Preprints of the 18th IFAC World Congress

Milano (Italy) August 28 - September 2, 2011

Comparison of a turbocharger model based

on isentropic eciency maps with a
parametric approach based on Eulers
turbo-machinery equation
Andreas Sidorow Rolf Isermann Francesco Cianone
Gerhard Landsmann
Institut of Automatic Control, Technical University Darsmtadt
(e-mail: asidorow@iat.tu-darmstadt.de

Institut of Automatic Control, Technical University Darsmtadt

(e-mail: risermann@iat.tu-darmstadt.de

GM Powertrain Europe, Turin, Italy (e-mail:

francesco.cianone@gm.com

GM Powertrain Europe, Ruesselsheim, Germany (e-mail:

gerhard.landsmann@gm.com

Abstract: Faults in the intake and exhaust path of turbocharged common rail Diesel engines can
lead to an increase of emissions and performance losses. Application of turbocharger models can
help to detect and diagnose more faults as standard fault detection methods. The modeling of the
turbocharger for onboard fault diagnosis can be obtained by dierent models. The dierences
between an approach based on the isentropic eciencies and an approach based on Eulers
turbo-machinery equation are investigated in this paper. The two models for a GT1749MV
turbocharger are parameterized with data from the engine testbed. The comparison is applied
by issues of measured model inputs, number of intern parameters, parameterization eort and
model accuracy. Both models are compared regarding the application for onboard diagnosis.
Keywords: Turbocharger modeling, heat transfer, isentropic eciency, Eulers turbo-machinery
equation
1. INTRODUCTION
Dynamic models of turbocharged Diesel engines are
needed for purposes of simulation assisted development of
engine parts, control and diagnosis functions. The intake
and exhaust path is an important part of the engine model,
due to its growing complexity and nonlinear dynamics.
The turbocharger model has a signicant inuence on the
performance of the complete air and exhaust path model.
Mrosek and Isermann (2010) and Shaaban (2004) show
that consideration of the heat transfers in the turbocharger
housing leads to better model accuracy. Turbocharger
eciency maps usually delivered by the manufacturer
are gained from the turbocharger hot gas test bench.
They contain measurement points at medium to high
turbocharger speeds with negligible eects of heat transfer
and dont reproduce the inuence of pulsations occurring
at the engine. Parameterization of the turbocharger using
extrapolation from these measurement points, Guzzella
(2009), leads nonetheless to insucient modeling results.
The alternative parameterization way is to use measurements from engine test bench in the operation region with
low to high turbocharger speeds in order to take the heat
transfer as well as engine pulsations into account.
In Mrosek and Isermann (2010) a uiddynamic semiphysical model based on Eulers turbo machinery equation
Copyright by the
International Federation of Automatic Control (IFAC)

with consideration of the heat ows in the turbocharger

is presented. In Sidorow et al. (2011) the separation of
aerodynamic compression respectively expansion from the
heat ow is applied on a thermodynamic model based on
isentropic eciencies. The modeling approaches are both
veried on measurements from the the dynamic engine
test bench at Institute of Automatic Control. In this
contribution both methods are compared.
2. TURBOCHARGER MODELS
The model of turbocharger with speed as output is
obtained from models of the compressor power , turbine
power and friction power combined in (1) taking the
inertia of compressor and turbine wheels and shaft into
account, Merker et al. (2005).

=
(1)
(2)2
Friction power is modeled using viscous friction equation
= (2)2 2
(2)
with friction coecient . The power models of the
compressor and turbine compose of massow and adiabatic
enthalpy dierence submodels
=
, ,
(3)
=
, ,
(4)

10627

Preprints of the 18th IFAC World Congress

Milano (Italy) August 28 - September 2, 2011

a)

2,

with , = 1050
the specic heat capacity at constant
pressure for the exhaust gas and 3 , 4 not measured
temperatures.

