i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 4 1 ( 2 0 1 4 ) 1 9 0 e1 9 9

Available online at www.sciencedirect.com

ScienceDirect
w w w . i i fi i r . o r g

journal homepage: www.elsevier.com/locate/ijrefrig

Empirical models for efficiency and mass flow rate

of centrifugal compressors

Xiande Fang*, Weiwei Chen, Zhanru Zhou, Yu Xu

Institute of Air Conditioning and Refrigeration, Nanjing University of Aeronautics and Astronautics, 29 Yudao Street,
Nanjing 210016, PR China

article info abstract

Article history: Centrifugal compressors are widely used in refrigeration systems, heat, power and
Received 18 December 2013 chemical plants, turbochargers, and vehicle engines, including flight vehicles. Empirical
Received in revised form models of efficiency and mass flow rates of centrifugal compressors are required for the
3 February 2014 design, energy analysis, performance simulation, and control- and/or diagnose-oriented
Accepted 3 March 2014 dynamic simulation of such compressors and systems. This work presents a compre-
Available online 12 March 2014 hensive review of empirical models for efficiency and mass flow rates of centrifugal
compressors. An experimental investigation of a centrifugal compressor for air-cycle
Keywords: refrigeration systems of aircraft is carried out. Then, the available models are evaluated
Centrifugal compressor with the experimental data of the refrigeration centrifugal compressor and two turbo-
Empirical model charger centrifugal compressors. It is found that the most models for centrifugal com-
Efficiency pressors of vehicle engines and turbochargers are not satisfactory for refrigeration
Mass flow rate centrifugal compressors, and thus accurate models for refrigeration centrifugal compres-
sors need to be developed.
ª 2014 Elsevier Ltd and IIR. All rights reserved.

Modèle empirique pour l’efficacité et le débit massique de

compresseurs centrifuges
Mots clés : Compresseurs ; Modèle empirique ; Efficacité ; Débit massique

(Letlow and Jenkins, 1998; Ashford and Brown, 2000). Cen-

1. Introduction
Centrifugal compressors are widely used in air-cycle refrig-
eration systems such as aircraft environmental control sys-
tems (Ordonez and Bejan, 2003; Conceição et al., 2007; Fang
et al., 2010). It is also used for vapor-cycle refrigeration
trifugal compressors have been standard in turbochargers and
some vehicle engines, including jet engines, where they are
driven by gas turbines (Sieros et al., 1997; Eriksson et al., 2002;
Canova, 2004), and in industries such as oil refineries, chem-
ical and petrochemical plants, heat and power plants, and

* Corresponding author. Tel./fax: þ86 25 8489 6381.

E-mail addresses: xd_fang@yahoo.com, xd_fang@nuaa.edu.cn (X. Fang).
http://dx.doi.org/10.1016/j.ijrefrig.2014.03.005
0140-7007/ª 2014 Elsevier Ltd and IIR. All rights reserved.
 i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 4 1 ( 2 0 1 4 ) 1 9 0 e1 9 9 191

Nomenclature Greek symbols

g polytropic exponent
ai constant (i ¼ 0, 1, 2, .)
hc isentropic efficiency
cp constant pressure specific heat, J kg1 K1
pc compression ratio
dc blade wheel diameter, m
r density, kg m3
Gc mass flow rate, kg s1
fc normalized compressor mass flow rate
k isentropic exponent
fcorr corrected mass flow parameter, kg s1
M Mach number
j dimensionless head parameter
n rotational speed, rev s1 (RPS)
nnond non-dimensional speed, n/nref Subscripts
pffiffiffiffiffiffiffi
ntc rotational speed parameter, n= Tin , rev s1 K1/2 corr corrected parameter
p pressure, Pa exp experimental
R gas constant, J kg1 K1 in inlet
R2 coefficient of determination out outlet
R2c corrected coefficient of determination pred predicted
s2 residual mean square ref value at a reference point
T temperature, K top maximum value on a speedline
Uc blade tip speed, m s1 s isentropic
Vd displacement volume, m3 rev1
Wc compression work, W

natural gas processing plants where they are driven by electric
motors or by steam or gas turbines.
The models of compressor efficiency and mass flow rate
are necessary for performance study, energy analysis, and
control- and/or diagnose-oriented dynamic simulation
(Eriksson et al., 2002; Canova et al., 2009; Chu et al., 2012). The
compressor efficiency is also called the compressor isentropic
efficiency or the compressor adiabatic efficiency.
Mathematical models of compressors can be built either
theoretically or experimentally. Theoretical modeling is based
on physical laws of mass, energy and momentum conservation
(Schiffmann and Favrat, 2010; Liu et al., 2012; Chamoun et al.,
2013). The high complexity of this modeling approach and its
computational requirement make it unsuitable for control-
and/or diagnose-oriented dynamic simulation and less
attractive than compact and time-efficient models in design,
analysis and performance simulation. Experimental modeling
can yield compact and time-efficient mathematical models,
which not only meet the need of control- and/or diagnose-
oriented dynamic simulation, but also possibly apply to
design, analysis and performance simulation. An experiment-
based model is often referred to as an empirical model or a
semi-empirical model, and is also called a mean value model in
some literature regarding turbochargers, vehicle engines, and
jet engines (Jensen et al., 1991; Biteus, 2004; Canova et al., 2009).
There are a number of mean value models for compressor
efficiency and mass flow rates of turbochargers, vehicle en-
gines, and jet engines (Jensen et al., 1991; Kolmanovsky et al.,
1997; Mueller, 1997; Sieros et al., 1997; Guzzella and Amstutz,
1998; Eriksson et al., 2002; Biteus, 2004; Andersson, 2005;
Canova et al., 2009). However, a systematic review has not
been found. Some papers presented brief review, but the
number of the models covered was very limited. Moraal and
Kolmanovsky (1999) compared the models of Jensen et al.
(1991), Mueller (1997), and Kolmanovsky et al. (1997) to the
data of three turbocharger compressors. Andersson (2005)
compared the models of Jensen et al. (1991) and his own
with the measured data.
Most empirical efficiency and mass flow models of cen-
trifugal compressors are derived from and applied to com-
pressors of vehicle engines, turbocharger, and jet engines. For
refrigeration compressors, there are no such models found.
Due to lack of available models, most control and perfor-
mance modeling of refrigeration centrifugal compressors
opted for theoretical modeling (Schiffmann and Favrat, 2010;
Liu et al., 2012), look-up tables (Tsujikawa, 1985; He et al.,
2004), or neural network methods (Fast et al., 2009; Wang
et al., 2011; Li and Li, 2012).
Our literature survey has not found any paper providing an
overall review on the empirical models of efficiency and mass
flow rates of centrifugal compressors and any empirical model
for refrigeration centrifugal compressors. This paper presents
a comprehensive review on empirical models of the
compressor efficiency and mass flow rates of centrifugal
compressors. The reviewed models are evaluated with the
experimental data of two turbocharger centrifugal compres-
sors and a centrifugal compressor for air-cycle refrigeration
systems. The experimental investigation of a centrifugal
compressor for air-cycle refrigeration systems is conducted to
provide data for assessing the potential of centrifugal
compressor models of turbochargers and vehicle engines to
refrigeration centrifugal compressors.
2. Review of empirical models of compressor
efficiency and mass flow rates
2.1. Compressor performance map
Empirical models of compressor efficiency and mass flow rate
are essentially a mathematical interpretation of the
compressor performance map. The operating behavior of a
centrifugal compressor can be defined by a map with pressure
ratio and non-dimensional mass or volume flow rate as co-
ordinates (Fig. 1). The useable portion of the map is limited by
the surge, choke and maximum permissible speed lines.
192 i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 4 1 ( 2 0 1 4 ) 1 9 0 e1 9 9

