Beruflich Dokumente
Kultur Dokumente
Abstract
For the design, implementation and testing of control systems hardware-in-the-loop (HIL) simulation is increasingly being
required, where some of the control-loop components are real hardware, and some are simulated. Usually, a process is simulated
because it is not available (simultaneous engineering), or because experiments with the real process are too costly or require too much
time. The real-time requirements for such simulations depend on the time-scale of the process and the simulated components involved.
This paper gives, "rst, an overview of the various kinds of real-time and HIL simulation. Then, two cases are considered. First, HIL
simulation for relatively slow processes, like those in basic industries or heating systems, is discussed. Then, the HIL simulation of
combustion engines is shown in detail. The required models are described. Comparisons between real-time simulations and
measurements on real diesel engines and trucks are shown. The goal of the HIL system is to develop new control algorithms and to
investigate the e!ect of faults, both in sensors and actuators, and the engine itself. 1999 Elsevier Science td. All rights reserved.
Keywords: Real-time simulation; Hardware-in-the-loop simulation; Diesel engines; Turbochargers; Vehicle dynamics; Transputers;
Digital signal processors; Simulator testbenches
*Coresponding author. Tel. #49 6151 16 2114; fax: #49 6151 (1) The real process can be operated together with the
29 3445; e-mail: risermann@iat.tu-darmstadt.de simulated control by using hardware other than
0967-0661/99/$ - see front matter 1999 Elsevier Science Ltd. All rights reserved.
PII: S 0 9 6 7 - 0 6 6 1 ( 9 8 ) 0 0 2 0 5 - 6
644 R. Isermann et al. /Control Engineering Practice 7 (1999) 643}653
E Testing of the e!ects of faults and failures of actuators, include multibody systems (Schaefer, 1993; Rieger and
sensors and computers on the overall system; Schiehlen, 1995; Kortuem et al., 1992), brake systems
E Operating and testing of extreme and dangerous oper- (Sailer and Essers, 1996; Sailer, 1996) and combustion
ating conditions; engines (Spindler and Doll 1990; Savaglio, 1993; Woer-
E Reproducible experiments, frequently repeatable; mann, 1994; Kimura, and Maeda, 1996; Isermann et al.,
E Easy operation with di!erent man-machine interfaces 1997).
(e.g., in cockpit-design and the training of operators); Some of these HIL simulations were in-house develop-
E Saving of cost and development time. ments within companies, especially in the "elds of aircraft
and automobiles, and were published only partially or
2.2. Historical development not at all. Typical con"gurations consist of either trans-
puter or DSP cards, a host PC and special interface cards
The "rst approaches to HIL simulation were probably for actuators and sensors. O!-the- shelf software exists
realized for (real-time) -ight simulation, where the early for the basic operating system and some special pro-
goals were to simulate the instruments with a "xed- cesses, sensors and actuators. Commercially available
cockpit &&linktrainer'' (1936), (N.N., 1964; Marienfeld, HIL simulators are, for example, described in (Hansel-
1965), and later on in addition to move a cockpit accord- mann, 1993; ETAS, 1997).
ing to aircraft motions, e.g. for the training of pilots.
Here, the cockpit and the pilot were real, and the motions 2.3. HIL simulation for slow and fast processes
were generated by electrical and hydraulic actuators.
First generations used analog tube controllers and ana- Until now HIL simulation with real controller hard-
log motion simulations, which were subsequently re- ware has mostly been realized for relatively fast processes
placed by analog and then process computers (N.N., like aircraft, automobiles or combustion engines and
1953; Anderson, 1962). their components. The sampling time of the controller
HIL motion simulators were also built for the dynamic is about +1}10 ms (sampling frequency f +
A A
testing of vehicle components (e.g., suspensions, bodies) 100 Hz}1 kHz), and the sampling time for the simulation
with hydraulic or electrical actuators (testing machines). of the mechanical processes +0.5}10 ms.
