Dr.

Rainer Kurz is Manager of Systems

Analysis and Field Testing for Solar
Turbines, Incorporated, in San Diego,
California. His organization is responsible
for predicting gas compressor and gas
turbine performance, for conducting appli-
cation studies, and for field performance
tests on gas compressor and generator
packages.
Dr. Kurz attended the University of the
Federal Armed Forces in Hamburg,
Germany, where he received the degree of a Dipl.-Ing. in 1984 and
the degree of a Dr.-Ing. in 1991. He has authored numerous publi-
cations in the field of turbomachinery and fluid dynamics, is an
ASME Fellow, and a member of the Turbomachinery Symposium
Advisory Committee.
Dr. Klaus Brun is a Principal Engineer
at Southwest Research Institute, Mechani-
cal and Materials Engineering Division, in
San Antonio, Texas. His experience in-
cludes positions in business development,
project management, engineering, and
marketing at Solar Turbines, General
Electric, and ALSTOM Power. He is the
inventor of the Single Wheel Radial Flow
Gas Turbine and co-inventor of the
Planetary Gear Mounted Auxiliary Power
Turbine. He has authored over 30 papers on turbomachinery,
given numerous technical lectures and tutorials, and published a
textbook on Gas Turbine Theory. He is the incoming Chair of the
ASME-IGTI Oil & Gas Applications Committee, a member of the
Gas Turbine Users Symposium Advisory Committee, and a past
member of the Electric Power and Coal-Gen Steering Committees.
Dr. Brun received his Ph.D. and M.Sc. degrees (Mechanical and
Aerospace Engineering, 1995, 1992) from the University of
Virginia, and a B.Sc. degree (Aerospace Engineering, 1990) from
the University of Florida.
ABSTRACT
Field testing of compressor packages driven by gas turbines or
electric motors requires the accurate determination of efficiency,
flow, head, and power in sometimes less than ideal working envi-
ronments. This paper discusses field performance testing of
centrifugal compressors and measurement uncertainties one can
expect when following appropriate test guidelines. The primary
focus is on compressor performance for applications where the
compressor is controlled by speed variation. Special consideration
will be given to testing of compressors with two sections and mul-
ticompressor trains. The paper addresses the issues of planning and
organization of field tests and necessary instrumentation, but
primarily focuses on data reduction, data correction, test uncer-
tainty, and the interpretation of test data.
INTRODUCTION
Centrifugal gas compressors are used in many applications in
the oil and gas industry, such as in pipelines, for gas gathering, gas
reinjection, gas lift, gas storage, in onshore as well as offshore
environments. The predominant driver for these compressors is
two shaft gas turbines, and, to some extent, variable speed electric
motors. Both types of drivers have in common that the speed of the
compressor can be varied easily over a large range. To some extent,
variable speed electric motors are used in similar tasks.
Performance testing of compressor packages is becoming increas-
ingly frequent because economic pressures demand that the
efficiency, power, fuel flow, capacity, and head of an installation be
verified to assure a projects return on investment. Test results may
have significant financial implications for the compressor and gas
turbine manufacturers and their customers. They may be the basis
of future decisions on plant modifications or extensions, or may
serve as baseline data for monitoring purposes. Field tests also
provide the operator and the equipment manufacturers with infor-
mation complementary to the data collected during factory testing.
Thus, for the end user and the manufacturer, an accurate determi-
nation of the package field performance is critical. This paper
discusses problems and challenges related to the field performance
testing of gas turbine or electric motor driven, variable speed com-
pressor sets.
Insights from over 100 field performance tests on centrifugal
compressors are described. Such field tests provide the user with
valuable operation and maintenance data and the manufacturer
with information complementary to the data gathered through
factory testing.
This paper also describes in detail how raw test data (i.e.,
measured pressures, temperatures, flows, and speeds) are
converted into meaningful data that can be used for evaluating the
performance of a gas turbine or a gas compressor. Details on
instrumentation will not be provided by this paper, but can be
found elsewhere (Kurz, et al., 1999). The data reduction is
explained based on the basic relationships of pressure, tempera-
ture, flow, and head, as well as the operational characteristics of
53
SITE PERFORMANCE TEST EVALUATION FOR
GAS TURBINE AND ELECTRIC MOTOR DRIVEN COMPRESSORS
by
Rainer Kurz
Manager of Systems Analysis and Field Testing
Solar Turbines Incorporated
San Diego, California
and
Klaus Brun
Principal Engineer
Southwest Research Institute
San Antonio, Texas
centrifugal gas compressors, with special consideration of test
uncertainties. Methods of data correction are explained. A few
remarks on general procedures, valid for any type of test, including
the application of test uncertainties are included.
Assume a performance test has just been concluded and columns
of pressures, temperatures, and speeds have been measured. Fuel
and process gas compositions are known. The test was conducted
using a reasonably accurate test protocol (ASME PTC 10, 1997;
Kurz, et al., 1999). The test was conducted to gain information
about the performance of the gas turbine driven centrifugal com-
pressor set. The question is now (actually, this question should
have been asked prior to the test), what to do with all that data? In
this paper, the authors want to review methods to reduce the raw
test data to information about the performance of the compression
system.
SYSTEM DESCRIPTION AND PERFORMANCE MAPS
The system discussed consists of a two-shaft gas turbine or a
variable speed electric motor driving one or more centrifugal com-
pressors (Figure 1). Possibly, a gearbox is used between the driver
and the driven equipment or between compressors. Figure 2
outlines the test data taken on a single compressor:

