Long Term Calibration Stability of Critical Flow Nozzles and Laminar
Flowmeters, National Conference of Standards Laboratories Conference Proceedings,
Albuquerque, NM, USA, pp. 443-462, 1998.

Wright, J. D. and Mattingly, G. E., NIST Calibration Services for Gas Flow Meters:
Piston Prover and Bell Prover Gas Flow Facilities, NIST Special Publication 250-49,
National Institute of Standards and Technology, Gaithersburg, Maryland, August, 1998.

Wright, J. D., Gas Flow Measurements and Standards: Development and Validation of a
PVTt Primary Gas Flow Standard, Doctoral thesis to Kogakuin University, Tokyo,
Japan, September, 2003A.

Wright, J. D., What Is the Best Transfer Standard for Gas Flow?, FLOMEKO,
Groningen, the Netherlands, May, 2003B.

64

NIST Test Number: 836-123456-03-01 Page 1 of 6
Calibration Date: July 2, 2003
Appendix: Sample Calibration Report

REPORT OF CALIBRATION

FOR

A CRITICAL FLOW NOZZLE

July 7, 2003

Mfg.: Meter Builders, Inc.
Serial No: 1234
Throat Diameter: 0.125 in (0.3175 cm)

submitted by

Flowmasters, Inc.
Metertown, MD

Purchase Order No. A123 dated June 23, 2003

The flow meter identified above was calibrated by flowing filtered dry air at a constant
rate through the flow meter and then into a volumetric prover (the NIST 677 L PVTt
standard). The PVTt standard determines mass flow, , by measuring the change in
density of gas diverted into a known volume for a measured period of time.
m&
1
The flow
meter was tested at five flows and at each flow, three (or more) measurements were
gathered on two different occasions and used to produce averages at each of these flows.
As a result, the tabulated data for this test are averages of six or more individual
calibration measurements.

The nozzle was installed in an assembly that meets the ISO Standard for critical nozzles
2

and a photograph of the installation is shown in Figure 1. The nozzle temperature (
and pressure were measured with NIST sensors, (Keithley SN 687848, thermistor #3,
and Paroscientific SN 75971). Stagnation temperature, T was calculated from the
measured temperature via the following equation, using a recovery factor,
)
1
T
) (
1
P
0
r , of 0.75:
( ) T T M r
0 1
2
1
1
2
1 = +

(1)

1
Wright, J. D., Johnson, A. N., and Moldover, M. R., Design and Uncertainty Analysis for a
PVTt Gas Flow Standard, J. Res. Natl. Inst. Stand. Technol., 108, 21-47, 2003.
2
Measurement of Gas Flow by Means of Critical Flow Venturi Nozzles, ISO 9300: 1990 (E),
International Organization for Standardization, Geneva, Switzerland, 1990.
REPORT OF CALIBRATION Gas Flow Meters
Flowmasters, Inc. Purchase Order No. A123
and the stagnation pressure, , was calculated via the equation:
0
P
P P M
0 1
2
1
1
1
2
= +

(2)
where is the specific heat ratio and M is the Mach number in the approach pipe, both
based on and T . The largest of these corrections is 0.0067% for pressure.
1
P
1

Figure 1. A photograph of the flow meter installation.

The Reynolds number is included in the tabulated data and it was calculated using the
following expression:

=
d
m
Re
& 4

(3)
where is the mass flow of gas, is the nominal nozzle throat diameter, and & m d is the
gas viscosity, all in consistent units so that is dimensionless. The viscosity of the gas
was calculated using best-fit equations which are based on the NIST gas properties
database
Re
3
:
3
0 3
2
0 2 0 1 0
T a T a T a a + + + =
(4)

3
Lemmon, E. W., McLinden, M. O., and Huber, M. L., Refprop 23: Reference Fluid
Thermodynamic and Transport Properties, NIST Standard Reference Database 23, Version 7,
National Institute of Standards and Technology, Boulder, Colorado, 2002.
where has units g / (cm s), and T is in K. The polynomial coefficients, a are in turn
polynomial functions of pressure

0 i
3
0 3
2
0 2 0 1 0
P b P b P b b a
i i i i
i
+ + + = (5)
where is in kPa and the coefficients for dry air are:
0
P
i
b
j
-1.48711900E-06 2.36550580E-08 -4.74728350E-11 3.80951780E-14
8.03433280E-07 -2.06447690E-10 4.86819380E-13 -3.87442880E-16
-7.32108840E-10 6.60598760E-13 -1.65670740E-15 1.31012510E-18

( ) =
i
b
j

4.44444440E-13 -7.23905720E-16 1.87289560E-18 -1.47306400E-21
(6)
where i number and number. row = column = j

The discharge coefficient C was calculated from the expression:
d
C
m R T
d P C
d
=

4
0
2
0
&
(7)
where R is the gas constant (the universal gas constant, 8.314471 J / (mol K), divided by
the gas molecular weight, 28.9646431 g/mol). The critical flow factor, C
, was
calculated from the expression:
1
1
*
1
2
+
+
=

C
(8)
where is the specific heat ratio, calculated using the same system of polynomials
described above with the following coefficients:
1.40035540E+00 2.31594790E-04 2.32361020E-08 3.48369500E-13
-2.51273630E-06 -1.63018610E-06 -2.01773710E-10 -4.02631530E-15
9.40776360E-08 4.19213690E-09 5.90056690E-13 1.42526470E-17
( ) =
i
b
j