The left hand boundary of the map, where the air flow is too where the compressor compression ratio pc and the
low for the current pressure ratio, is the surge line. This is compressor blade tip speed Uc are defined as
basically “stalling” of the air flow at the compressor inlet. With
pout
too small a volume flow and too high a pressure ratio, the flow pc ¼ (3)
pin
can no longer adhere to the suction side of the blades, resulting
in the interruption of the discharge process. Operation to the
Uc ¼ pdc n (4)
left of the surge line represents a region of flow instability. This
region is characterized by mild flutter to wildly fluctuating The compressor corrected mass flow parameter fcorr is
boost and resultant noise known as “surging” from the defined as

sffiffiffiffiffiffiffiffi
compressor. Continued operation within this region can lead

to premature turbo failure due to heavy thrust loading. pref Tin
fcorr ¼ Gc (5)
The right hand boundary of the map is the choke line. The pin Tref
maximum centrifugal compressor volume flow rate is nor-
where the reference point is usually at 15  C and 101.3 kPa.
mally limited by the cross-section at the compressor inlet.
The compressor rotational speed parameter ntc is defined as
When the flow at the wheel inlet reaches sonic velocity, no
further flow rate increase is possible. The choke line can be n
ntc ¼ pffiffiffiffiffiffiffi (6)
recognized by the steeply descending speed lines at the right Tin
on the compressor map. In addition to the rapid drop of The compressor (isentropic, adiabatic) efficiency hc is
compressor efficiency past this line, the compressor speed defined as
will also be approaching or exceeding the allowable limit.
The top boundary of the map is the maximum permissible Wc;s
hc ¼ (7)
speed line, where the compressor is approaching its Wc
maximum shaft speed. Past this point, the compressor is where Wc,s and Wc are the compressor isentropic compression
liable to “overspin” and cause catastrophic bearing failure. work and actual compression work, respectively.
Empirical models of compressor efficiency and mass flow
 k1 
rate are generally proposed to provide a mean for determining Wc;s ¼ Gc cp Tin pck  1 (8)
the efficiency and mass flow rates for normal operational
conditions, i.e. the region within the surge and choke lines
Wc ¼ Gc cp ðTout  Tin Þ (9)
and the maximum permissible speed line.
Substituting Eqs. (8) and (9) into Eq. (7), it follows that
2.2. Basic parameters ,
 k1  T 
out
hc ¼ pck  1 1 (10)
For a centrifugal compressor, the normalized compressor Tin
mass flow rate fc and the dimensionless head parameter j are
defined as 2.3. Compressor mass flow rate models

Gc (1) Jensen et al. (1991) model

fc ¼ p 2 (1)
d Ur
4 c c in

 k1  Studying the compressor of a small turbocharged diesel

cp Tin pck  1 engine, Jensen et al. (1991) proposed a compressor mass flow
j¼ 1 2
(2) model based on the normalized compressor mass flow rate fc,
U
2 c
the dimensionless head parameter j, and the Mach number M
as the following:

A1 þ A2 4
j¼ (11a)
A3  4c

Ai ¼ ai1 þ ai2 M; i ¼ 1; 2; 3 (11b)

where coefficients aij are determined through the least square

fit on experimental data, and the Mach number is defined as

Uc
M ¼ pffiffiffiffiffiffiffiffiffiffiffiffi (12)
kRTin

where R is the gas constant.