Suction pressure and temperature

Discharge pressure and temperature

Flow and gas composition

Speed
Figure 1. Compressor Station with Three Gas Turbine Driven
Centrifugal Compressors.
Figure 2. Test Data for Centrifugal Compressors.
Usually, the manufacturer of the gas compressor provides a set
of performance maps for the equipment (Figure 3).
For the compressor, there are maps showing suction or discharge
pressure over the standard flow for various operating speeds and
Figure 3. Typical Gas Compressor Performance Map.
maps that show isentropic or polytropic head over actual compres-
sor inlet flow, also for various operating speeds. For the purpose of
this test data evaluation, only the head-versus-actual flow maps are
useful because the pressure-versus-flow maps are only valid if the
compressor suction conditions (pressure, temperature, gas compo-
sition) are exactly the same as indicated on the map.
The authors will use these maps, which are based on the manu-
facturers predictions of the equipment performance, to compare
their test data against.
DATA REDUCTIONHOWTO GET POWER,
FLOW, AND EFFICIENCY FROM ALL THESE
PRESSURES AND TEMPERATURES
Single Gas Compressor
The flow through the compressor (as well as the gas turbine fuel
flow) has been measured using one of several possible flow
measuring devices. If the device is a flow orifice, the relationship
between the flow and the measured temperatures and pressures is
as follows:
(1)
C and E are discharge coefficients and the velocity approach factor,
respectively, and d is the orifice throat diameter. The coefficients
can be determined either from the orifice manufacturers data
sheets or from such codes as ASME PTC 19.5 (1971) or ISO 5167
(1980).
Other devices (venturi, pitot-type probes, etc.) have formally
similar relationships between the flow and the measured pressures
and temperatures. Devices that do not use the pressure differen-
tials, such as turbine flowmeters, ultrasonic flowmeters, and
coriolis flowmeters will be supplied by their respective manufac-
turers with appropriate methods to calculate actual flow and
standard flow or mass flow. It must be noted that, while the
standard flow through the flow measuring device and the com-
pressor are identical (as long as no leaks or flow divisions are
present), the actual flow will be different because the pressure and
temperature at the compressor nozzle will be different from the
actual flow through the flow measuring device. For now, one can
state that any flow measuring device will provide one with either
the standard flow (SQ) or the mass flow (W). (Standard conditions
can be 60F and 14.70 psia, 60F and 14.73 psia, or 15C (59F)
PROCEEDINGS OF THE THIRTY-FOURTH TURBOMACHINERY SYMPOSIUM 2005 54
W C E d p =

4
2
2
1

and 760 mm Hg (14.7 psia). Many countries use normal condi-

tions, such as 0C (273.15 K, 32F) and 1013.25 mbar (1 atm, 14.7
psia.)
The knowledge of pressure and temperature at the compressor
inlet nozzle enables us to calculate the actual flow (Q
s
) with:
(2)
The density in the above equations has to be determined using
an equation of state. The general relationship is:
(3)
The compressor head (H) can be determined from the measure-
ment of suction and discharge pressure and temperature. The
relationship between the pressure, temperature, and the enthalpy
(h) is defined by the equations of state described below.
By using the equations of state, the relevant enthalpies for the
suction, the discharge, and the isentropic discharge state can be
computed. The isentropic head (H
*
) is:
(4)
The actual head H is:
(5)
[In US units, the enthalpy difference (Btu/lb) has to be multiplied
by the mechanical equivalent of heat (778.3 ft lb/Btu) to get the
head (ft lb/lb)].
The isentropic and polytropic efficiencies then become:
(6)
It should be noted that the polytropic efficiency is defined
similarly to the isentropic efficiency, using the polytropic process
instead of the isentropic process for comparison. The actual head,
which determines the absorbed power, is not affected by the
selection of the polytropic or isentropic process. However, the
isentropic head is unambiguously defined by the users process
data (i.e., gas composition, suction pressure and temperature,
discharge pressure), while the polytropic head for full definition
additionally requires the compressor efficiency or the discharge
temperature.
The task of gas compression is to bring gas from a certain
suction pressure to a higher discharge pressure by means of
mechanical work. The actual compression process is often
compared to one of two ideal processes.
The compression process is isentropic if the process is friction-
less and no heat is added to or removed from the gas during
compression. With these assumptions, the entropy of the gas does
not change during the compression process, and the process is
reversible. Because there is no heat transfer across the system
boundaries, the process is often referred to as reversible adiabatic.
The polytropic compression process is like the isentropic cycle
reversible, but it is not adiabatic. It can be described as an infinite
number of isentropic steps, each interrupted by isobaric heat
transfer. This heat addition allows the process to yield the same
discharge temperature as the real process.
Either process can be used for the definition of the operating
point and the efficiency of the compressor. It should be noted that
the absorbed compressor power is not impacted by this choice,
because it solely depends on the actual head. Since the site test will
in many instances be performed at conditions very similar to the
design point, the isentropic definition has the advantage that the
isentropic head (and thus the operating point) is fully defined by
the process conditions (gas, suction and discharge pressure, and
suction temperature), while the polytropic head additionally
depends on the compressor efficiency, which in itself is the subject
of the test.
The comparison of the actual process with a polytropic process,
as opposed to an isentropic process, has the advantage that the effi-
ciency for an aerodynamically similar point is less dependent on
the actual pressure ratio. However, it has the disadvantage that the
polytropic head for a given set of operating conditions depends on
the efficiency of the compressor, while the isentropic head does
not.
With the flow from above, the aerodynamic or gas power of the
compressor, then, is determined to be:
(7)
The absorbed power (brake power) P is calculated by dividing
the internal power (gas power) by the mechanical efficiency
m
:
(8)
After considering the mechanical efficiency (
m
) (typically
around 98 to 99 percent), which accounts for bearing, seal, and
windage losses, the absorbed (or brake) power (P) of the com-
pressor becomes:
(9)
Related to measurements of head and flow is also the determi-
nation of the surge point or the surge line. The main challenge lies
in the fact that steady-state conditions are required for any of the
measurements discussed herein. By definition, surge is a nonsteady
condition. Even close to surge, most readings start to fluctuate. The
determination of flow at surge is, thus, much more inaccurate than
measurements further away from surge.
The method to use increased vibration levels as an indication of
surge, or incipient surge, is even more inaccurate because the
increased vibration levels might be generated by the onset of
rotating stall (which is by no means identical with the onset of
surge) or other conditions.
Equations of State
The aerothermodynamic performance of a gas compressor is
defined by enthalpy and entropy differences, so an additional
problem arises: enthalpies and entropies cannot be measured
directly, but have to be calculated by the use of an equation of state
(EOS). The state of any fluid consisting of known components can
be described by any given pair of its pressure, specific volume, and
temperature. Equations of state approximate these relationships.
The equations can also be used to calculate enthalpy and entropy
from the condition of a gas given by a pressure and a temperature
(Baehr, 1981).
The simplest equation of state is the equation for a perfect gas:
(10)
Real gases and in particular gas mixtures, however, display
complex relationships between pressure, volume, and temperature
(p-v-T). EOS use semiempirical equations to describe these rela-
tionships, in particular the deviations from perfect gas behavior:
(11)
They also allow for the calculation of properties that are derived
from the p-v-T relationships, such as enthalpy (h) and entropy (s).
SITE PERFORMANCE TEST EVALUATION FOR GAS TURBINE AND ELECTRIC MOTOR DRIVEN COMPRESSORS 55
Q
W
or Q
SQ
s
s
s
std
s
= =

