A components library for simulation and analysis of aircraft

electrical power systems using Modelica

Martin R. Kuhn1 , Antonio Griffo 2 , Jiabin Wang 2 , and Johann Bals1

1German Aerospace Center (DLR), Münchner Strae 20, 82234 Wessling, Germany, Phone:
+49 (8153) 28-2461, Fax: +49 (8153) 28-1441, Email: forname.surname@dlr.de, URL:
http://www.robotic.de
2 Department of Electronic and Electrical Engineering , The University of Sheffield , Mappin

Street, S1 4EX, Sheffield UK , Phone: +44 (0)114 2225817 , Email: a.griffo@sheffield.ac.uk,

j.b.wang@sheffield.ac.uk

June 12, 2009

Acknowledgments
This research is being conducted in the frame of the MOET project (More-Open Electrical Technologies),
a FP6 European Integrated Project. See www.moetproject.eu for details.

Keywords
<<Aerospace>>, <<Modelling>>, <<Simulation>>

Abstract
A library of components for simulation and analysis of large vehicular electrical power systems using
Modelica language is presented. Components are described using different levels of model complexity,
catering for both detailed high fidelity transient switching dynamics and averaged value descriptions
which, being time invariant, are a computationally efficient and useful tool for design, stability and
sensitivity analyses. The merits of employing the Modelica based modelling tool are discussed, and its
utilities and effectiveness are demonstrated through a test system consisting of a three-phase, variable
frequency synchronous generator which feeds high voltage DC loads via an auto-transformer rectifier
unit.

Introduction
Future ”more electric” terrestrial, marine and aerospace vehicular electrical power systems will be based
on the interconnection of a wide range of components, resulting in a significantly more complex elec-
trical distribution system with multiple distributed loads most of which are supplied and controlled by
power electronic converters [1]. Due to the destabilizing effect of tightly regulated loads which result in
equivalent negative impedance behavior, these electrical networks are susceptible to instabilities [2],[3].
In order to investigate potential instability causes and develop corrective actions detailed investigations
by means of numerical simulations are required. Time-domain simulation using detailed nonlinear, time
varying power system models (which will be referred to as behavioral models) can be employed for
accurate transient performance evaluation, stability assessment and power quality investigation. Sev-
eral commercial and non-commercial software tools are available for detailed behavioral simulations
[4],[5],[6],[7]. Despite the highest level of model accuracy, the main drawback of behavioral modeling
is the need for vast computational resources and time. Since switching transients from power electronic
devices do not normally have a significant influence on systems stability, state-space averaging tech-
niques have been developed to derive equivalent non-linear time-invariant system models (which will be
referred to as functional models) [8]. Due to the unavailability of validated averaged models suitable for

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on February 1, 2010 at 11:08 from IEEE Xplore. Restrictions apply.
 stability assessment and the lack of commercial libraries for simulation and analysis of functional mod-
els, a dedicated library has been developed using the Modelica language [9]. It is demonstrated that the
proposed library, coupled with the commercial software Dymola [10] can be effectively employed as a
tool for modeling and simulation of complex power systems, both at a functional and behavioral level. It
is also demonstrated how important simulation steps such as system’s initialization, linearization around
an operating point, formulation of the Jacobian matrix and parametric analysis can be automatically car-
ried out in order to provide a reliable and computationally efficient tool for systems analysis. The paper
describes in detail the use and the advantages of Modelica language for power systems modeling, as
well as provides details on the developed library of components. Detailed descriptions of the two-level
modeling of the key power systems components will be given in subsequent sections, as well as thorough
simulations of a test system containing the modeled ATRU fed by a three-phase synchronous generator,
with transients initiated by changing in load conditions. Numerical simulations are reported to clearly
show the capabilities of the proposed framework in terms of behavioral and functional systems model-
ing. The potentiality of the proposed modelling framework in stability analysis will also be described
and numerical results of small-signal analysis of a test system, validated with time-domain simulations
will also be presented.