(2) Mueller (1997) model

Mueller (1997) used the following polynomial model to

simulate the compressor of turbocharged spark ignition (SI)
Fig. 1 e Compressor performance map. engines:
 i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 4 1 ( 2 0 1 4 ) 1 9 0 e1 9 9 193

j ¼ A1 þ A2 4c þ A3 42c (13a) (7) Biteus (2006) model

A1 ¼ a1 þ a2 Uc þ a3 U2c (13b) Biteus (2006) presented the following mass flow rate model
for turbocharged SI engine compressors:
2A3 Gc;top    1=2  1=4
A2 ¼ (13c) 1 1 1
Uc Gc ¼ a1 1  þ a2 n 1  þ a3 n 1  þ a4 n (18)
pc pc pc

Gc;top ¼ a4 þ a5 Uc þ a6 U2c (13d) (8) Sieros et al. (1997) models

A3 ¼ a7 þ a8 Uc þ a9 U2c (13e) Sieros et al. (1997) proposed three types of mass flow
models for centrifugal compressors of jet engines: Nonlinear
(3) Kolmanovsky et al. (1997) model model, simple linear model and generalized linear models. All
of them use the non-dimensional speed, nnond, defined as
Kolmanovsky et al. (1997) proposed the following model for
nnond ¼ n nref (19)
compressors used in automobile turbocharged diesel engines:
8  (a) Nonlinear model, which can be reduced to
n h  p
>
> þ b 
c
 for pc < pc;top
>
< 1 1 exp a5 1
a2 þ a3 nnond þ a4 n2nond
fc pc;top
¼  pc ¼ a0 þ a1 nnond þ (20)
fc;top >  p a5 þ Gc þ a6 nnond
>
> c
: 1  ba5 1 for pc > pc;top
pc;top This model resulted in a mean error of 1% for all com-
(14a) pressors examined by the authors.
fc;top ¼ a1 ntc þ a2 n2tc (14b) (b) Simple linear model, which can be reduced to

Gc þ pc ¼ a1 þ a2 nnond ðGc  pc Þ2 (21)

pc;top ¼ a3 þ a4 f2c;top (14c)
This form is implicit, which is difficult to be used. For all of
b ¼ a6 expð  a7 ntc Þ (14d) the compressors examined by the authors, the simple linear
model had the mean error between 2.5 and 3.2%.
where fc,top is the maximum normalized compressor mass (c) Generalized linear model 1.
flow rate on each speedline, and pc,top is the compression ratio The simplest form of the generalized linear models is
corresponding to fc,top.
Y ¼ A1 þ A2 X (22a)
(4) GuzzellaeAmstutz (1998) model Y ¼ Gc þ pc (22b)
2
Studying the control of diesel engines, Guzzella and X ¼ nnond ðGc  pc Þ (22c)
Amstutz (1998) proposed the following model for the A1 ¼ a1 þ a2 nnond (22d)
compressor:
A2 ¼ a3 þ a4 nnond (22e)
2
pc ¼ A1 þ A2 ðfcorr  A3 Þ (15) For all of the compressors examined by the authors, this
where Ai (i ¼ 1,2,3) depend on the shaft speed n only, and can form had the mean error between 1.0 and 2.0%.
be parameterized as polynomial functions. (d) Generalized linear model 2.
A further increase in accuracy is achieved by using a
(5) Andersson (2005) model second-degree polynomial of the form

pc ¼ a1 þ A1 X þ A2 X2 (23a)
Andersson (2005) proposed the following model for the
compressor of turbocharged SI engines: X ¼ nnond ðpc þ Gc Þ2 (23b)
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi A1 ¼ a2 þ a3 nnond þ a4 n2nond (23c)
4c ¼ a1 1  a2 j2 ; (16)
A2 ¼ a5 þ a6 nnond (23d)
where 1  a2j2  1.
for which the mean error was <1% for most of the test
compressors.
(6) Eriksson et al. (2002) model

Eriksson et al. (2002) proposed the following mass flow rate

model for turbocharged SI engines compressors:
2.4. Compressor efficiency models
Gc ¼ a1 Pr þ a2 Pr þ a3 Pr
2 3
(17a)
(1) Jensen et al. (1991) model
0 02
Pr ¼ pc þ a4 n þ a5 n (17b)
Jensen et al. (1991) proposed the following efficiency model
0 5 4
where n ¼ 5  10 (n  8  10 ). for the compressor of a small turbocharged diesel engine:
194 i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 4 1 ( 2 0 1 4 ) 1 9 0 e1 9 9

hc ¼ a1 42c þ a2 4c þ a3 (24a) (5) Eriksson et al. (2002) model

ai1 þ ai2 M Eriksson et al. (2002) proposed the following compressor

ai ¼ ; i ¼ 1; 2; 3 (24b) efficiency model for turbocharged SI engines:
ai3  M
"     #
(2) Guzzella and Amstutz (1998) model 2
Gc Gc
Wc;s ¼ U2c a1 þ a2 þ a3 (30a)
Uc Uc
Guzzella and Amstutz (1998) proposed the following model
for the compressor used in diesel engines:
Wc ¼ b1 Gc þ b2 Gc Uc þ b3 U2c þ b4 Uc þ b5 (30b)
hc ¼ hc;max  c Qc T
(25a) Substituting Eq. (30) into Eq. (7), the compressor efficiency hc
can be obtained.


cT ¼ 4corr  4corr;hmax ; pc  pc;hmax (25b)
(6) Sieros et al. (1997) model
 
a11 a12
Q¼ (25c)
a12 a22 Sieros et al. (1997) proposed two types of linear (simple
linear and generalized linear) models for the efficiency of
where fcorr,hmax and pc, hmax are the corrected mass flow
centrifugal compressors of jet engines.
parameter and the compression ratio corresponding to the
(a) Simple linear model
maximum efficiency hmax, respectively. This model requires
hmax and its location in terms of fcorr,hmax and pc, hmax. This Y ¼ a1 þ a2 X (31a)
point may not be included in the measured data, in which case
hmax, fcorr,hmax and pc, hmax need to be determined using curve Y ¼ ðpc þ hc Þ2 (31b)
fitting or similar methods.
(3) Andersson (2005) model
X ¼ ðpc þ hc þ 1=pc Þðpc þ hc Þ (31c)
When estimating the compressor performances of turbo-
charged SI engines, Andersson (2005) modified the (b) Generalized linear model 1
GuzzellaeAmstutz (1998) model by substituting pc to
hc ¼ A1 þ A2 pc þ A3 p2c (32a)
1 þ (pc1)0.5 as the following:

hc ¼ hc;max  cT Qc (26a) A1 ¼ a1 þ a2 nnond þ a3 n2nond (32b)

h  pffiffiffiffiffiffiffiffiffiffiffiffiffiffi i
cT ¼ 4corr  4corr;hmax ; 1 þ pc  1  pc;hmax (26b) A2 ¼ a4 þ a5 nnond (32c)