1
1

3
b)

3
3

,
,

4,

applied for the diabatic expansion process in the turbine

illustrated in Fig. 1 b). The specic heat outow before
the expansion , , adiabatic expansion , and specic
heat outow after the expansion , can be expressed in
the equation
, = , (3 4 )
(10)
= , + , + ,
(11)
= , [(3 3 ) + (3 4 ) + (4 4 )] (12)

Fig. 1. Compression process a) and expansion process b)

in a schematic h-s diagramm
The compressor- and turbine massow models are given
according to Zahn (2007), Zahn and Isermann (2008) by
2
, , 1 )
(5)

, = (
1
3

, = ( , )
(6)
4
and are not the focus of this paper.
The compression process is illustrated in Fig. 1 a) using hs-diagramm with lines of constant pressure 1 and 2 and
measured temperatures 1 and 2 . The diabatic enthalpy
dierence , , obtained from the measured temperatures represents both compression process and heat transfer. In order to separate the adiabatic compression from
the heat inow, the diabatic process can be divided into
three parts. Thereby heat transfer into the compressor is
assumed to occur on two lumped locations on the ow
path (Shaaban (2004)). The specic heat inow before
compression , leads to temperature increase from 1
to 1 . The adiabatic compression of the intake air from
pressure 1 to 2 is attended with temperature increase

. The specic heat inow after compression

from 1 to 2

, leads to temperature increase from 2

to 2 . The
introduced parts of diabatic compression can be expressed
by means of thermodynamic enthalpy denition, compare
Zahn and Isermann (2008)
, = , (2 1 )
(7)
= , + , + ,
(8)

= , [(1 1 ) + (2
1 ) + (2 2
)] (9)

the specic heat capacity at conwith , = 1080

stant pressure for the fresh air. Similar considerations are

Previous works of Shaaban (2004), Mrosek and Isermann

(2010) show that turbocharger is part of a complex thermal system which contains dierent heat ows between
compressor, turbine, engine housing, lubrication oil and
environment. The most signicant heat input before and
after the adiabatic compression is assumed to come from
the turbine over the housing, see Mrosek and Isermann
(2010). According to Newtons law of heat transfer the
temperatures before and after adiabatic compression can
be obtained from equations
,, ,
1 = 1 +
(3 1 )
(13)
,

,, ,

2
= 2
(3 2 )
(14)
,

and the temperatures before and after adiabatic expansion

concerning the heat outow from the turbine are given by
,, ,
(3 1 )
(15)
3 = 3
,

,, ,
4 = 4 +
(4 1 )
(16)
,

with heat transfer coecients ,, , ,, , ,, , ,,

and heat transfer areas , , , , , and ,
The parameterization of the adiabatic enthalpy dierences using measured temperatures is applied together
with the heat transfer models. Thereby the nonmeasured

, 3 and 4 are estimated which

temperatures 1 , 2
can be interpreted as a shift of the line describing adiabatic compression and adiabatic expansion in Fig. 1 as
is implied by the grey arrows. The thermodynamic and
uid-dynamic approach for adiabatic enthalpy dierences
, , , is introduced in the following sections.
2.1 Thermodynamic approach
The turbocharger modeling with thermodynamic approach is based on isentropic eciencies of the compressor
and turbine according to Sidorow et al. (2011) and considers heat transfer in turbocharger housing. In contrary to
Merker et al. (2005), Guzzella (2009) model is parameterized from the engine test bench measurements instead of
extrapolation of the OEMs turbocharger maps.
Compressor
The adiabatic enthalpy dierence of the
compressor according to equations (8), (9) depends on the

unmeasured temperatures 1 and 2

. The temperature

1 is obtained from (13). The temperature after the com

pression 2
can be calculated using of isentropic eciency
of the compressor which shows the entropy increase