N
Here, for example, the excitation of wheels by a road The limits on the real-time simulation of these rela-
surface is simulated. Another interesting type of HIL tively fast processes are given by
motion simulation is the vehicle driving simulator (Dros- E the possibility of using parallel algorithms;
dol and Panik, 1985; Suetomi et al., 1991). Dynamic motor E the kind of integration method used;
teststands, where the engines are real and the vehicles and E the sti!ness of the simulated di!erential equation sys-
gears are simulated by some other hardware (an electrical
tems (expressed as a ratio: largest/smallest eigenvalue).
DC or AC motor) together with a digital process com-
puter, is a special kind of HIL simulator (Thun, 1987; HIL simulators can also be used to develop and test real
Pfei!er, 1997). controller hardware and software for slow processes, i.e.
With the development of digital electronic control processes in the power, chemical and basic industries.
systems for vehicles, for example, ABS (antiblock braking The advantages of &&moving the process "eld into the
systems) for brakes, and ASR (TCS: traction control laboratory'' mentioned in Section 2.1 are the same. How-
system) for drive chains and automatic gears, the HIL ever, in order to save observation and development time,
simulators followed through various stages of develop- the simulated process should be run faster than in real
ment (Kempf et al., 1987; Huber et al., 1988). First ver- time, i.e. to simulate one month of operating time in one
sions used high-performance workstations and process day, one hour or even some minutes. Then, the limits are
computers. However, the amount of real-time simulation usually not given by the speed of the process simulation,
was very limited (Klinker, 1992; Wagner and Furry, but by the possible speed-up of the real controller hard-
1992). The availability of parallel computers, in the form ware. Examples for these limits a!ect
of transputers (RISC processors with on-chip RAM and E speed-up of timers, capture}compare units and watch-
high-speed communication links) and digital signal pro-
dog timers;
cessors (DSP: microprocessors with e$cient operation of E speed-up of process interfaces (ADC, DAC);
additions and multiplications by parallel processing with E speed-up of real actuators;
di!erent memories and buses for the program and the E time constants of analog components.
data) then opened the way for real-time simulations of
complete hydraulic systems, sensors, actuators and sus- However, even speed-up factors of 5 or thereabouts may
pension systems, see for instance (Fennel et al. 1992; save considerable time in testing and debugging, com-
Hanselmann, 1993). Further research then showed pared to "eld tests. By such HIL simulation for slow
how more comprehensive mechanical systems can be processes, not only can development time be saved, but
simulated in real time by parallel computers; examples also commissioning time at the plant.
646 R. Isermann et al. /Control Engineering Practice 7 (1999) 643}653
In the following section, the HIL simulation of a very by domain-speci"c software tools (e.g. ADAMS, SPICE).
fast process is shown, the dynamic simulation of Diesel First object-oriented languages for multidisciplinary
engines with real-time simulation of each cylinder work- modeling are Omola (Andersson, 1990) Dymola (Elm-
ing cycle, drive chain and vehicle. qvist, 1978) and Modelica (N.N., 1997).
The second method is experimental modeling or identi-
,cation. Here, input and output signals are measured,
3. Hil simulation of diesel engines with turbocharger and for a selected class of models, the model is adapted to
the process behaviour by minimization of an error
3.1. Process models for real-time simulations measure between the process and the model. For linear
of combustion engines models, identi"cation methods have reached a very ma-
ture state; for instance, methods of Fourier analysis,
For real-time simulation, dynamic models for the sub- correlation, and parameter estimation. For nonlinear
systems comprising the diesel engine, the turbocharger models, di!erent approaches have been formulated. Clas-
and the longitudinal vehicle dynamics are required, see sical methods are mostly based on polynomial approxi-
Fig. 4. Several compromises have to be made with regard mators, e.g. Hammerstein, Wiener or nonlinear di!erence
to the model complexity and the calculation time. For an equation (NDE) models. Neural networks are especially
implementation on a multiprocessor system, a parallel attractive for the general task of nonlinear dynamic sys-
structure of the overall model is aimed at. To ensure the tem identi"cation due to their universal approximation
user's acceptance of the simulator, adaptation to the real capabilities.