( )
=

p
Z p T R T ,
( ) ( ) H h p s h p T
d s s
*
, , = = 0
( ) ( ) H h p T h p T
d d s s
= , ,

*
*
= =
H
H
H
H
p
p
P Q H
p
Z R T
Q H
g
= =
1 1
1
1 1
1
( ) ( )
[ ]
P P
W
h p T h p T
W H
G m
m
t t t t
m
= = = / , ,
*
*

2 2 1 1
P
P
g
m
=

pv p RT = = /
( )
p
Z p t R T

= ,
Because EOS are semiempirical, they might be optimized for
certain facets of the gas behavior, such as liquid-vapor equilibri-
ums and not necessarily for the typical range of temperatures and
pressures in various compression applications. Because different
EOS will yield different values for density, enthalpies, and
entropies, the EOS has to be agreed upon before the test.
Usually, it is not possible to select a most accurate EOS to
predict enthalpy differences, since there usually is no calibration
normal to test against. All the frequently used EOS (RK, BWR,
BWRS, LKP, RKS, PR) show reasonably correct enthalpies. It is
just not possible to decide which of them is more accurate for a
given application (Kumar, et al., 1999). Therefore, it is recom-
mended to use the EOS for test data reduction that was also used
for the performance prediction. This procedure is also recom-
mended in VDI 2045 (1993) to avoid additional test uncertainties.
Most equations of state used in gas compression applications are
either cubic equations (RK and its derivatives SRK and PR), and
Benedict-Webb-Rubin derivatives (BWR, BWRS, LKP). Beinecke
and Luedtke (1983) have conducted thorough evaluations on the
accuracy of the Lee-Kesler-Ploecker (LKP) method, the Benedict-
Webb-Rubin-Starling (BWRS) method, and the Soave-Redlich-
Kwong (SRK) method. All of the EOS mentioned can predict the
properties of hydrocarbon mixtures quite accurately over a wide
range of pressures. Still, deviations of 0.5 to 2.5 percent and
greater in the values for gas density are common. Even more
important than the compressibility factor is the calculation of the
enthalpy and entropy using the EOS. Because derivatives of the
EOS have to be used to perform these calculations (Poling, et al.,
2001), the deviations can be even larger than for the compressibil-
ity factor.
Figure 4 shows the effect of different EOS on the results for a
given set of typical test data. The isentropic efficiency was calcu-
lated based on four equations of state, using the Redlich-Kwong
equation as a reference. Depending on the pressure ratio, the four
different EOS deliver four different results for the same measured
conditions. For the calculations in the example, the following con-
ditions were used. Suction condition was always at T
1
= 20C
(68F) and p
1
= 50 bar (725 psia). The gas was compressed to
varying end pressures (p
2
) with T
2
chosen such that the reference
EOS (RK) yields 80 percent efficiency. The results are shown in
Figure 4. Differences as high as 2 percent exist among the EOS
models. Clearly, it cannot be concluded that a certain EOS will
always lead to higher efficiency than other EOS.
Figure 4. Isentropic Efficiency Differences among EOS for a
Natural Gas Mixture (when p
1
= 50 bar (725 psia), T
1
= 20