10628

Preprints of the 18th IFAC World Congress

Milano (Italy) August 28 - September 2, 2011

during the adiabatic compressing process in relation to

isentropic compression process, Watson and Janota (1982):
,
=
(17)
,
The isentropic compression is shown as vertical line in the

- diagram (Fig. 1 a)) from point 1 to 2,

and can be
expressed as:
)
(
(18)
, = , 2,
1
The temperature 2, is obtained from Merker et al.
(2005)
) 1
(
2

(19)
2, = 1
1
with isentropic exponent = 1, 399 of air. Substitution
of (19) in (18) and (18) in (17) yields
(( ) 1
)

2
, 1

1
1
(20)
=
,
In order to avoid the calculation in (20) the isentropic
eciency is modeled as neuronal net of type LOLIMOT
(see Nelles (97)) with inputs turbocharger speed and
compressor massow
, which are corrected by reference
conditions and according to Merker et al. (2005),
Guzzella (2009):
= (, ,
, )
(21)

, =
(22)
1

1
(23)

, =

Solving (20) for , yields

((
)
) 1
, 1
2
, =
1

(24)

Turbine The adiabatic enthalpy dierence of the turbine

is introduced in (10) concerning Fig. (1 b)). The temperature 3 is calculated using (14). The temperature after the
adiabatic expansion process 4 can be calculated with help
of isentropic eciency , see Watson and Janota (1982)
,
=
(25)
,
with isentropic expansion (see Fig 1 b)).
)
(
(26)
, = , 4,
3
Applying equation
( ) 1

4,
= 3
(27)
3
with isentropic exponent = 1, 361 of exhaust gas,
compare Merker et al. (2005), and substitution analog to
the compressor yields:
,
=
(28)
(
( ) 1 )

, 3 1 3
According to Guzzella (2009) depends on turbine blade
speed ratio

1
1

with prewhirl
without prewhirl
1

1
1

)
2
2,

2,
2
2,
2,
2,,

2,
ideal
2 real

=2,,

2,
=2,
2

Fig. 2. Impeller velocity triangles according to Zahn and

Isermann (2008)
=

3
)
(

( ) 1

2, 3 1 34

(29)

which is calculated inter alia from , and variable geometry actuator . Thereby is 3 diameter of the turbine
wheel. Due to measurement heat transfer inuence, is
interpolated by polynomials, see Sidorow et al. (2011) and
further is modeled according to Zahn (2007) as LOLIMOT:
= ( , )
(30)
The dependency of on is considered in (29). Solving
(28) for , yields:
(
( ) 1 )

(31)
, = , 3 1
3
2.2 Fluid dynamic approach
The turbocharger modeling with uiddynamic approach
is based on Eulers equation of turbo-machinery. The adiabatic enthalpy dierence of compressor and turbine are
calculated according to Watson and Janota (1982), Zahn
and Isermann (2008), Mrosek and Isermann (2010) using
the parametric approach derived from physical equations
of uiddynamic theory.
Compressor The adiabatic enthalpy dierence of compressor is derived from the velocity triangles of the impeller
in- and outlet shown in Fig. 2. From Eulers equation of
turbo-machinery yields:
, = 2 2, 1 1
(32)
with peripheral velocities 2 and 1 and peripheral components of the absolute velocity at the impeller inlet 1
and outlet 2, , see Fig. 2. Neglecting impeller inlet component yields:
, = 2 2,
(33)
with diameter of the compressor wheel 2 . From Fig. 2 b)
yields
2, = (2 2, cot (2, ))
(34)

10629

Preprints of the 18th IFAC World Congress

Milano (Italy) August 28 - September 2, 2011

Table 1. Compressor model parameters

3
3

4
4
4
4

with exit swirl

textwithout exit swirl

Fig. 3. Turbine velocity triangles according to Zahn and

Isermann (2008)
with the meridional component of the absolute velocity
2, and slip factor , compare Stodola (1945)
2,
(35)
=
2,,
where
2, = 2,,
(36)
with simplied approach
= 2
(37)
Substitution of the (37), (36) and (35) yields:
2
(38)
=1
2 2, cot 2,
Using the mass ow continuity equation and compressor
blade width 2 one gets:

2, =
(39)
2 2 2
with compressor outlet air density
2
(40)
2 =

2
By substitution of (34) and (39) in (33) and taking (38)
into account, adiabatic enthalpy dierence , can be
expressed as:
(
)

2
, = (2 )
(41)
cot 2,
2 2
Turbine The expansion in the turbine can be modeled
analog to the previously described. Approach similar to
(32) is assumed:
, = 3 3 4 4
(42)
Neglecting the swirl at turbine outlet yields:
, = 3 3
(43)
where 3 is turbine diameter.

method

compressor parameters

thermodynamic
uiddynamic

,, , , ,, , , (, ,
, )
cot(2 )
,, , , ,, , , ,

as polynomial function of 4th order dependent on the guide

vanes actuator position :
cot 3 = ( )
(45)
3 is derived from the mass ow continuity equation

3 =
(46)
3 3 3
with turbine blade width 3 and the exhaust gas density
3
3 =
(47)
3
Finally the adiabatic enthalpy dierence is expressed by

cot 3
(48)
, =
3 3
3. COMPARISON
Modeling approaches presented in last sections contain parameters which have to be obtained applying identication
methods. The eects of ow unsteadyness are considered
in both models rst by using the pressures before and after
the compressor respectively turbine as inputs of massow
as well as of enthalpy dierence models. Further engine
test bench measurements containing gas pressure and
massow oscillations are used for parameter estimation.
In following the dierence in the parameterization eort
of both models is outlined. Furthermore the modeling
approaches are sompared concerning model inputs and
model accuracy.
3.1 Comparison concerning identication issues
Unknown parameters of compressor and turbine model
have to be estimated iteratively by minimizing the model
error which is given by mathematical norm between the
calculated model outputs and measurements. The compressor parameters to be estimated are summarized in the
table 1 for the thermodynamic and uiddynamic approach.
The model parameters of the turbine are summarized in
the analog manner in the table 2 for both methods.
The parameters of the compressor model can be identied from measured adiabatic enthalpy with corrected

temperatures 1 and 2
, (13), (14) and , from the
corresponding thermodynamic respectively uiddynamic
calculation. The parameters from table 1 have to be estimated minimizing compressor model error

= , , (2
1 ) 2
(49)

According to velocity triangle at turbine inlet shown in

Fig. 3 a) the equation
3 = 3 cot 3
(44)
with meridian component of the absolute velocity at the
turbine inlet 3 is derived. Considering the variable turbine geometry actuator, the parameter cot 3 is modeled
10630

Table 2. Turbine model parameters

method

turbine parameters

thermodynamic

,, , , ,, , , ( , )

uiddynamic

,, , , ,, , ,

cot 3 ( )
3

Preprints of the 18th IFAC World Congress

Milano (Italy) August 28 - September 2, 2011

out measured data. However the reliability is limited by

the heat transfer rate which increases with opening turbine
geometry actuator and lower engine speed. Both thermodynamic and uiddynamic turbocharger models contain
additionally the compressor- and turbine massow models, which are calculated by (5) and (6) and have to be
parametrized separately minimizing the error between the
measured and modeled massow. Finally the turbocharger
speed can be calculated according to (1).