system must be possible with reasonable e!ort. There The advantage of system identi"cation is that less time
are two main ways of developing mathematical process is consumed in the development process. However, the
models: quality of the resulting models is highly dependent on the
heoretical modeling is based on physical laws, ex- use of appropriate model structure assumptions, and on
pressed by equations. After simplifying assumptions, the measurement data used. Theoretical models often
these equations are stated for single process elements. have a high complexity because of their physical orienta-
They can be subdivided into balance equations for the tion. However, there is a possibility of model reduction,
storage of mass, energy and momentum, constitutive e.g. by simplifying to input/output models with shorter
equations for sources, transformers and converters, and calculation times.
phenomenological equations for irreversible processes Sometimes the structure of the mathematical model
like dissipative elements or sinks. The interconnections of can be determined in an analytical way, but several of the
the process elements are described by continuity equa- parameters are unknown. In this case, combination with
tions (node law) and compatibility equations (closed cir- a parameter-identi"cation technique, e.g. parameter es-
cuit law). Based on these equations, an overall model can timation, is convenient.
be calculated. For lumped-parameter systems, the pro- For given equations several simulation software tools
cess model can be represented in either state-space form already exist, like MATLAB/SIMULINK, ACSL or
or input/output form, i.e. di!erential equations or trans- MATRIX . For the development of the engine simulator,
fer functions. For distributed-parameter systems, in a combined theoretical}experimental route was taken.
general partial di!erential equations are obtained. The This will be illustrated by presenting the di!erent sub-
development of theoretical models is partially supported models in the following subsection.
Calculation of a complete thermodynamic cycle is not < L
p (h )"p A ,
feasible, due to the calculation time restrictions and the A@?QGA !1 A < (h )
big e!ort required for the adaptation. On the other hand, A !1
h 3[1803, 5403]. (1)
a simple static model based on a simple engine operating !1
map (average torque) is not su$cient for dynamic invest- < (h ) denotes the current cylinder volume, < the
igations. The presented torque model considers a com- !1 X
cylinder volume at the beginning of compression, and
promise between both variants, see Fig. 5. n the polytropic coe$cient. Fig. 6 represents the typical
The approach is based on the idea of decoupling the basic pressure for one working cycle. For comparison,
calculation of the average torque and the dynamic in- the measured cylinder pressure p (h ) for the "red
dicated torque over one working cycle: AGLB !1
operation is shown. The di!erence between the two
First, the average cylinder torque over one work- curves is the part that is used for generating the e!ective
AGLB
ing cycle is determined using a multidimensional map. torque. The pressure at the beginning of compression
The map can be generated either by processing measured p essentially depends on the charge}air pressure, and
data (e.g., on an engine test stand) or by using suitable can be determined by a simpli"ed calculation of the
software tools (in general, the calculation of a thermo- charge cycle (Zapf, 1970).
dynamic cycle (Pischinger, 1989; Zapf, 1970)). Instead of The basic pressure is converted to the basic torque
look-up tables, neural networks can be used. They re- (h ) by using the mechanical laws of the crankgear
A@?QGA !1
quire fewer parameters, and can be adapted if new data (Sinsel et al., 1997; Maass and Klier, 1981). Here, the
become available. resulting gas forces and the oscillating mass forces are
Below the dynamic indicated cylinder torque taken into consideration.
(h ) for one working cycle in dependence on the E+ective torque. The e!ective torque (h ) is
AGLB !1 ACDD !1
crank angle h is determined, based on a given average given by the di!erence between the indicated cylinder
!1
engine torque. It consists of a zero mean basic torque torque (h ) and the zero mean basic torque
AGLB !1
(h ), describing the mass and gas forces, and an (h ):
A@?QGA !1 A@?QGA !1
e+ective torque (h ), describing the gas forces re-
ACDD !1 (h )" (h )! (h ). (2)
sulting from combustion. The mean value of the e!ective ACDD !1 AGLB !1 A@?QGA !1
part is equivalent to the average cylinder torque . This part of the torque essentially results from the com-
AGLB
This approach ensures that the mean value of the dynam- bustion during the expansion cycle. Fig. 7 shows the
ically indicated cylinder torque (h ) is determined subdivision of the indicated torque into basic and e!ec-
AGLB !1
correctly, even if the modeling of the dynamics is only tive torques. The represented indicated torque is based
approximate. The dynamic cylinder torque (h ) at the on measured cylinder pressure data acquired on a truck
A !1
crankshaft is given by the di!erence between (h ) engine (Mercedes-Benz OM 402). (h ) can be ap-
AGLB !1 ACDD !1
and the cylinder friction torque . proximated by the following equation:
ADPGA
Basic torque. The basic torque (h ) is deter- K (h )"ahY e\@FY!1
A@?QGA !1 ACDD !1 !1
mined using a simpli"ed combustion chamber model.