C
(68F) and varying p
2
, T
2
chosen to give = 80 percent for RK
EOS). (Courtesy Kumar, et al., 1999)
EOS may differ from program to program because sometimes
different mixing rules are used, different interaction parameters
between the gases are assumed, or a different treatment of the ideal
gas portion in the EOS is used. Poling, et al. (2001), give an
overview of the theory behind equation of state procedures.
Considerations for Trains with Multiple Compressors
In trains with multiple compressors, each compressor is treated
individually, both as far as pressure, temperature, flow measure-
ments, and gas compositions are concerned, but also with regards
to the design points. The latter requirement is due to the fact that
site conditions rarely allow both (or all three) compressors to
operate at their respective design points at the same time.
Therefore, their power consumption has to be determined individ-
ually, and later added up. If all compressors are completely
instrumented, the power requirement of the train (and thus the
power generated by the driver) can be determined.
Considerations for Compressors with Multiple Sections
The particular challenge for compressors with multiple sections
is to correctly separate the absorbed power for the individual
sections. The difficulty arises from the fact that there can be sig-
nificant mass transfer (due to leakage across the division wall) and
possibly heat transfer (again, across the division wall) from section
to section. It should be noted that the measurement of the overall
power consumption of the compressor is not affected by these
internal transfers. However, they can lead to observed efficiencies
that are too high for the first section, and too low for the second
section, or vice versa. For compressors with multiple sections, the
absorbed power is:
(12)
This relationship is valid, as long as all flows in and out of the
system are considered. Internal leakage does not affect it.
The main difficulty in the determination of the performance of
individual sections lies in the fact that the interstage leakage has an
impact on the observed section performance. The interstage
leakage can be determined by either:

Measuring the flow into the first section inlet, the first section
discharge, and the second section inlet

Measuring the flow into the first section inlet, measuring the
flow into the second section inlet, and estimating the leakage flow
based on theoretical considerations or factory test data.
Either method will yield the inlet flow used in the calculations
above.
For machines in back-to-back configuration (Figure 5), the
measured inlet temperatures and pressures will be used to calculate
actual inlet flow and the enthalpies of the gas flowing into the
respective sections. To calculate the enthalpy rise over the section,
the measured discharge pressures can be used. The measured
discharge temperatures will be corrected for the effect of the
division wall leakage by:
(13)
This procedure yields the suction flow, suction temperatures,
and discharge temperatures for the calculation of section head,
section efficiency, and section absorbed power as outlined in
Equations (2) through (9). The isentropic head of either section is
not affected by the leakage, while the discharge temperature
necessary for the calculation of the polytropic section head (and
the section efficiency) can be derived from the calculation above.
The method described above assumes that section I has a lower
discharge pressure than section II, and therefore the leakage is
from the section II to section I.
PROCEEDINGS OF THE THIRTY-FOURTH TURBOMACHINERY SYMPOSIUM 2005 56
P P
m
G Section i
i
n
=
=

1
1

, _
( )
[ ]
( )
H
W
H W W W H
T f H p gas
I
I
meas
I I
L L L
I I
2
1
2 1
2 2 2
1
= +
=
,
, ,
Figure 5. Configurations for High Pressure Ratio Compressor
TrainsTwo-Body Tandem, Straight Through (Compound)
Compressor, Back-to-Back Compressor. Usually, Intercoolers Are
Used to Reduce the Exit Temperature from the First Section to a
Lower Inlet Temperature into the Second Section.
For compound machines (Figure 5), the leakage over the
division wall can normally be neglected because the pressure dif-
ferential is usually only due to pressure loss in the intercooler and
the piping.
Using either procedure, the combined absorbed power of the
sections with the added mechanical losses yields the shaft power of
the compressor.
TEST CONDITIONS VERSUS REFERENCE
CONDITIONSCOMPARING TEST DATA
Usually, the operating points for the test are determined by the
facility and the test may not be conducted at some desired
condition. Also, when test data are taken over time for condition
monitoring purposes, the data are taken for different operating con-
ditions. Therefore, there is a need to compare data taken or
predicted at different conditions. Note that the gas compressor test
may serve several purposes, for example:

To determine the compressor performance

To load the gas turbine to full load to determine the gas turbine
output and full-load heat rate