  A3 ¼ a 6 (32d)
a11 a12
Q¼ (26c)
a12 a22 (c) Generalized linear model 2

(4) Canova et al. (2009) model hc ¼ A1 þ A2 pc þ A3 logpc (33a)

Defining the corrected compression work Wcorr as A1 ¼ a 1 (33b)

 sffiffiffiffiffiffiffiffi
pref Tref A2 ¼ a2 þ a3 nnond (33c)
Wcorr ¼ Wc (27a)
pin Tin
A3 ¼ a4 þ a5 nnond (33d)
Canova et al. (2009) represented Wcorr of automotive diesel
engines as a function of the corrected mass flow rate as the This model has the same accuracy as the generalized linear
following: model 1, while reducing the regression constants from six to
five.
Wcorr ¼ A1 þ A2 4corr þ A3 42corr þ A4 43corr (27b)

where the coefficients Ai (i ¼ 1, 2, 3, 4) are linear functions of

the corrected speed, ncorr, and can be obtained by fitting the 3. Experiment of an air-cycle refrigeration
compressor maps. centrifugal compressor
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

ncorr ¼ n Tref Tin (28) The above reviewed models are used for centrifugal com-
pressors of turbochargers and vehicle engines. In order to
When Eq. (27b) is fitted using experimental data, the
evaluate their applicability to refrigeration centrifugal
compressor efficiency model can be determined by
compressors, a test rig for an aircraft environmental control
substituting Eqs. (27) and (8) into Eq. (7),
system (ECS) with a bootstrap air cycle machine (ACM) was
 k1  qffiffiffiffiffiffi built, with which experiments were conducted. In an air-
Gc cp Tin pck  1 ppref Tref

hc ¼
T in in
(29) cycle ECS, the bleed air (compressed hot air from the en-
Wcorr gine compressor) expands through the cooling turbine of
 i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 4 1 ( 2 0 1 4 ) 1 9 0 e1 9 9 195

the ACM, where the thermal energy of the air is converted

Table 1 e Sensors and uncertainties of the test rig.
into the turbine shaft work. The turbine shaft work can be
used to drive a coaxial compressor and/or fan to enhance Parameter Sensor Range Uncertainty
the refrigeration efficiency (Fang and Xu, 2011). An ACM is Temperature Platinum 50eþ250  0.5  C
called the bootstrap ACM if its compressor is coaxial with resistance
thermometer
the cooling turbine. A bootstrap ACM may be a type of 2-
( C)
wheel, 3-wheel or 4-wheel (Ordonez and Bejan, 2003;
Pressure Pressure 0e0.2; 0e1.0 0.5%
Conceição et al., 2007), depending on how the turbine transducer
shaft work is used. Aircraft ACM compressors are generally (MPa)
centrifugal. The tested ECS is a 2-wheel bootstrap one, Mass flow rate Gas turbine 400e4000 1.0%
which is also the ACM commonly seen in ground applica- flowmeter
tions of air-cycle refrigeration systems. (kg h1)

The test rig is shown in Fig. 2. The pressurized air

compressor is used to provide pressurized air. The air storage
using Eq. (9), and the compression efficiency hc is calculated
tank is used as a buffer so that the inlet pressure of the mass
using Eqs. (7)e(9).
flow meter can be stable. The electric heater heats the pres-
surized air so that the outlet air of the electric heater be-
comes the pressurized hot air having the pressure and
temperature similar to bleed air. The simulated bleed air
4. Evaluation of the reviewed models for
enters the ACM compressor, where its pressure is boosted
application to refrigeration centrifugal
further. The heat exchanger between the ACM compressor
compressors
and ACM turbine is used to control the inlet temperature of
4.1. Criteria for evaluation
the turbine. The cabin simulation chamber is used to simu-
late the aircraft cabin pressure that is usually lower than the
Commonly used criteria for comparing a set of models with
ground pressure in flight, and the vacuum pump is used to
their constants determined with the same database are the
control the pressure in the cabin simulation chamber. The
residual mean square s2, the coefficient of determination R2,
heat exchanger is a cross-flow aireair heat exchanger typi-
and the corrected coefficient of determination R2c (Fang and
cally used in aircraft ECSs. The ACM turbine is a centripetal
Dai, 2010; Myers, 2005; Fang and Xu, 2013).
cantilever type turbine commonly seen in aircraft air-cycle
ECSs. N h i2
1 X
The positions of the sensors used in the test rig are given in s2 ¼ yexp;i  ypred;i (34)
N  m i¼1
Fig. 2. The mass flow meter is an orifice meter, the pressure
sensors are Pitot tubes, the temperature sensors installed in
Pn  2
pipes are platinum resistance temperature sensors, and the i¼1 yexp;i  ypred;i
temperature sensors in the cabin simulation chamber are R ¼1P
2
 2 (35)
n
i¼1 yexp;i  ymean
copper-constantan thermocouples. Their uncertainties are
given in Table 1.
There are 54 data points obtained in the experiments, as ðN  1Þð1  R2 Þ ðN  1Þs2
R2c ¼ 1  ¼1P  2 (36)
listed in Table 2. The compressor inlet pressures range from Nm n
i¼1 yexp;i  ymean
151 to 558.5 kPa, the out pressures from 276.8 to 927.3 kPa, and
the pressure ratios from 1.16 to 2.05. The compressor inlet where yexp and ypred are the experimental and predicted
temperatures range from 282.1 to 284.2 K, and the outlet values, respectively, ymean is the mean experimental value, N
temperatures from 325.8 to 391.4 K. The mass flow rates range is the number of observations, and m is the number of pre-
from 0.256 to 0.728 kg s1, the rotational speeds range from 827 dictors (m ¼ 5 in Eq. (33) for instance).
to 1248 RPS, and the shaft powers range from 18.8 to 45.5 kW. The coefficient of determination R2 is a statistic index that
The compression work Wc is obtained through calculation gives some information about the goodness of fit of a model.