First, the basic cylinder pressure p (h ) is calculated, with h "h !3603. (3)
A@?QGA !1 !1 !1
which is an approximation of the cylinder pressure for
the non-"red engine. During the compression and com-
bustion cycle, p (h ) follows a polytropic change of
A@?QGA !1
state. During the intake and exhaust cycle, p (h ) is
A@?QGA !1
equivalent to the pressure at the beginning of compres-
sion p :
A
p (h )"p ,
A@?QGA !1 A
h 3[03, 1803]R[5403, 7203],
!1
Fig. 7. Cylinder torque, subdivided into basic and e!ective torque. Fig. 8. E!ective cylinder torque: simulated, and derived from measure-
ments.
1 3
K (h ) dh " . (4)
7203 ACDD !1 !1 AGLB
2. The maximum of the approximating function is at
crank angle h "h . While the crank angle for
!1 !1K?V
maximum pressure is in#uenced by the engine load,
engine speed and start of injection, the crank angle for
maximum torque is nearly independent of these
quantities. This fact is due to the mechanics of the Fig. 9. Indicated cylinder torque: simulated, and derived from measure-
crankgear. Thus h can be assumed as a constant. ments.
!1K?V
d
K " (5)
dh ACDD F!1F!1K?V"0.
!1 3.3. Turbocharger model
Using Eqs. (4) and (5), the parameters a und b can be
obtained: Fig. 10 schematically represents the exhaust turbo-
charger with waste gate. The charging process has a non-
4 7203 2 linear static input/output behavior, as well as a strong
a" AGLB b" . (6)
(h ) h dependency at the dynamics at the operating point.
!1K?V !1K?V In general, the static behavior of the turbocharger may
Using Eqs. (3) and (6), (h ) can be calculated, de- be su$ciently described by characteristic maps (look-up
ACDD !1
pending on . Fig. 8 shows the e!ective torque tables) of the compressor and turbine. However, if the
AGLB
simulated and derived from measurements of the cylinder dynamics of the turbocharger need to be considered,
pressure. Using this approach, the dynamics of the over- basic mechanical and thermodynamical modeling is re-
all cylinder torque (h ) can be reproduced with quired, see (Zinner, 1985; Pucher, 1985).
AGLB !1
su$cient accuracy. In Fig. 9, the simulated and recon- Practical applications have shown that these methods
structed cylinder torque for one working cycle is depic- are capable of reproducing the characteristic dynamic
ted. This good agreement can also be obtained at other behavior of the turbocharger. The model quality, how-
speed/torque operating points. Only at low load do small ever, essentially depends on accurate knowledge of sev-
deviations occur, due to the simpli"ed model assump- eral process parameters, which have to be laboriously
tions mentioned above. derived or estimated, in most cases by analogy. A further
R. Isermann et al. /Control Engineering Practice 7 (1999) 643}653 649
Mercedes-Benz truck engine series 500 and 900 is pre- the information received from other units of the IES. The
sented. First, a short survey of the engine management is data is transmitted to the pump-line-nozzle control unit
given. (PLD) via the CAN-bus. The PLD control unit contains
Electronic engine management. The electronic engine the engine control functions, especially the precise con-
management is part of the Integrated Electronic System trol of the magnetic injection valves. For that, the opti-
(IES), and consists of two control units connected by mal beginning and duration (pulse width) of injection for
a CAN-bus interface, Fig. 13. each cylinder, depending on the reference values ob-
The vehicle-engine-control unit (FMR) contains all the tained from the FMR control unit, are calculated. By
functions concerning the vehicle, e.g.: means of multidimensional maps and di!erent sensor
signals (sensors for engine speed, charge-air pressure, and
E detection of the accelerator-pedal position;
temperatures of fuel, coolant and charge air), an optimal
E engine brake control;
operating condition for fuel consumption, emissions and
E cruise control;
driving comfort is achieved.