Modelica for dynamical systems modeling

Modelling and simulation of electrical, mechanical, electro-mechanical and other physical systems usu-
ally consists of writing a system of algebraic-differential equations in the form:
ẋ = f(x, y, u) (1)
0 = g(x, y) (2)
which describe the evolution in time of state-space variables x of the system with inputs u. Algebraic
relationships describe the instantaneous memoryless dynamics of variables y which are implicitly defined
as functions of the state variables x by (2) which usually result from the mathematical description of the
interconnection of different subsystems.
Small signal stability analysis requires the construction of a state-space model in the differential form:

ẋ = f(x, h(x), u) (3)

where (2) has been explicitly solved to yield y = h(x), and the subsequent linearization of (3) around
a steady-state equilibrium point corresponding to a specified operating condition x0 determined by the
inputs u0 , to yield:

Δx
˙ = A · Δx + B · u (4)
with the equilibrium point obtained as the solution of:

0 = f(x0 , u0 ). (5)

Jacobian matrix A and B are obtained as:

∂f ∂f
A= ,B = (6)
∂x |x0 ,u0 ∂u |x0 ,u0
Solution of (2) for y and calculation of steady-state conditions require algebraic manipulations which are
often extremely difficult to be analytically performed for all but very simple systems, and therefore are
usually tackled by iterative numerical calculations at simulation stage which usually slows down compu-
tation by introducing fictious algebraic loops. Furthermore analytical calculation of linearized equations
can also be difficult and some simulation software such as Simulink in these cases performs linearization
by resorting to numerical perturbation of the system’s equations about the specified equilibrium point.
The Modelica language based simulation program Dymola has been selected for the building of the pro-
posed library and simulation environment thanks to its capability to symbolically pre-process a Modelica
model and to transform implicit equation systems to state space form, where possible, allowing for effi-
cient simulation. The transformed equations are provided as C-Code that is compiled and embedded in
Dymola’s own simulation engine, as stand-alone programme or in other simulation environments (e.g.
in Simulink or Matlab). Parameter variations, initialization, pre and post-processing of data can be au-
tomatically carried out. Those operations can be performed directly in Dymola and also controlled via a
Matlab interface which is part of the distribution. Most of the calculations are carried out in analytical
symbolic form, therefore simplifying design and avoiding loss of precision and increase in computation
burden which would result from iterative numerical procedure.

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on February 1, 2010 at 11:08 from IEEE Xplore. Restrictions apply.
 Figure 1: Schematic of a simplified electric power system

Modelica is an object oriented multi-domain modeling language for component-oriented modeling of

complex systems. In contrast to data flow-oriented languages with directed inputs and outputs, such
as the widely known Matlab and its associated Simulink tool, Modelica employs an equation based
modeling, which results in a faster modeling process and a significantly increased reusability of the
models. The language is developed under open source license since 1996 by the non-profit Modelica
Association, based at Linköping University, Sweden. Many free Modelica libraries for different physical
domains modeling are available. The Modelica standard library contains a large collection of compo-
nents to model among others analogue and digital electronics, electrical machines, 1-dim. translational
and rotational mechanical systems, as well as input/output control blocks. Modelica has also been used
for simulation of electric power systems and power electronics components [11],[12]. Free and com-
mercial simulation environments acting as Modelica front-end are available with useful functionalities
to graphically construct and simulate Modelica models and perform post-processing operations. Besides
the potentialities of its computational algorithms the Dymola simulation environment has also useful
capabilities in modeling and handling large libraries of components.