Fig. 2 e Test rig of 2-wheel bootstrap ECS.

196 i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 4 1 ( 2 0 1 4 ) 1 9 0 e1 9 9

Table 2 e Experimental data of the ACM compressor.

Tin (K) Tout (K) pin (kPa) pout (kPa) n (R s1) Uc (m s1) Gc (kg s1) pc hc Wc (kW)
283.8 359.3 406.5 758.1 998 291.3 0.346 1.86 0.73 26.2
283.7 357.7 384.0 723.3 995 290.6 0.356 1.88 0.76 26.4
283.4 354.3 349.1 664.0 988 288.6 0.367 1.90 0.81 26.1
283.6 349.6 526.6 922.1 929 271.3 0.394 1.75 0.75 26.2
283.4 348.9 520.8 912.1 928 271.0 0.394 1.75 0.75 26.0
283.1 344.3 558.5 927.3 861 251.4 0.371 1.66 0.72 22.8
282.8 339.0 527.4 827.4 845 246.8 0.369 1.57 0.69 20.9
282.6 331.6 459.0 716.2 827 241.5 0.382 1.56 0.78 18.8
282.6 330.9 410.5 639.1 840 245.4 0.397 1.56 0.79 19.3
282.2 334.2 384.0 595.2 844 246.5 0.397 1.55 0.72 20.7
282.1 337.0 340.9 550.2 877 256.0 0.368 1.61 0.75 20.3
283.9 364.2 273.0 554.2 1094 319.6 0.389 2.03 0.79 31.4
283.8 364.9 258.7 515.1 1096 320.2 0.392 1.99 0.76 31.9
283.6 352.6 254.0 494.1 1096 320.1 0.394 1.95 0.86 27.4
283.6 332.9 432.6 501.6 833 243.3 0.561 1.16 0.25 27.8
283.3 343.1 337.0 431.4 924 269.8 0.481 1.28 0.35 28.9
283.0 353.3 275.2 389.0 1005 293.6 0.414 1.41 0.42 29.2
282.9 370.5 202.0 324.0 1108 323.6 0.351 1.60 0.47 30.9
282.5 385.0 151.0 276.8 1214 354.4 0.278 1.83 0.52 28.6
282.6 381.7 168.0 295.8 1180 344.6 0.292 1.76 0.50 29.0
282.5 391.4 210.3 402.1 1248 364.3 0.411 1.91 0.53 45.0
282.6 384.8 233.5 423.2 1207 352.3 0.439 1.81 0.51 45.1
282.8 376.8 262.8 449.8 1159 338.6 0.469 1.71 0.50 44.3
282.9 366.3 312.0 469.0 1093 319.2 0.538 1.5 0.42 45.1
283.1 353.1 392.0 552.0 1002 292.7 0.597 1.41 0.42 42.0
283.3 331.3 487.7 622.0 914 266.8 0.686 1.28 0.43 33.1
283.5 327.1 542.2 666.3 871 254.4 0.728 1.23 0.39 31.9
283.0 362.4 337.5 512.4 1063 310.5 0.570 1.52 0.45 45.5
284.1 369.8 312.0 631.0 1071 312.6 0.322 2.02 0.74 27.8
284.1 369.5 303.7 620.2 1078 314.8 0.325 2.04 0.75 27.9
283.9 367.2 287.0 582.3 1070 312.5 0.332 2.03 0.76 27.8
283.8 363.3 270.0 548.2 1077 314.5 0.336 2.03 0.80 26.9
283.7 362.2 259.0 525.0 1081 315.7 0.339 2.03 0.81 26.7
283.6 364.3 245.8 503.0 1093 319.1 0.342 2.05 0.80 27.7
283.4 363.6 228.9 448.0 1083 316.3 0.346 1.96 0.75 27.9
283.3 364.3 215.5 406.0 1083 316.2 0.342 1.88 0.69 27.8
283.2 360.5 186.0 308.3 1049 306.3 0.294 1.66 0.57 22.9
283.3 361.7 188.5 340.4 1049 306.3 0.287 1.81 0.67 22.6
283.8 364.3 256.4 506.4 1044 304.9 0.256 1.98 0.76 20.7
283.6 359.0 229.8 448.2 1041 304.0 0.264 1.95 0.79 20.0
283.4 356.1 208.0 398.0 1042 304.2 0.271 1.91 0.79 19.8
282.8 353.2 193.0 351.2 1041 303.9 0.280 1.82 0.75 19.8
283.8 383.2 186.6 335.7 1206 352.2 0.331 1.80 0.52 33.0
283.8 379.2 202.1 347.7 1169 341.5 0.349 1.72 0.50 33.4
284.1 364.2 201.1 398.0 1070 312.5 0.438 1.98 0.76 35.3
284.1 342.8 332.0 457.0 987 288.1 0.488 1.38 0.46 28.8
284.2 331.7 389.0 496.0 921 268.8 0.553 1.28 0.43 26.4
284.2 324.8 415.0 512.0 891 260.2 0.574 1.23 0.43 23.4
284.1 339.0 403.0 533.3 948 276.8 0.592 1.32 0.43 32.6
283.9 350.2 324.7 519.5 1035 302.1 0.514 1.60 0.62 34.2
283.6 362.8 270.9 474.0 1111 324.4 0.461 1.75 0.62 36.7
283.5 373.2 239.0 409.0 1159 338.4 0.412 1.71 0.52 37.1
283.3 371.0 201.0 373.0 1222 356.9 0.381 1.86 0.62 33.5
283.3 369.0 205.0 379.0 1213 354.3 0.382 1.85 0.64 32.9