1
0.5

0
0.5
reachable operation
area for considered
engine

1
1.5
2
3000

2000
1000
0

ntc,corr in 1/s

0.02

0.03

0.05

0.04

0.06

0.07

0.08

dm/dtc,corr in kg/s

Fig. 4. Compressor eciency map with corresponding

engine test bench measurement points

The thermodynamic model contains, in addition to constant parameters, two neuronal nets (, ,
, )
and ( , ) of type LOLIMOT, which have to be parameterized in every iteration step. These models compose
of several local linear models, which are interpolated by
Gaussians, see Nelles (97). The amount of parameters for
a LOLIMOT net with inputs and local linear models is
dened by = ( +1)+2 . The number of parameters
of isentropic eciencies in thermodynamic model is shown
in table 3. Contrary to the thermodynamic model, the uTable 3. LOLIMOT models of and

1
0.8

0.6
0.4

model

inputs

local models

parameters

2
2

7
3

49
21

0.2
0
4000

iddynamic model contains only constant parameters which

have to be estimated. Summing up, the 74 parameters
of the thermodynamic approach and the circumstance of
training the neuronal nets in every iteration step show a
higher parameterizing eort then the 10 parameters of the
uiddynamic model.

operation area with

engine test bench
measurement points

3000

0.2

2000
0.15
1000
n

tc,corr

in [1/s]

0.1
0

0.05
0
dm/dt

c,corr

in [kg/s]

Fig. 5. Compressor eciency map calculated with OEM

data
Therefore nonlinear global optimization with Matlab function under estimated constraints similar to
Mrosek and Isermann (2010) is applied. In gure 4 the calculated compressor map with corresponding measurement
data from the engine test bed is shown. The reachable
operation area is marked by black lines. In gure 5 the
eciency map calculated from hot gas test-bench measurement data provided by the OEM is shown. One can
see that measurement data points are mostly located in
operation area with high turbocharger speeds above 1500
1/s. The parameters of the turbine model (table 2) are
estimated in analog manner using model error
= , , (3 4 ) 2
(50)
In the next step the estimated compressor and turbine parameters are used for parameterizing of friction according
to (2). Therefore the power balance error
=
,
, 2
(51)
is minimized. For balancing of turbocharger powers the
optimization error can be constructed as weighted combination of (49), (50) and (51), accepting increase of complexity due to selection of weights and rising amount of
parameters. The turbocharger eciency maps obtained
over heat transfer estimation motivated by physics are
expected to be more accurate at lower speed turbocharger
operation region than the extrapolated OEM maps with-

3.2 Measured input analysis

The thermodynamic and uiddynamic models of compressor power (3) are analyzed in the following. The massow
model according to (5) is the common for both approaches
and reqires, regarding the feedback of , measured inputs
1 , 2 and 1 . Applying the thermodynamic approach, the
adiabatic enthalpy dierence , is calculated using
the temperature at compression begin 1 . In contrary
to this, , calculated using uiddynamic approach
has the gas density after the compression (40) as input.
Considering the heat transfers (13), (14) both modeling
approaches need the measured temperature 3 . Fluiddynamic model requires an additionally measurement of 2 .
By analyzing of both approaches of turbine power (4) using
equations (6), (16), (24) and (48) can be derived that
the inputs for both models are 3 , 4 , 3 , 4 and .
The required measured inputs are shown in the Fig. 6.
Summing up the uiddynamic model needs one measured
input, the temperature 2 , more than the thermodynamic
model.
3.3 Comparison concerning model accuracy
The practical application of introduced models for onboard
diagnosis is veried on measurement from the test bench
with a turbocharged 1.9 liter Opel Common Rail Diesel
engine. Experiment is designed with as wide as possible covering of the engine operation region. The control