Power system modeling

A simplified schematic for a representative power system for ’more electric’ aircraft is shown in Fig. 1.
A three-phase synchronous generator provides power to the AC bus whose voltage is regutated by an
automatic voltage regulator (AVR) that provides the field excitation to the generator. A direct symmetric
eighteen-pulse autotransformer and rectifier unit (ATRU) is used to rectify AC voltage and power the DC
loads. AC and DC capacitors may also be present for power quality and stability improvement.
Signal-flow simulation programs such as Simulink requires the user to explicitly describe the intercon-
nections between different components in a system. For an electrical system these relationships may be
given by voltage/current balance equations. Each component is usually represented with a block having
a predefined set of inputs and outputs. Causality dictates that the outputs at each time instant are a func-
tion of the inputs at that instant and the internal states. Figure 2 shows the block diagram that describes
a signal-flow modelling of the power system of Fig. 1, while Fig. 3 shows the block diagram for the
description of the full order synchronous generator model. Thick arrows in Fig. 2 are used to denote
AC values which can be given by either three-phase abc or two-phase dq quantities, depending on the
modelling and the level of detail the user is interested in.

Figure 2: Block diagram for signal flow power system simulation

Although a signal-flow based modelling is appealing for its simplicity, the need to explicitly define
interface equations for the interconnection of different components, which can in some cases be given
by complex algebraic or differential/algebraic equations, while taking into account the causality of each
block can make the modelling and simulation of complex systems with a large number of interconnected

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on February 1, 2010 at 11:08 from IEEE Xplore. Restrictions apply.
 Figure 3: Synchronous generator block diagram

components awkward. Furthermore it is not possible to integrate different levels of modelling details
such as functional and behavioural models in the same simulation environment.
As an example of the powerful modeling capabilities of Modelica, an extract of the code for the modeling
of a synchronous generator is presented. This simple description is compared with the complex modeling
that would result in a traditional signal-flow causal based modeling environment such as Simulink, as
reported in the block diagram in Fig. 2 [13].
• Voltage equations

u_qs = Rs*i_qs + we*psi-ds + der(psi_qs);

u_ds = Rs*i_ds - we*psi_qs + der(psi_ds);
-Rkq*i_kq = der(psi_kq);
-Rkd*i_kd = der(psi_kd);
u_fd = Rfd*i_fd + der(psi_fd);

• Flux linkage equations

Psi_qs = L_ls*i_qs + K_sat*L_mq*(i_qs+i_kq);

psi_ds = L_ls*i_ds + K_sat*L_md*(i_ds+i_fd+i_kd);
psi_kq = L_lkq*i_kq + K_sat*L_mq*(i_qs+i_kq);
psi_kd = L_lkd*i_kd + K_sat*L_md*(i_ds+i_kd+i_fd);
psi_fd = L_lfd*i_fd + K_sat*L_md*(i_ds+i_kd+i_fd);

• Magnetizing currents

i_md = (i_fd + i_ds + i_kd)/k_fd;

i_mq = (i_qs + i_kq)/k_fd;

• Saturation equations

i_m = sqrt(i_mdˆ2 + i_mqˆ2);

Lm_sat = sqrt(2)*v_sat/(w_e*i_m);
K_sat = Lm_sat/Ls_fd;
SaturationTable.input = i_m;
SaturationTable.output = v_sat;

• Mechanical equations

Te = 3/2*Pp*(psi_ds*i_qs-psi_qs*i_ds);
der(w_m) = Te - visc;
w_e = der(theta);

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on February 1, 2010 at 11:08 from IEEE Xplore. Restrictions apply.
 • abc ↔ dq transformation

[u_qs; u_ds] = 2/3*[ cos(theta) cos(theta-2/3*pi) cos(theta+2/3*pi);

-sin(theta) -sin(theta-2/3*pi) -sin(theta+2/3*pi)]*[u_a; u_b; u_c];

Simbols and variables have an intuitive meaning, details are reported in [14]. The Modelica command
der define the derivative operator. It is clear that there is no need to specifically define an input/output
relationship for each block.
Differently to any causal-based simulators, algebraic loops are automatically solved, in Modelica based
simulations, at compilation stage by employing powerful symbolic manipulations, therefore no iterative
procedures are required resulting in faster simulations. Symbolic manipulation also allows Modelica
based simulators to automatically generate linearised equations and Jacobian matrix of the system which
would have to be derived manually or numerically with signal-flow based modeling.