E reduction of the engine torque in conjunction with
Setup of the simulator test bench. Fig. 14 shows the
anti-slip control;
setup of the simulator test bench for the investigation of
E output of a reference value for the engine speed, for
the engine control system as mentioned above. It may be
speed control of accessories.
subdivided into the following parts:
The FMR control unit calculates the reference values for
E real-time computer system, including I/O-modules;
the engine torque, brake torque and engine speed, de-
E interface, consisting of the sensor and actuator inter-
pending on the driver's commands, the sensor signals and
face;
E PLD control unit, including real actuator components,
either stand-alone or in combination with the real
FMR;
E PC with graphical user interface;
E control panel.
Real-time computer system (<ersions 1 and 2). To ful"ll
the real-time constraints necessary for an HIL simula-
tion, powerful computer hardware is required. For the
"rst version of the simulator test bench, a transputer
system with 11 processors (T805) was developed. Four
transputers were used to calculate the simulation models,
and seven for I/O communication. The models for the
di!erent subsystems are calculated in parallel to increase
the computing power. The real-time computer system for
the second simulator version is based on a dSPACE-
Fig. 12. Vehicle model. system equipped with digital signal processors and
Fig. 13. Engine control system of the new Mercedes-Benz truck engine series 500 and 900.
R. Isermann et al. /Control Engineering Practice 7 (1999) 643}653 651
Fig. 14. Setup of the simulator test bench with simulated engine and vehicle, and real engine control unit and injection actuators.
a DEC Alpha processor. This system has the advantage real valve opening time, and to determine the pulse width
of higher computing power, which makes a parallel cal- and the beginning of injection. These quantities are trans-
culation unnecessary. It also o!ers the possibility to ferred to the real-time computer system for engine simu-
realize all the models in MATLAB/SIMULINK, to use lation. By this way, the real behaviour of the injection
all the bene"ts of a graphical simulation environment. pumps is included.
Special I/O-modules (Digital-I/O-module, D/A conver- PD control unit and FMR control unit as test
ter, A/D converter and CAN-interface) are used for the piece. The simulator test bench was set up with the
communication with the interface. objective of testing the PLD control unit, either stand
Interface. The coupling of the simulator and the control alone or in combination with the FMR control unit. In
unit is implemented by means of a special interface, which the "rst case, the necessary FMR functions are simulated
can be subdivided into a sensor and an actuator interface. by the computer system. The data transfer is done via the
The sensor interface generates the necessary sensor signals, engine CAN-bus. In the second case, the PLD and the
like temperatures and pressures. The pulses of the cam- FMR control units are connected directly via the engine
shaft and crankshaft inductive speed sensors are generated CAN-bus. The computer system emulates other IES
by a board speci"cally designed for high-speed signal units in this operation mode by transmitting the data via
generation. For that purpose, a look-up table with the the IES-CAN-bus to the FMR.
pulse signals versus the crankshaft angle is stored o!-line PC with graphical user interface. For e$cient use of
in memory. During the simulation, the signals are period- the simulator, a comfortable experimental environment
ically read out, synchronously with the simulated engine is needed. The simulator operation is performed with
speed. This realization guarantees high #exibility in form- a Windows user interface on the host PC which copies
ing the pulses and adapting di!erent gear wheels. The the functionality of a real truck-cockpit. All the relevant
sensor interface also contains relay electronics to simulate simulation quantities can either be visualized on-line, or
sensor faults like interruptions and short-circuits. recorded for o!-line analysis. To ensure reproducible
The actuator interface consists mainly of the injection results, a driver simulation is implemented, which can
pumps (pump-line-nozzle injection system), which are automatically follow a given speed cycle by operating the
integrated into the simulation test bench as real compo- accelerator, brake, clutch and gears.
nents, because the combination of the PLD control unit Control panel. As an alternative, interactive &&driving''
and the injection pumps represents a mechatronic unit of the simulator can be performed manually, using a con-
that is di$cult to model. A special electronic device trol panel where the most important cockpit functional-
measures the magnetic valve currents to reconstruct the ities are realized.