Its weakness, however, is that for a particular model its value
tends to increase when the sample size is reduced and the
model thus gets closer to being saturated. R2c corrects this
overestimation problem by considering the number of pre-
dictors in the model and is generally considered superior
especially when comparing models with different numbers of
predictors. s2 plays an important role in hypothesis testing in
multiple regressions. It can also provide important
information in an exercise from which one hopes to select the
best model for prediction. A common practice is to choose the
candidate model with the smallest value of s2. From Eq. (36) it
can be seen that R2c is identical to s2 in terms of ranking.
Therefore, R2c is chosen as the criterion for evaluating the
above models. Besides, the mean absolute deviation MAD and
the percentage of data points within relative deviation (RD) of
5% and 10% are used as supplemental criteria.
 i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 4 1 ( 2 0 1 4 ) 1 9 0 e1 9 9 197

Table 3 e Mass flow models: Predictions vs. the ACM

compressor data.
a b
Model R2c MAD, %
Jensen et al. 0.919 4.6 66.7 90.7
Mueller 0.895 5.6 56.6 83.3
Kolmanovsky et al. 0.832 5.6 72.7 81.5
Guzzella-Amstutz 0.586 14.4 31.5 50.0
Andersson 0.773 8.9 40.7 70.4
Eriksson et al. 0.617 12.6 29.6 44.4
Biteus 0.620 12.7 33.3 50.0
Sieros et al.: Nonlinear 0.375 26.9 11.1 27.8
Sieros et al.: Simple linear 0.424 20.5 9.3 25.9
Sieros et al.: Generalized linear 1 0.413 19.9 14.8 29.6
Sieros et al.: Generalized linear 2 0.648 16.1 22.2 46.3
a
Percentile of data points within the RD of 5%, (%).
b
Percentile of data points within the RD of 10%, (%).

ypred;i  yexp;i
RDi ¼
yexp;i
1 XN
MAD ¼ jRDi j
N i¼1
4.2. Evaluation of mass flow models with the ACM
compressor data
The 54 experimental data points of the 2-wheel ACM
compressor are used to evaluate the reviewed models.
The above mass flow models present mass flow in different
ways: the mass flow rate Gc, the mass flow parameter fc, or
the corrected mass flow parameter fcorr. Most of compressor
performance maps use either fc or fcorr, and the two have the
same evaluation results. Besides, fcorr has the same unit as Gc.
Therefore, fcorr is chosen as target for comparison. The results
are listed in Table 3.
Table 3 shows that in most cases the ranking by R2c co-
incides with that by MAD. Based on accuracy, the Jensen et al.
(1991) model performs best with R2c ¼ 0.919 and MAD ¼ 4.6%,
following which are the Mueller (1997) model (R2c ¼ 0.919 and
MAD ¼ 5.6%) and Kolmanovsky et al. (1997) model (R2c ¼ 0.832
and MAD ¼ 5.6%).
Fig. 3 shows the RD distribution of the corrected mass flow
parameter of the top three models. It can be seen that Jensen
et al. (1991) model has the smallest MAD when
fcorr  0.12 kg s1, and the Kolmanovsky et al. (1997) model has
the smallest MAD when fcorr > 0.12 kg s1.
4.3. Evaluation of efficiency models with the ACM
compressor data
(37)
Fig. 3 e RD of corrected mass flow parameter of the top
three models against the ACM compressor data.
(38)
when pc > 1.2. The biggest deviations of the top three models
occur at pc < 1.2.
The Sieros et al. simple linear model is implicit, and the
GuzzellaeAmstutz (1998) model and the Andersson (2005)
model need the point of the maximum efficiency hmax to be
known. These requirements increase difficulties of imple-
menting them.
5. Comparison of the model predictions with
turbocharger centrifugal compressor data
The experimental data of turbocharger centrifugal compres-
sors from Moraal and Kolmanovsky (1999) and Jung (2003) are
used for the comparison. The data were given in graphs,
which do not provide enough information for evaluating some
models. The parameter ranges provided by Moraal and
Kolmanovsky (1999) are the rotational speeds from 90,000 to
140,000 rpm, the corrected mass flow rate from 0.0410 to
0.2619 kg s1, and the pressure ratio from 1.15 to 1.80. The
parameter ranges of the Jung (2003) experiment are the
normalized rotational speeds in the range from 20,000 to
Table 4 e Efficiency models: Predictions vs. the ACM
compressor data.
Model R2c MAD, % a b

Jensen et al. 0.606 12.5 24.1 48.1

The results are listed in Table 4, from which it can be seen that Guzzella-Amstutz 0.491 19.6 37.0 46.3
the top three models in light of accuracy are the Canova et al. Andersson 0.433 22.4 24.1 44.4
(2009) model (R2c ¼ 0.947and MAD ¼ 4.9%), the Sieros et al. Canova et al. 0.947 4.9 59.3 88.9
(1997) generalized linear model 1 (R2c ¼ 0.931 and Eriksson et al. 0.543 13.4 24.1 46.3
Sieros et al.: Simple linear 0.886 7.0 50.0 74.1
MAD ¼ 5.5%), and the Sieros et al. (1997) generalized linear
Sieros et al.: Generalized linear 1 0.931 5.5 53.7 85.2
model 2 (R2c ¼ 0.926 and MAD ¼ 5.8%).
Sieros et al.: Generalized linear 2 0.926 5.8 53.7 87.0
Fig. 4 shows the RD distribution of the compressor effi- a
Percentile of data points within the RD of 5%, (%).
ciency of the top three models with compression ratio pc. It b
Percentile of data points within the RD of 10%, (%).
can be seen that the MADs of top three models are very close
198 i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 4 1 ( 2 0 1 4 ) 1 9 0 e1 9 9

Table 6 e Mass flow models: Predictions vs. the Jung

(2003) experimental data.
a b
Model R2c MAD, %
Eriksson et al. 0.963 5.4 61.4 82.5
Biteus 0.968 5.3 64.9 86.0
Sieros et al.: Nonlinear 1 0.005 100.0 100.0
Sieros et al.: Simple linear 0.812 16.4 19.3 49.1
Sieros et al.: Generalized linear 1 0.950 7.6 45.6 79.0
Sieros et al.: Generalized linear 2 0.999 0.03 100 100
a
Percentile of data points within the RD of 5%, (%).
b
Percentile of data points within the RD of 10%, (%).