To verify the performance of the entire train

For site tests it is always of advantage to compare the test data
with other, redundant measurements. For example, the gas
turbine driver full load performance may be known from a recent
factory test. If the compressor can be operated with the engine
running at full load, the compressor shaft power equals the engine
full load output. This engine performance can then be corrected
to factory test conditions (Kurz, et al., 1999), and should be rea-
sonably close to the factory test results. If the gas turbine fuel
flow can be measured, a similar comparison can be made for the
heat rate. If the results from the site test and the factory test are
reasonably close, the confidence in the site test results is
improved. Otherwise, reasons for the discrepancy should be
determined.
Similarity Conditions for Gas Compressors
The goal of a site test should be to create conditions that are as
close as possible to the original design conditions. The approach to
a site test is thus different from the approach to a factory test,
where a number of correction methods are used to correct per-
formance for vastly different conditions.
To compare test data for a centrifugal compressor, it is very
useful to use nondimensional parameters for head and flow.
Efficiency is already a nondimensional value.
Using the Flow Coefficient:
(14)
and Head Coefficient (isentropic or polytropic):
(15)
eliminates the requirement to test the compressor precisely at the
same speed as predicted. It allows one to compare data taken at
(somewhat) different speeds. (The same can be accomplished by
using the term Q/N and H/N
2
. However, these parameters are not
nondimensional.) This is because as long as and (as well as
some other parameters described below) are maintained, the aero-
dynamics of the compressor are not impacted by the different
operating condition. Figure 6 shows a typical, nondimensional
compressor map.
Figure 6. Nondimensional Compressor Map.
The other parameters that need to be maintained to accomplish
similarity (although with some possible deviations) are:
Machine Mach Number:
(16)
Machine Reynolds Number:
(17)
Isentropic Exponent:
(18)
Volume Flow Ratios:
(19)
Typically, only some of the similarity parameters can be brought
into accordance with the desired acceptance criteria, especially
when the gas composition during the test is different from the
design gas. The most important parameters are head and flow coef-
ficients and the machine Mach number. Maintaining the
volume/flow ratios is also desirable.
SITE PERFORMANCE TEST EVALUATION FOR GAS TURBINE AND ELECTRIC MOTOR DRIVEN COMPRESSORS 57

= =
Q
D u
Q
D N
s
tip
s
tip 4
1
2
4
1
3
2
,
,
( ) ( )

*
* *
, ,
= = = =
H H
D N
H H
D N
u
tip
p
p
u
p
tip
2 2
2 1
2
2 1
2
2 2
Ma
u
k Z RT
D N
k Z RT
u
s s s
tip
s s s
= =

Re
D N b
v
u
tip tip
s
=

k
v p
p v
=

( ) ( ) Q Q Q Q
s d
t
s d
a
/ / =
When keeping the flow coefficient the same as for the design
case, the velocity triangles at the inlet into the first stage remain the
same. Together with the head coefficient, this defines a singular
operating point of the compressor, as long as the fan law remains
applicable. If the volume flow ratios between inlet and outlet are
kept the same as for the design case, the velocity triangle at the
outlet of the compressor also will be the same. Generally, this
requirement involves keeping the same machine Mach number and
the same average isentropic exponents over the machine (at least
approximately).
For most applications, the Reynolds number similarity is of
lesser importance because the Reynolds numbers are relatively
high and clearly in the turbulent flow regime. Additionally, the loss
generation in centrifugal compressors is only partially due to skin
friction effects, i.e., due to effects that are primarily governed by
Reynolds numbers.
Certain deviations between design and test case for these param-
eters are acceptable and unavoidable. In general, as long as the
deviations between test and design stay within limits as described
in ASME PTC 10 (1997), or in VDI 2045 (1993), a simple correc-
tion based on the fan law can be used. Namely, the test point must
be at the same combination of and [Equations (14) and (15)]
as the design point. The limitations of the fan law are also
discussed by Brown (1991). Pipeline compressors, with usually
only one or two impellers per body, are typically less sensitive to
deviations from the above parameters (Figure 7). Multistage
machines show more sensitivity.
Figure 7. Effect of Machine Mach Number Variations on Gas
Compressor Performance. (Courtesy, Kurz and Fozi, 2002)
If the test conditions are considerably different from the design
conditions, for example outside of the limits established in ASME
PTC 10 (1997), or, in more general terms, when the fan law is no
longer applicable, easy corrections for Mach numbers and
volume/flow ratios are not available. Often, the design programs of
the compressor manufacturer can be used to recalculate the com-
pressor performance for the changed design conditions, which are
new curves for head coefficient versus flow coefficient, and effi-
ciency versus flow coefficient are generated for the new
conditions.
ASME PTC 10 (1997) assumes for a Type 1 test that the test gas
is almost identical to the gas for the specified acceptance condi-
tions. In a field test, the gas composition cannot be controlled by
the equipment manufacturer and the test gas might deviate from
the specified gas. In case the actual test gas deviates significantly,
the compressor performance can be recalculated for the actual test
gas.
Deviations also occur if the gas was specified incompletely, for
example, by only defining the specific gravity rather than a full gas
composition.
TEST UNCERTAINTIES
Test uncertainties need to be clearly distinguished from building
tolerances. Building tolerances cover the inevitable manufacturing
tolerances and the uncertainties of the performance predictions.
The actual machine that is installed on the test stand will differ in
its actual performance from the predicted performance by the
building tolerances. Building tolerances are entirely the responsi-
bility of the manufacturer.
Test uncertainties, on the other hand, are an expression of the
uncertainty of the measuring and testing process. For example, a
machine tested with 84 percent efficiency may have an actual effi-
ciency somewhere between 82 percent and 86 percent, assuming 2
percent test uncertainties.
The test uncertainty is basically a measurement of the quality of
the test. An increased test uncertainty increases the risk of failing
the test if the turbomachinery is actually performing better than the
acceptance level, but it reduces the risk of failing if the turboma-
chinery performance is lower than the acceptance level. Because it
is normal practice to use a lower performance than predicted as an
acceptance criterion, it is in the interest of the manufacturer as well
as the user to test as accurately as possible (Figure 8).
Figure 8. Test Uncertainty and Performance Tolerance.
Test uncertainties are caused by the following factors:

Instrument accuracy and calibration

Instrument location and installation

Number of instruments

Reading errors

Unstable process
When considering instrumentation tolerances, the whole
measuring chain needs to be taken into account. The instrument,
such as the resistance temperature detector (RTD), thermocouple,
or pressure transducer, has a certain accuracy and a certain quality
of calibration. However, the overall error is also influenced by the
location of the instrument (flow measurements with insufficient
straight runs), the way the instrument is installed (thermocouples
in thermowells without heat conductive paste or insufficient
immersion depth), potential reading errors (especially if gauges are
used), or the accuracy of the digital voltmeter, and the calibration
quality. The following are typical instrumentation tolerances:

Pressure 0.5 to 2.0 percent

Temperature 0.5 to 4F

Flow 0.5 to 2.0 percent

Gas composition 1.0 to 5.0 percent

Torque 1.0 to 1.5 percent

Equation of state 0.2 to 2.5 percent

Further, if the process shows fluctuations, it will influence the
accuracy of the test results.
The use of package instrumentation leads to a considerably lower
accuracy compared to tests with dedicated test instrumentation, espe-
cially due to higher calibration standards for the test instrumentation.
PROCEEDINGS OF THE THIRTY-FOURTH TURBOMACHINERY SYMPOSIUM 2005 58
Package instrumentation is normally selected to allow for suffi-
cient accuracy for trending. For trending purposes, the absolute
accuracy of a measurement is not important, but rather the differ-
ence from certain baselines. Package displays usually do not take
changes in gas composition into account. Furthermore, dedicated
test instrumentation is calibrated on a regular basis and maintained
continuously.
It is good practice to perform an uncertainty calculation to
determine what possible conclusions can be drawn from the test data.
For the uncertainty analysis, it is assumed that all measurement
parameters can be considered to be independent and that parame-
ters have associated statistical bounds, such as a 95 percent
confidence interval (u), rather than absolute limits of errors. All
parameters also are assumed to have Gaussian normal distributions
around their respective mean values, such that the uncertainties can
be properly combined using the root-square sum method. The total
uncertainty (F) for a given function, F = f (u
1
, u
2
, u
n
) is, thus,
determined from:
(20)
For this method, the overall uncertainty (F) has the same sta-
tistical meaning as the individual uncertainties (u). Namely, if u
represents a 95 percent confidence, then the result for the total
uncertainty (F) is also a 95 percent confidence interval.
While many test procedures use the rigorous application of
Equation (20) to determine the uncertainty of test results, the
method has its limitations: for complex relationships (e.g., when
equations of state have to be considered) the equation above is
rather difficult to use because the partial derivatives of all variables
are not easy to obtain.
An elegant way out is the following (Moffat, 1988). If a data
reduction program exists (e.g., a program that calculates compres-
sor shaft power from flow, pressure, and temperature
measurements), then the same program can be used to estimate the
uncertainty in the result. This is accomplished by sequentially per-
turbing the input values by their respective uncertainties and
recording their effects. Any term in Equation (20) can be approxi-
mated (assuming that the error is relatively small) by:
(21)
That means, that the contribution of the variable u1 to the uncer-
tainty in f can be found by calculating f twice: once with the
observed value of u1 and once for u
1
+ u
1
and then subtracting the
two values of f. When several variables are involved, the overall
uncertainty can be found by sequentially perturbing the individual
variables (u
1
) and then finding the square root sum of the squares
of the individual terms. This can be done on a spreadsheet.
To illustrate the aforementioned, the uncertainty for the com-
pressor head, flow, efficiency, and power is demonstrated in Table
1. The first column lists the measured parameters (SG, k50, k300,
pcp, and pct define the gas composition). All measured parameters
have their respective test uncertainties listed in column 3. The
respective readings from the instruments (averaged in the case of
multiple instruments for the same parameter) are listed in column
4. The following columns are each for one of the parameters in
column 1, and only that parameter is perturbed by the test uncer-
tainty for this parameter. For example, column 6 (for the discharge
pressure p2) changes only the value for p2 by its test uncertainty
(0.25 percent), and leaves all the other parameters as in column 4
[Equation (21)]. For each column, the values for head, flow effi-
ciency, and power are calculated and compared to the nominal
values from the data in column 4. The absolute test uncertainty is
then the square root of the sum of the differences squared
[Equation (20)].
Table 1. Spreadsheet for Test Uncertainty Calculation.
For example, the absolute uncertainty for power is 220 hp,
which amounts to 2.87 percent of the nominal value. In the
example, it also becomes immediately obvious that the largest con-
tributor to the uncertainty of the absorbed power is the inaccurate
measurement of the pipe inner diameter, while the (in this case
very accurate) measurements of temperature and pressure at
suction and discharge are the key contributors to the accuracy of
the efficiency measurement.
The beauty of this scheme lies in the fact that:

It does not matter whether the uncertainty is given as an absolute

or relative number.

The procedure can be implemented using any of the commercial

spreadsheet programs.