652 R. Isermann et al. /Control Engineering Practice 7 (1999) 643}653
4. Conclusion
References Nelles, O., Sinsel, S., & Isermann, R. (1996). Local basis function
networks for identi"cation of a turbo charger. Proc. ;KACC Int.
Anderson, O.K. (1962). Design tool (Research and development) simu- Conf. on Control +96, Exeter, UK, 2}5 Sept. 1996 (Vol. 1, pp. 7}12).
lator survey report. SAE Nat. Aerospace Engng. Manufacturing N.N. (1953). The redifon comet #ight simulator. De haveland gazette,
August 1953, 139}141.
Meeting, Los Angeles, Oct. 1962.
Andersson, M. (1990). Omola } An object-oriented language for model N.N. (1964). Schulung und praxis mit #ugsimulator. INERA<IA,
representation. Lic Tech thesis TFRT-3208, Department of Auto- 19(8), 1104}1107.
N.N. (1997). Modelica, a uni,ed object-oriented language for physical
matic Control, Lund Institute of Technology, Lund, Sweden.
Drosdol, J., & Panik, F. (1985). The Daimler-Benz driving simulator. systems modeling. Modelica homepage: http://www.Dynasim.se/
SAE 850334. Modelica/
Pfei!er, K. (1997). Fahrsimulation eines Kraftfahrzeugs mit einem dyna-
Elmqvist, H. (1978). A structured model language for large continuous
systems. Ph.D. dissertation. Report CODEN: ;FD2/(FR- mischen Motorpru( fstand. Dissertation TU-Darmstadt. Fortschr.-
1015). Department of Automatic Control, Lund Institute of Tech- Ber. VDI Reihe 12 Nr. 336. VDI-Verlag, DuK sseldorf, Germany.
Pischinger, R. (1989). hermodynamik der <erbrennungskraftmaschine.
nology, Lund, Sweden.
ETAS (1997). abCar. Hardware-in-the-loop testsysteme fu( r steuerger- Berlin, Germany: Springer.
a( te. ETAS, Schwieberdingen, Germany. Pucher, H. (1985). Au-adung von <erbrennungsmotoren. Sindel"ngen,
Germany: Expert-Verlag.
Fennel, H., Mahr, S., & Schleysing, R. (1992). ransputer-based real-
time simulator } a high performance tool for ABS and CS develop- Rieger, K., & Schiehlen, W. (1995). Echtzeitsimulation eines Fahrzeug-
ment. SAE 920643. modells mit aktiver Federung. <DI-Bericht Nr. 1189 (pp. 17}34).
DuK sseldorf, Germany: VDI-Verlag.
Germann, S. (1996). Modellbildung und modellgestu( tzte Regelung der
Fahrzeugla( ngsdynamik. Dissertation TU Darmstadt. Fortschr.-Ber. Sailer, U. (1996). Nutzfahrzeug-Echtzeitsimulation auf Parallelrechnern
VDI Reihe 12 Nr. 309. DuK sseldorf, Germany: VDI-Verlag. mit Hardware-in-the-oop. Renningen-Malmsheim, Germany:
Expert-Verlag.
Hanselmann, H. (1993). Hardware-in-the-loop simulation as a standard
approach for development, customization, and production test of Sailer, U., & Essers, U. (1996). Parallelverarbeitung als Weg zur Ech-
EC;1s. SAE 931953. tzeitsimulation in der Kfz-SteuergeraK teentwicklung. <DI-Bericht
Nr. 1283 (pp. 99}116). DuK sseldorf, Germany: VDI-Verlag.