Multilevel library concept

The proposed library is based upon a multilevel modelling concept and a base ”netcore” library devel-
oped previously by DLR under the European commission funded project VIVACE [15] which addressed
the need for a unitary framework for the integration and management of different suppliers’ models. The
motivation behind embedding multi-level models in one model is that a system architecture only has to
be built once and can be simulated at different levels of accuracy and at different speed depending on the
purpose of the study and results can be compared and validated between different levels of modelling.
In the library, each component model is a container for two sub-models which represent different model-
ing accuracies of the same system. The so-called ”behavioral” models are close to the hardware level and
are suited for detailed simulations as needed for power quality studies, with switching transients of power
electronics components fully captured. ”Functional” models are non-switching but possibly nonlinear
models obtained as a result of averaging techniques applied to switching transients. Functional models
can be used for linearization and for stability investigations applying classical methods of control theory
for linear time invariant systems. The frequency range in which functional models are representative
of the average dynamics is much lower than the switching frequency of power electronics devices. The
criteria for validating functional models are [16]:,
• the trajectories have to represent the moving averages along the trajectories of the original system,
• the stability properties of the original and approximate systems have to be identical,
• the models have to be valid for large signals,
• the open-loop models must be usable for closed-loop design.
The presence of both (three-phase) AC and DC subsystems in a typical vehicular power network requires
the use of a time invariant reference frame. For three-phase AC waveforms in the so called ”abc system”,
the DC component of the generalized multi-frequency averaging results in the classical description in
terms of a synchronously rotating reference frame, which decomposes periodic three-phase values into
their associate DQ0-axis stationary components [17].
The zero component is not included in this library since it is mainly relevant for asymmetric loads not
treated by functional models.
A multi-level model for the generator as shown in Figure 4 (a), is built by embedding functional (c)
and behavioral (d) models into a single container model (b), which can in turn be connected to other
multi-level models by an appropriate connection plug. A flag acting as a global variable can be set at
compilation stage enabling the translation of either the behavioral or the functional modeling level (object
in upper right corner in (a)). The equation system and simulation code only contains the selected levels.
In order to satisfy the need to interconnect models which might be represented either in three-phase
abc system (c) or in DQ reference (d), the Modelica concept of ”expandable connectors” is employed.
This data bus concept allows level-dependent models to be interconnected, while enabling only data
related to the selected level to be present after compilation. It contains either the abc current and voltage
variables or the DQ components plus the generator angular frequency. New multi-level models can be
implemented conveniently by using the infrastructure of the implemented ”netcore” library with base
classes, partial models, interfaces and models.

Components modelling
The presented components library was developed in context with the EC funded project ”More Open
Electrical Technologies” (MOET). Among others, several generators, (autotransformer-) rectifiers and
motor drives were embedded. The modeling framework is illustrated with respect to two essential

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on February 1, 2010 at 11:08 from IEEE Xplore. Restrictions apply.
 selects global level global

BitStream triggeredTransition
options
defaultLevel=level 3

a m=3

resistor
m=3

variableResistor
level=global option
issue5_1

functional

star
m=3
angw_dc

b
ground

flange1
behavioural

neutral
variable lines

c d dq lines
m=3 m=3 m=3 rl_qd_th...

ynode_...

star r l

~ ~

flange flange

friction friction

SMEE SMEE

abc lines neutral neutral

Figure 4: Example of a multi-level model

components of the power distribution network of a proposed aircraft power system: a three-phase syn-
chronous generator and an 18-pulse Auto Transformer Rectifier Unit (ATRU). The generator component
model is shown in figure 4. Both generators submodels share the same systems equations in the dq sys-
tem. The most essential difference between the two levels of the model is the use of the dq components
in the functional models connector ports while the behavioral model has the additional transformation
step to the abc system in the output. The generator’s output voltage is controlled with an AVR which
is similarly modelled using the two-level concept and employs a PI controller for output regulation
(with/without anti-windup limiters).
For the voltage feedback, the AC voltage magnitude of the behavioural model is obtained as:

VAC−beh = Vα2 +Vβ2 (7)

where:
⎤ ⎡
   Va 
Vα 2/3 −1/3 −1/3
= · ⎣ Vb ⎦ (8)
Vβ 0 1 −1
Vc

whereas for the functional model VAC− f unc = v2d + v2q .
Figure 5 shows the library structure of the generator. The package ”generator” contains the multi-level
models of the generator, here in different versions to include different modelling details. This model
can be used to build a system and the level can be selected globally or locally as a parameter. The
functional and behavioral sub-models and other components are placed in the components folder. In
order to validate the generator modelling, test systems are also provided in the library.
The ATRU model in figure 6 shows the model icon with the two embedded submodels. Both submod-
els use the same DC filters on the DC side. Only one of the level dependent models is translated at
compilation time depending upon the level set by the user.
On the AC side of each submodel the AC connectors are present. To discriminate between the two dif-
ferent modelling levels a yellow square connector carrying dq voltages and currents and angular velocity
informations is used for interconnecting ”functional” components. A blue dot represents a three-phase
connector carrying information on three-phase voltages and currents for ”behavioural” models. Both
connectors are connected to the green circle representing the expandable plug which allows the coexis-
tence of the two different modelling levels in the same block.
Several topologies of ATRU have been analyzed and included in the library. The direct-symmetric topol-
ogy without interphase reactors is presented here. The ATRU behavioural model contains a 9 phase

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on February 1, 2010 at 11:08 from IEEE Xplore. Restrictions apply.
 different issues
of the multi-level
sublibrary "generator" model

functional and behavioural

model and components

unitary testing
environment
for the generator

Figure 5: Library structure of the generator

p
~
a...
C=45e-6

functional model
C19
R1

R2
R=1

R=1

n2
C=130e-6

C=130e-6

C=45e-6
C27

C28

C18

behavioural model N

case

Figure 6: Multi level model of the ATRU

autotransformer unit, three six diodes bridge rectifier and some filtering components. The functional
models has been derived applying state-space averaging techniques. Detailed analytical derivations and
thourough validation of the derived models are reported in [18].

Numerical results
The modeling approach has been validated by time domain simulations. A test system representative
of the system in Fig. 1 has been assembled using the components in the library. It consists of the
synchronous generator and its associated AVR connected to the ATRU which supplies resistive loads at
the DC output. The simulation results with stepwise increasing load are reported in Fig. 7. In particular
the DC voltage and the generator’s d and q axis currents are shown. It is demonstrated how the functional
models trajectories clearly capture the moving averages along the trajectories of the detailed behavioural
model.

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on February 1, 2010 at 11:08 from IEEE Xplore. Restrictions apply.
 (a) generator, q-axis current (b) generator, d-axis current

(c) DC output voltage

Figure 7: Compared simulation results for generator/ATRU source

Stability analysis
The developed modelling environment can be easily used for network stability studies by analyzing the
influence of design parameters and operating conditions on system’s small-signal stability. As an ex-
ample, small-signal stability of the test system previously described has been analyzed, when a constant
power load (CPL) is connected to the DC output of the ATRU via a 200μF, 20μH LC passive filter.
The load power demand PCPL is increased gradually from 110 to 150 kW. For each loading condition
the system’s equilibrium and linerization are automatically calculated using Dymola algorithms. The
resulting eigenvalue locus is shown in Fig. 8. It is clear that the system is unstable for PCPL ≥ 130kW.
In order to confirm this result, time domain simulations using functional models are reported in Fig. 9
which shows the transients in DC voltage and current following a step change of 1 kW in load power
demand at 0.4s. Two conditions are reported: PCPL =128kW and PCPL =130kW. The unstable transients
with growing amplitude oscillations in the latter case confirm the results of small-signal analysis. More
complex systems with several loads can easily be analyzed. Detailed results are reported in [14].