Any value in the table can be the result of a complex, even

iterative, calculation.
More details on test uncertainty calculation can be found in
Moffat (1988), ASME PTC 19.1 (1985), Kurz, et al. (1999), and
Kurz (2001).
INTERPRETATION OF TEST DATA
If the test data deviate from the predictions or from other test
data by more than the level of test uncertainty, the reasons must be
explored. Assuming the test data are reduced correctly, it must be
determined whether the test conditions were close enough to the
conditions for the prediction.
It is usually desirable to use redundant data for comparison.
Examples follow.
Determine the shape of the head-flow and flow efficiency
curves, and compare them with predictions. If the curves are just
shifted to the left or right, the flow measurement is suspect.
Another necessary step is to compare the whole measured --
curve with the predicted curve. For compressors, it might be found
that the head-versus-flow curves have just shifted horizontally,
which points to an incorrect flow measurement. If some points of
the curve match the predictions and others do not match, variations
of the gas composition during the test could be the cause. Data
from a site test (Figure 9) for a compressor station close to several
wells serves as an example. The solid line represents the prediction
for head and efficiency. The symbols indicate test data taken
during two tests. Test 1 experienced significant fluctuations in gas
composition, while test 2 was somewhat more stable.
Additional evidence may come from a comparison between
compressor absorbed power and expected driver available power:
SITE PERFORMANCE TEST EVALUATION FOR GAS TURBINE AND ELECTRIC MOTOR DRIVEN COMPRESSORS 59
F u
f
u
u
f
u
u
f
u
n
n
=

1
1
2
2
2
2 2

( ) ( ) u
f
u
f u u f u
1
1
1 1 1

+
Figure 9. Data Scatter Due to Unstable Gas CompositionTest 1
Experienced Fluctuations in the Gas Composition. Test 2, on the
Same Compressor, Was More Stable.
determine the absorbed power and compare it with the expected
power from the driver. For a gas turbine, full load factory test data
are usually available. The compressor should be operated at a point
that requires the gas turbine to operate at full load. The absorbed
compressor power should be close to the as-factory-tested gas
turbine power (corrected to the site test conditions regarding
ambient conditions and power turbine speed), assuming the gas
turbine is in new and clean condition. For electric motor driven
compressors, the motor, gearbox, and variable frequency drive
(VFD) efficiencies can be used to compare the measured electric
power consumption to the absorbed compressor power.
Head and efficiency lowa comparison to available other test
data should be made. If the head was already low in a factory test,
then the results from the site test may just confirm the factory test
findings. A wrong flow measurement can make the compressor
look like it is not producing the correct head and efficiency (refer
above). Other reasons include damaged impellers or damaged
seals. Both issues can be eliminated by visual inspection, if
possible. Damaged balance piston seals can also be detected by
monitoring the pressure (or flow) in the balance piston return line.
Ingested inlet strainers that are caught in the inlet (or other obstruc-
tions) cause significant pressure drop between the measurement
location and the actual compressor inlet, thus generating the
impression that compressor is low in head and efficiency.
Any data taken must be corrected to the same datum conditions.
For gas compressors, the nondimensional curves are a good tool.
However, large deviations in Mach number (Figure 7), especially
in multistage machines, need to be avoided.
On gas compressors, the effects of different Mach numbers or
different volume/flow ratios (Q
s
/Q
d
) may be responsible for the
deviations. In such cases, it is always helpful to repeat the predic-
tion procedure for the actual test conditions.
In many instances, redundant measurement can increase the
confidence in the results. The compressor gas power can be
checked by comparing the results with the gas turbine power and
heat rate from the factory test, corrected to site test conditions
(Kurz, 1999). In this case, it is also recommended to thoroughly
clean the gas turbine air compressor prior to the test; 3 percent and
more engine power have been recovered after cleaning the air com-
pressor. Electric motors allow a convenient measurement of the
electric power input. Corrected by the motor efficiency, the
gearbox efficiency, the losses in the variable frequency drive (if
applicable), the motor shaft power can be calculated and compared
to the measured compressor power.
It is good practice to perform a test uncertainty calculation as
part of the data evaluation. Obviously, data with an uncertainty of
3 percent cannot yield conclusions that require an accuracy of 1
percent. If the test point does not match the prediction or other test
results, a test uncertainty ellipse (Figure 10) can be drawn. The two
axes of the ellipse represent the test uncertainties for the parame-
ters on the x and y axis, respectively. If it still covers the prediction,
the test results might be correct. The uncertainty ellipse in Figure
10 expresses the fact that not only is the measured head subject to
test uncertainties, but also the measured flow. When comparing
field test results with factory tests, the influence of test uncertain-
ties in both tests must be considered. Whatever factory test results
are available, they can be used for comparison and verification
purposes. Whatever the deviation might be, it is best if it can be
detected, discussed, and possibly corrected during the test. Agood
relationship with a trusted manufacturer can help in finding causes
for discrepancies.
Figure 10. Comparison of Multiple Test Points, Their Respective
Test Uncertainties, with the Acceptance Point.
Another reason for test data discrepancies can be found in the
way the test is conducted. If the test data have not been taken while
the equipment operated under steady-state conditions, they may
not be useful. Before readings are taken for any individual test
point, steady-state operating conditions must be achieved. Steady-
state is achieved if all of the following apply during the 10-minute
interval.