10631

Preprints of the 18th IFAC World Congress

Milano (Italy) August 28 - September 2, 2011

a) thermodynamic approach
p1
p2c
T1
T3

p1
p2c
T1
T2c
T3

Pc
Compressor

p3
p4
svgt
T3
T1

Table 4. Comparizon results

b) fluiddynamic approach

.
mc

p3
p4
svgt
T3
T1

Pt
Turbine

Pc
Compressor

.
mt

.
mc

.
mt

in 1/min

3000

Eng

2000
1000

100

200

300

400

500
time in s

600

700

800

900

1000

100

200

300

400

500
time in s

600

700

800

900

1000

40
30

qInj in mm /cyl

inputs

RMSE in 1/s

thermodynamic
uiddynamic

74
10

10
11

60.57
55.53

isentropic eciencies of the compressor and turbine. The

uiddynamic model contains one measured input more
then the thermodynamic. Both models show good accuracy in the considered operation range.
5. CONCLUSION
In this contribution two dierent turbocharger modeling
approaches have been presented. Both approaches consider heat outow from the turbine and heat inow into
the compressor and are parameterized from measurements
accomplished on the dynamic engine test bench at Institute of Automatic Control. The introduced modeling
approaches are compared concerning identication issues,
parameterization eort, analysis of model inputs, accuracy. The uiddynamic model shows better performance
concerning the investigated criteria of comparison. Both
models are applicable for onboard fault diagnosis.

4000

20
10
0
0

ACKNOWLEDGEMENTS
The presented contribution is developed in cooperation
between the Institute of Automatic Control at TUDarmstadt and GM Powertrain Europe.

Fig. 7. Excitation signals for testbench measurement

3000
2500

REFERENCES

thermodynamic
fluiddynamic
measured

2000
1500

tc

n in 1/s

parameters

Pt
Turbine

Fig. 6. Comparison of both modeling approaches concerning measured model inputs

0
0

approach

1000
500
0
0

100

200

300

400

500
time in s

600

700

800

900

1000

Fig. 8. Model comparison: turbocharger speed

loops of high pressure exhaust recirculation (HPEGR) and
charge pressure are in series operation. The excitation signals contain steps in the fuel injection rate between 0
3
and 35
and engine speed between 1000 and 3500
rotations per minute, as shown in Fig. 7. The simulation
results can be found in Fig. 8. The model accuracy is calculated by root mean square error (RMSE), see Isermann
(2005). For fault diagnosis application of the models, high
model accuracy is signicant, see Sidorow et al. (2011).
The RMSE values of 55.53 1/s for the uiddynamic model
and 60.57 1/s for the thermodynamic model show excellent
performance of both models.
4. RESULTS
The results of the comparison can be summarized in table
4. Due to manageable amount of parameters uiddynamic
model is more suitable for the onboard applications. The
thermodynamic approach has a higher parameterization
eort, due to application of neuronal networks for the

Guzzella, L. (2009). Introduction to Modeling and Control

of Internal Combustion Engine Systems. Springer.
Isermann, R. (2005). Fault Diagnosis Systems. Springer.
Merker, G.P., Schwarz, C., Stiesch, G., and Otto, F.
(2005). Simulating Combustion. Springer, Berlin.
Mrosek, M. and Isermann, R. (2010). On the parametrisation of the turbocharger power and heat transfer models.
IFAC AAC 2010, Munich, Germany.
Nelles, O. (97). Lolimot - local linear model trees for
nonlinear dynamic system identication. at - Automatisierungstechnik, 45, 163174.
Shaaban, S. (2004). Experimental investigation and extended simulation of turbocharger non-adiabatic performance. Ph.D. thesis, Hanover University , Institute of
mechanical engineering.
Sidorow, A., Isermann, R., Cianone, F., and Landsmann,
G. (2011). Model based fault detection of the air and
exhaust path of diesel engine including turbocharger
models. In SAE 2011 World Congress, Detroit.
Stodola, A. (1945). Steam and Gas Turbines. McGrawHill, reprinted by Peter Smith, New York.
Watson, N. and Janota, M. (1982). Turbocharging the
Internal Combustion Engine. London, Basingstoke: The
Macmillan Press.
Zahn, S. (2007). Development of a crank angle based engine model for real-time simulation. In 2nd Conference:
Engine Process Simulation and Supercharging.
Zahn, S. and Isermann, R. (2008). Crank angle synchronious modelling and real-time simulation of diesel engines for ecu function developement and testing. AVEC.

10632