Conclusion
A library of components for modelling, simulation and stability analysis of complex vehicular power
systems has been described. The library, built using the modelling language Modelica, is able to handle
different levels of modelling complexity and details, employing in a unified framework both detailed
behavioural models and functional averaged models. The benefits of describing a complex system with
Modelica in terms of modelling simplicity and capability of automatically performing procedures such as
initialization and linearization, have been highlighted. Numerical results have been reported to demon-
strate the modelling approach and its potential for stability analysis of complex vehicular power systems.

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on February 1, 2010 at 11:08 from IEEE Xplore. Restrictions apply.
 Figure 8: Eigenvalues locus as a function of constant power load demand

Figure 9: DC voltage and currents transients for load power PCPL =128kW (a), and PCPL =130kW (b).

References
[1] A. Emadi, M. Ehsani, and J.M. Mille. Vehicular Electric Power Systems: Land, Sea, Air, and Space
Vehicles. CRC Press, 2003.
[2] A. Emadi, A. Khaligh, C.H. Rivetta, and G.A. Williamson. Constant power loads and negative
impedance instability in automotive systems: definition, modeling, stability, and control of power

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on February 1, 2010 at 11:08 from IEEE Xplore. Restrictions apply.
 electronic converters and motor drives. IEEE Trans. on Vehicular Technology, 55(4):1112–1125,
July 2006.
[3] L. Han, J. Wang, A. Griffo, and D. Howe. Stability assessment of ac hybrid power systems for
more electric aircraft. In Proc. IEEE Vehicle Power and Propulsion Conference VPPC ’08, pages
1–6, 3–5 Sept. 2008.
[4] Synopsys Inc. Saber. www.synopsys.com, 2009.
[5] The MathWorks. Simpowersystems. www.mathworks.com/products/simpower, 2009.
[6] Plexim. Plecs. www.plexim.com, 2009.
[7] Alternative Transient Program. Atp. www.emtp.org, 2009.
[8] M. R. Kuhn, M. Otter, and L. Raulin. A multi level approach for aircraft electrical systems design.
In Proceedings of the 6th International Modelica Conference, 2008.
[9] Modelica association. Modelica. www.modelica.org, 2009.
[10] Dynasim AB. Dymola. www.dynasim.se, 2009.
[11] M. Larsson. Objectstab-an educational tool for power system stability studies. IEEE Trans. Power
Systems, 19(1):56–63, Feb. 2004.
[12] H. Elmqvist, F.E. Cellier, and M. Otter. Object-oriented modeling of power- electronic circuits
using dymola. In First Joint Conference of International Simulation Societies, 1994.
[13] Chee-Mun Ong. Dynamic Simulation of Electric Machinery. Prentice Hall, 1998.
[14] A. Griffo and J. Wang. Stability assessment of electrical power systems for ’more electric’ aircraft.
In 13th European Conference on Power Electronics and Applications, 2009.
[15] VIVACE Project. Value improvement through a virtual aeronautical collaborative enterprise.
www.vivaceproject.com, 2007.
[16] P.T. Krein, J. Bentsman, R.M. Bass, and B.L. Lesieutre. On the use of averaging for the analysis of
power electronic systems. 5(2):182–190, April 1990.
[17] P.C. Krause, O. Wasynczuk, and S.D. Sudhoff. Analysis of Electric Machinery and Drive Systems.
WileyBlackwell, 2002.
[18] A. Griffo and J. Wang. State-space average modelling of synchronous generator fed 18-pulse diode
rectifier. In 13th European Conference on Power Electronics and Applications, 2009.

Authorized licensed use limited to: UNIVERSITY OF NOTTINGHAM. Downloaded on February 1, 2010 at 11:08 from IEEE Xplore. Restrictions apply.