-3.01845240E-10 -3.79086500E-12 -5.81559640E-16 -1.61335590E-20
(9)

The calibration results are presented in the following table and figure.
0.9780
0.9800
0.9820
0.9840
0.9860
0.9880
0.9900
35,000 85,000 135,000 185,000 235,000 285,000
Re
Cd
Figure 2. Calibration results for 0.125 in (0.3175 cm), SN 1234 nozzle

Table 1. Calibration results for 0.125 in (0.3175 cm), SN 1234 nozzle from the 677 L
PVTt standard.
0
P
[kPa]
0
T

[K]
m&
[lb
m
/s]
C
[]
Re
[]
d
C
[]
r
U
[%]
104.07 295.73 0.004189 0.68505 41471.1 0.9808 0.08
249.69 295.26 0.010117 0.68550 100147.9 0.9859 0.08
399.90 294.87 0.016256 0.68597 160854.2 0.9877 0.08
546.52 294.63 0.022261 0.68643 220116.4 0.9887 0.08
685.88 294.49 0.027979 0.68686 276392.8 0.9893 0.08

An analysis was performed to assess the uncertainty of the results obtained for the meter
under test.
4, 5, 6
The process involves identifying the equations used in calculating the

4
International Organization for Standardization, Guide to the Expression of Uncertainty in
Measurement, Switzerland, 1996 edition.
5
Taylor, B. N. and Kuyatt, C. E., Guidelines for Evaluating and Expressing the Uncertainty of
NIST Measurement Results, NIST TN 1297, 1994 edition.
calibration result (measurand) so that the sensitivity of the result to uncertainties in the
input quantities can be evaluated. The 67 confidence level uncertainty of each of the
input quantities is determined, weighted by its sensitivity, and combined with the other
uncertainty components by root-sum-square to arrive at a combined uncertainty (U ).
The combined uncertainty is multiplied by a coverage factor of 2.0 to arrive at an
expanded uncertainty (U ) of the measurand with approximately 95 confidence level.
%
c
e
%

As described in the references, if one considers a generic basis equation for the
measurement process, which has an output, , based on input quantities, , y N
i
x
) , , , (
2 1 N
x x x y y K =
(10)
and all uncertainty components are uncorrelated, the normalized expanded uncertainty is
given by,
( ) ( ) ( )
=
|
|
.
|
\
|
= =
N
i i
i
i
c e
x
x u
s k
y
y U
k
y
y U
1
2
2
(11)
In the normalized expanded uncertainty equation, the u are the standard
uncertainties of each input, and are their associated sensitivity coefficients, given
by,
s )' x (
i
s ' s
i
y
x
x
y
s
i
i
i
=
(12)
The normalized expanded uncertainty equation is convenient since it permits the usage of
relative uncertainties (in fractional or percentage forms) and of dimensionless sensitivity
coefficients. The dimensionless sensitivity coefficients can often be obtained by
inspection since for a linear function they have a magnitude of unity.

For this calibration, the uncertainty of the discharge coefficient has components due to
the measurement of the mass flow by the primary standard, ( ) % 025 . 0 = m u & ,
7
as well as
the pressure, u , and temperature, ( ) % 02 . 0 = P ( ) % 03 . 0 = T u , measurements at the meters
under test. The sensitivity coefficients for mass flow and pressure are 1, and the
sensitivity coefficient for temperature is . This uncertainty analysis assumes that the
user will use the same values for the throat diameter and the critical flow factor given
herein and that the uncertainties in these quantities are correlated and cancel.

To measure the reproducibility
8
of the test, the standard deviation of the discharge

6
Coleman, H. W. and Steele, W. G., Experimentation and Uncertainty Analysis for Engineers,
John Wiley and Sons, 2
nd
ed., 1999.
7
Wright, J. D., Johnson, A. N., and Moldover, M. R., Design and Uncertainty Analysis for a
PVTt Gas Flow Standard, J. Res. Natl. Inst. Stand. Technol., 108, 21-47, 2003.
8
Reproducibility is herein defined as the closeness of agreement between measurements with the
flow changed and then returned to the same nominal value.
coefficient at each of the nominal flows was used to calculate the relative standard
uncertainty (the standard deviation divided by the mean and expressed as a percentage).
The reproducibility was root-sum-squared along with the other uncertainty components to
calculate the combined uncertainty. Using the values given above results in the expanded
uncertainties listed in the data table and shown as error bars in the figure.

For the Director,
National Institute of Standards and Technology

Dr. Pedro I. Espina
Leader, Fluid Flow Group
Process Measurements Division
Chemical Science and Technology Laboratories

NIST Sp250 63 PVTT GasFlowCal

Uploaded by

Copyright:

Available Formats

NIST Sp250 63 PVTT GasFlowCal

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

NIST Sp250 63 PVTT GasFlowCal

Uploaded by

Copyright:

Available Formats

NIST Measurement Services:

Gas Flowmeter Calibrations with the 34 L and 677 L PVTt

Table 1 presents the flow ranges covered by the

(1/8 to 1), A/N 37 degree flare (1/4 to 1), national pipe

You might also like