Operating speed constant within 5 rpm

Fluctuations of the efficiency reading no larger than +/- 0.5

points from average, while head and actual flow remain within +/-
0.5 percent from average, respectively.
The driving gas turbine (where applicable) must be heat soaked
if the compressor test point requires full load, to avoid drift. The
time requirement to achieve heat soaking should be provided by
the gas turbine manufacturer. As a rule of thumb, one hour is
required for smaller engines (below, say 8000 hp), while larger
engines may require two or more hours. Awell conducted test will
yield repeatable and reliable results. In Figure 11, the results of
several tests at two different stations, including data for four
identical compressors, tested consecutively, can serve as proof for
this statement.
CONCLUSIONS
Based on the information provided in this paper, the test
engineer can convert raw test data into meaningful data that can be
used for evaluating the performance of a gas turbine or a gas com-
pressor. The data reduction was explained based on the basic
relationships of pressure, temperature, flow, and head, as well as
the operational characteristics of gas turbines and centrifugal gas
compressors. The appropriate use of performance maps was
explained.
NOMENCLATURE
c
p
= Specific heat at constant pressure
C = Discharge coefficient
E = Velocity approach factor
PROCEEDINGS OF THE THIRTY-FOURTH TURBOMACHINERY SYMPOSIUM 2005 60
Figure 11. Site Test Results for Four Aerodynamically Identical
Compressors at Two Different Sites.
EOS = Equation of state
BWRS = Benedict-Webb-Rubin-Starling
LKP = Lee-Kesler-Ploecker
PR = Peng-Robinson
RK = Redlich-Kwong
SRK = Soave-Redlich-Kwong
f = Function
h = Enthalpy
H = Head
HR = Heat rate
k = Isentropic exponent
LHV = Lower heating value
Ma = Mach number
MW = Molecular weight
N = Speed (s
1
)
p = Pressure
P = Power
Q = Volumetric flow
R = Gas constant
Re = Reynolds number
SQ = Standard flow
T = Temperature
u = Function variable
u = Velocity of the impeller tip
v = Specific volume
W = Mass flow
Z = Compressibility factor
= Density
= Torque
= Efficiency
= Flow coefficient
= Head coefficient
Subscripts
a = Acceptance
amb = Ambient
d = Compressor discharge nozzle
GB = Gearbox
I = Inlet
L = Leakage
M = Mechanical
packg = Package
s = Compressor suction nozzle
SL = Sea level
t = Test
tip = Impeller tip
Superscripts
* = Isentropic
p = Polytropic
I = Section 1
REFERENCES
ASME PTC 10, 1997, Compressors and Exhausters, American
Society of Mechanical Engineers, New York, New York.
ASME PTC 19.1, 1985, Measurement Uncertainties, American
Society of Mechanical Engineers, New York, New York.
ASME PTC 19.5, 1971, Fluid Meters, American Society of
Mechanical Engineers, New York, New York.
Baehr, H. D., 1981, Thermodynamik, Berlin, Germany: Springer.
Beinecke, D. and Luedtke, K., 1983, Die Auslegung von
Turboverdichtern unter Beruecksichtigung des realen
Gasverhaltens, VDI Berichte, (487), pp. 271-279.
Brown, R.N., 1991, Fan Laws, the Use and Limits in Predicting
Centrifugal Compressor Off Design Performance,
Proceedings of the Twentieth Turbomachinery Symposium,
Turbomachinery Laboratory, Texas A&M University, College
Station, Texas, pp. 91-100.
ISO 5167, 1980, Measurement of Fluid Flow by Means of Orifice
Plates, Nozzles and Venturi Tubes Inserted in Circular Cross-
Section Conduits Running Full, International Organization
for Standardization, Geneva, Switzerland.
Kumar, S. K., Kurz, R., OConnell, J. P., 1999, Equations of State
for Compressor Design and Testing, ASME Paper 99-GT-12.
Kurz, R., Brun, K., Legrand, D.D., 1999, Field Performance
Testing of Gas Turbine Driven Compressor Sets, Proceedings
of the Twenty-Eighth Turbomachinery Symposium, Turbo-
machinery Laboratory, Texas A&M University, College
Station, Texas, pp. 213-230.
Kurz, R., 2001, Site Performance Testing of Gas Turbine Driven
Centrifugal Compressors, GMRC Gas Machinery Confer-
ence, Austin, Texas.
Moffat, R. J., 1988, Describing the Uncertainties in Experimental
Results, Experimental Thermal and Fluid Science, 1:3-17.
Poling, B. E., Prausnitz, J. M. and OConnell, J. P., 2001, The
Properties of Gases and Liquids, Fifth Edition, New York,
New York: McGraw Hill.
VDI 2045, 1993, Acceptance and Performance Tests on Turbo
Compressors and Displacement Compressors, Verein
Deutscher Ingenieure e.V., Dsseldorf, Germany.
BIBLIOGRAPHY
AGA Report No 3, 1991, Orifice Metering of Natural Gas,
American Gas Association, Washington, D.C.
SITE PERFORMANCE TEST EVALUATION FOR GAS TURBINE AND ELECTRIC MOTOR DRIVEN COMPRESSORS 61
ASME PTC 22, 1997, Gas Turbine Power Plants, American
Society of Mechanical Engineers, New York, New York.
Kurz, R. and, Fozi, A. A., 2002, Acceptance Criteria for Gas
Compression Systems, ASME Paper GT-2002-30282.
Robinson, R. L. and Jacobi, R. H., 1965, Better Compressibility
Factors, Hydrocarbon Processing, 4.
PROCEEDINGS OF THE THIRTY-FOURTH TURBOMACHINERY SYMPOSIUM 2005 62