5.2. Comparisons of the efficiency models

The evaluation results of the efficiency models of Jensen et al.

(1991), GuzzellaeAmstutz (1998), Andersson (2005), and Sieros
et al. (1997) against the Jung (2003) data are listed in Table 6,
where the models of Eriksson et al. (2002) and Canova et al. (2009)
Fig. 4 e RD distribution of efficiency with compression ratio
do not show up because of the insufficient data information.
of the top three models against the ACM compressor data.
It can be seen from Table 7 that the Sieros et al. (1997)
generalized linear model 2 performs best with R2c ¼ 0.980 and
110,000 rpm, the corrected mass flow rate from 0.0833 to MAD ¼ 4.8%, followed by the Sieros et al. (1997) generalized
0.0722 kg s1, and the pressure ratio from 1.01 to 1.48. linear model 1 (R2c ¼ 0.978 and MAD ¼ 5.3%) and the Jensen
et al. (1991) model (R2c ¼ 0.971 and MAD ¼ 6.1%). The rank-
5.1. Comparisons of the mass flow models ings showing in Tables 7 and 4 are quite different, indicating
that the performance of an efficiency model varies with
The evaluation results of the mass flow models based on the compressor type.
Moraal and Kolmanovsky (1999) data are listed in Table 5. It
can be seen that the top three models are those of Jensen et al.
(1991), Mueller (1997) and Kolmanovsky et al. (1997), with the 6. Conclusions
R2c of 0.992, 0.989, and 0.915 respectively. This ranking is the
same as that based on the ACM compressor discussed above. (1) A comprehensive review of the empirical models for
The evaluation results of the mass flow models of Eriksson efficiency and mass flow rates of centrifugal compres-
et al. (2002), Biteus (2006), and Sieros et al. (1997) based on the sors are conducted, and eleven mass flow models and
Jung (2003) data are shown in Table 6. It can be seen that the eight efficiency models are found. All those studied
Sieros et al. (1997) nonlinear model performs best (R2c ¼ 1 and models were developed for vehicle engine and turbo-
MAD ¼ 0.005%), followed by the Sieros et al. (1997) linear charger compressors.
model 2 (R2c ¼ 0.999 and MAD ¼ 0.03%) and the Biteus (2004) (2) All the models have constants needing to be determined
model (R2c ¼ 0.968 and MAD ¼ 5.3%). These results are quite by experimental data. Among the commonly used
different from those with the ACM compressor data and the criteria for comparing a set of models with their con-
Moraal and Kolmanovsky (1999) data, implying that the per- stants determined with the same database, the cor-
formance of a mass flow model varies with compressor type. rected coefficient of determination R2c is the favorable
choice. The mean absolute deviation MAD is used as a
criterion in this study because its explicitness for
Table 5 e Mass flow models: Predictions vs. experimental
showing deviations. It is shown that the ranking by
data in Moraal and Kolmanovsky (1999).
MAD largely coincides with that by R2c.
a b
Model R2c MAD, %
Jensen et al. 0.992 1.6 96.4 100.0
Mueller 0.989 2.1 96.4 100.0 Table 7 e Efficiency models: Predictions vs. the Jung
Kolmanovsky et al. 0.915 14.2 46.4 71.4 (2003) experimental data.
Guzzella-Amstutz 0.792 30.8 14.3 25.0 a b
Andersson 0.613 33.2 14.3 17.9 Model R2c MAD
Eriksson et al. 0.800 26.8 21.4 46.4 Jensen et al. 0.971 6.1 58.0 80.0
Biteus 0.882 19.3 28.5 50.0 Guzzella-Amstutz 0.927 10.9 34.0 58.0
Sieros et al.: Nonlinear 0.564 44.2 10.7 25.0 Andersson 0.790 22.2 12.0 30.0
Sieros et al.: Simple linear 0.598 88.8 3.6 14.3 Sieros et al.: Simple linear 0.919 11.6 20.0 58.0
Sieros et al.: Generalized linear 1 0.691 35.4 7.1 21.4 Sieros et al.: Generalized linear 1 0.978 5.3 56.0 88.0
Sieros et al.: Generalized linear 2 0.602 55.8 7.1 25.0 Sieros et al.: Generalized linear 2 0.980 4.8 60.0 86.0
a a
Percentile of data points within the RD of 5%, (%). Percentile of data points within the RD of 5%, (%).
b b
Percentile of data points within the RD of 10%, (%). Percentile of data points within the RD of 10%, (%).
 i n t e r n a t i o n a l j o u r n a l o f r e f r i g e r a t i o n 4 1 ( 2 0 1 4 ) 1 9 0 e1 9 9
(3) A test rig for aircraft air-cycle refrigeration system with
a bootstrap ACM having a centrifugal compressor was
built, and experiments were conducted from which 54
experimental data were obtained.
(4) The applicability of the reviewed models for vehicle
engine and turbocharger centrifugal compressors to
refrigeration centrifugal compressors is evaluated
against the experimental data of the bootstrap ACM
compressor. For mass flow rates, the models of Jensen
et al. (1991), Mueller (1997), and Kolmanovsky et al.
(1997) are among the top three in light of accuracy. For
efficiency, the top three models are the Canova et al.
(2009) model and the Sieros et al. (1997) generalized
linear models. It is found that for refrigeration centrif-
ugal compressors accurate models remain a problem
unsolved though some existing models are acceptable.
(5) The reviewed models are evaluated with turbocharger
centrifugal compressor data of Moraal and
Kolmanovsky (1999) and Jung (2003). Some models
were said to have very high accuracy in the source