Huber, W. et al. (1988). Simulation, performance and quality evaluation of
ABS and ASR. SAE 880323. Savaglio, C. (1993). Hardware-in-the-loop simulation } an engine
Isermann, R., Germann, S., WuK rtenberger, M., Halfmann, C., & Holz- controller implementation. SAE 930204.
Schaefer, P. (1993). Echtzeitsimulation aktiver MehrkoK rpersysteme auf
mann, H. (1996). Model-based control and supervision of vehicle
dynamics. Fisita Congress, Prag, Czech Republic, 17}21 June 1996. Transputernetzen. Fortschr.-Ber. VDI Reihe 11 Nr. 202. DuK sseldorf,
Isermann, R. (1997). Mechatronic systems } A challenge for control Germany: VDI-Verlag.
Sinsel, S., Scha!nit, J., & Isermann, R. (1997). Hardware-in-the-loop
engineering. Proc. 1997 American Control Conf., Albuquerque, USA,
4}6 June 1997. (Vol. 5, pp. 2617}2632). simulation von dieselmotoren fuK r die Entwicklung moderner
Isermann, R., Sinsel, S., & Scha!nit, J. (1997). Hardware-in-the-loop motormanagementsysteme. Moers, Germany: <DI-agung.
Spindler, W., & Doll, M. (1990). Ermittlung des stationaK ren und des
simulation of diesel-engines for the development of engine control
systems. Proc. 4th IFAC-=orkshop on Algorithms and Architecture instationaK ren Betriebsverhaltens von kleinen schnellaufenden Die-
for Real-ime Control, AARC+97, Vilamoura, Portugal, 9}11 selmotoren. Forschungsberichte <erbrennungskraftmaschinen: Heft
454, FVV, Gesamtproze{analyse-Thermodynamik PKW-Diesel-
April 1997 (pp. 90}92).
Kempf, D., Bonderson, L.S., & Slafer, L.I. (1987). Real time simulation motoren, Vorhaben 384, Abschlu{bericht.
for application to ABS development. SAE 870336. Suetomi, T. et al. (1991). he driving simulator with large amplitude
motion system. SAE 910113.
Kimura, A., & Meda, I. (1996). Development of engine control system
using real time simulator. Proc. IEEE Int. Symp. Comput. Aid. Thun, H.-J.v. (1987). Dynamischer VerbrennungsmotorenpruK fstand mit
Contr. Systems Design, Dearborn, 15}18 Sept. 1996 (pp. 157}163). Echtzeitsimulation des Kraftfahrzeug-Antriebsstranges. Automobil-
technische Zeitschrift } AZ, 89(1).
Klinker, W. (1992). Hardware-in-the-loop. S;N News, (pp. 21}23).
October 1992. Wagner, J.R., & Furry, J.S. (1992). A real time simulation environment
Kortuem, W., Sharp, R.S., & Departer, A.D. (1992). Review of multibody for the veri"cation of automotive electronic controller software.
Int. J. <ehicle Des., 13(4), 365}377.
computer codes for vehicle system dynamics: Problem requirement
codes and experiences with benchmark problems. Carl Cranz Gesel- Woermann, R. (1994). Ein beitrag zur echtzeitsimulation technischer
lschaft, Oberpfa!enhofen, Germany. systeme hoher dynamik mit diskreten modellen. Dissertation Univer-
sitaK t Gesamthochschule Kassel.
Maass, H., & Klier, H. (1981). Kra( fte, Momente und deren Ausgleich in
der <erbrennungskraftmaschine. Berlin, Germany: Springer. Zapf, H. (1970). Untersuchungen zur vorausberech-nung der ladun-
Marienfeld, H. (1965). Flugsimulation. Dissertation TU Darmstadt. gsendtemperatur in viertakt-die-selmotoren. Motortechnische Zeit-
schrift } MZ, 12.
Nelles, O., & Isermann, R. (1996). Basis function networks for interpola-
tion of local linear models. Proc. 35th IEEE Conf. on Decision and Zinner, K.A. (1985). Au-adung von verbrennungsmotoren. Berlin,
Control. Kobe, Japan, Dec. 1996, (Vol. 1, pp. 470}475). Germany: Springer.