literature, which is not consistent with the evaluation in
this study. The main reason is that the source files only
verified the proposed model(s) in narrow parameter
ranges for the given compressor(s). On the other hand,
the evaluation results in this work cannot be used to
rank the reviewed models because different data sour-
ces may yield quite different evaluation results.
(6) The model performance and ranking vary with com-
pressors. Therefore, careful assessment should be made
in selecting suitable models for a given compressor.
references
Andersson, P., 2005. Air Charge Estimation in Turbocharged Spark
Ignition Engines. Thesis No. 989. Department of Electrical
Engineering, Linkoping University, Sweden.
Ashford, R., Brown, S., 2000. F-22 Environmental Control System/
thermal Management System (ECS/TMS) Flight Test Program
e Downloadable Constants, an Innovative Approach. SAE
2000-01-2265.
Biteus, J., 2004. Mean Value Engine Model of a Heavy Duty Diesel
Engine. LITH-ISY-R-2666. Department of Electrical
Engineering, Linkopings universitet, Sweden.
Canova, M., 2004. Development and validation of a control-
oriented library for the simulation of automotive engines. Int.
J. Engine Res. 5 (3), 219e228.
Canova, M., Midlam-Mohler, S., Guezennec, Y., Rizzoni, G., 2009.
Mean value modeling and analysis of HCCI diesel engines with
external mixture formation. J. Dyn. Syst. Meas. Control 131
(011002), 1e14.
Chamoun, M., Rulliere, R., Haberschill, P., Peureux, J.-L., 2013.
Modelica-based modeling and simulation of a twin screw
compressor for heat pump applications. Appl. Therm. Eng. 58,
479e489.
Chu, F., Wang, F.L., Wang, X.G., Zhang, S.N., 2012. Performance
modeling of centrifugal compressor using kernel partial least
squares. Appl. Therm. Eng. 44, 90e99.
199
Conceição, S.T., Zaparoli, E.L., Turcio, W.H.L., 2007.
Thermodynamic Study of Aircraft Air Conditioning Air Cycle
Machine: 3-wheel X 4-wheel. SAE 2007-01-2579.
Eriksson, L., Nielsen, L., Brugard, J., Bergstriim, J., Pettersson, F.,
Andersson, P., 2002. Modeling of a turbocharged SI engine.
Annu. Rev. Control 26, 129e137.
Fang, X.D., Dai, Q.M., 2010. Modeling of turbine mass flow rate
performances using the Taylor expansion. Appl. Therm. Eng.
30 (13), 1824e1831.
Fang, X.D., Dai, Q.M., Yin, Y.X., Xu, Y., 2010. A compact and
accurate empirical model for turbine mass flow
characteristics. Energy 35, 4819e4823.
Fang, X.D., Xu, Y., 2011. Development of an empirical model of
turbine efficiency using the Taylor expansion and regression
analysis. Energy 36, 2937e2942.
Fang, X.D., Xu, Y., 2013. Correlations for two-phase friction
pressure drop under microgravity. Int. J. Heat Mass Transfer
56, 594e605.
Fast, M., Assadi, M., De, S., 2009. Development and multi-utility of
an ANN model for an industrial gas turbine. Appl. Energy 86,
9e17.
Guzzella, L., Amstutz, A., 1998. Control of diesel engines. IEEE
Control Syst. Mag. 18 (5), 53e71.
He, J., Zhao, J.Q., Sun, W., 2004. Dynamic simulation of bootstrap
air cycle refrigeration components for aircraft environmental
control system. J. Syst. Simul. 16 (4), 727e729.
Jensen, J.P., Kristensen, A.F., Sorenson, S.C., Houbak, N.,
Hendricks, E., 1991. Mean Value Modeling of a Small
Turbocharged Diesel Engine. SAE 910070.
Jung, M., 2003. Mean-value Modelling and Robust Control of the
Air Path of a Turbocharged Diesel Engine. Ph.D. thesis.
Department of Engineering, University of Cambridge, UK.
Kolmanovsky, I.V., Moraal, P.E., van Nieuwstadt, M.J., Criddle, M.,
Wood, P., 1997. Modeling and Identification of a 2.0 L
Turbocharged DI Diesel Engine. Ford internal technical report
SR-97-039.
Letlow, J.T., Jenkins, L.C., 1998. Development of an Integrated
Environmental Control System. SAE 981544.
Li, S.J., Li, F., 2012. Prediction of cracking gas compressor
performance and its application in process optimization.
Chin. J. Chem. Eng. 20 (6), 1089e1093.
Liu, X.H., Sun, D.K., Sun, X.F., Wang, X.Y., 2012. Flow stability
model for fan/compressors with annular duct and novel
casing treatment. Chin. J. Aeronaut. 25, 143e154.
Moraal, P., Kolmanovsky, I., 1999. Turbocharger modeling for
automotive control application. SAE Trans. 108 (3), 1324e1338.
Mueller, M., 1997. Mean Value Modeling of Turbocharged Spark
Ignition Engines. Master’s thesis. DTU, Denmark.
Myers, R.H., 2005. Classical and Modern Regression with
Applications, second ed. High Education Press, Beijing, China.
Ordonez, J.C., Bejan, A., 2003. Minimum power requirement for
environmental control of aircraft. Energy 28, 1183e1202.
Schiffmann, J., Favrat, D., 2010. Design, experimental
investigation and multi-objective optimization of a small-
scale radial compressor for heat pump applications. Energy 35,
436e450.
Sieros, G., Stamatis, A., Mathioudakis, K., 1997. Jet engine
component maps for performance modeling and diagnosis. J.
Propuls. Power 13, 665e674.
Tsujikawa, G.S., 1985. Digital controlled closed loop air cycle
development. SAE 851319.
Wang, X.D., Hirsch, C., Kang, S., Lacor, C., 2011. Multi-objective
optimization of turbomachinery using improved NSGA-II and
approximation model. Comput. Methods Appl. Mech. Eng. 200
(9e12), 883e895.