5086 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 52, NO.

6, NOVEMBER/DECEMBER 2016

An Advanced HIL Simulation Battery Model for

Battery Management System Testing
Jorge Varela Barreras, Christian Fleischer, Andreas Elkjær Christensen, Maciej Swierczynski, Erik Schaltz,
Søren Juhl Andreasen, and Dirk Uwe Sauer

Abstract—Developers and manufacturers of battery man-

agement systems (BMSs) require extensive testing of controller
Hardware (HW) and Software (SW), such as analog front-end and
performance of generated control code. In comparison with the
tests conducted on real batteries, tests conducted on a state-of-the-
art hardware-in-the-loop (HIL) simulator can be more cost and
time effective, easier to reproduce, and safer beyond the normal
range of operation, especially at early stages in the development
process or during fault insertion. In this paper, an HIL simulation
battery model is developed for purposes of BMS testing on a
commercial HIL simulator. A multicell electrothermal Li-ion
battery (LIB) model is integrated in a system-level simulation.
Then, the LIB system model is converted to C code and run in real
time with the HIL simulator. Finally, in order to demonstrate the
capabilities of the setup, experimental results of BMS tests over a
certain set of exemplary cases are shown.
Fig. 1. Overview of typical BMS functions and interfaces.
Index Terms—Battery management system (BMS), hardware-
in-the-loop (HIL), lithium ion.
heating/cooling system, run fault tolerance routines, control the
I. INTRODUCTION charging process, or provide a diagnosis of battery states such as
well-known limitation of Li-ion batteries (LIBs) is the power capability, state-of-charge (SOC), state-of-health (SOH),
A need for reliable electronic protection, i.e., a battery man-
agement system (BMS), to avoid electric or thermal abuse. The
or remaining run-time. Fig. 1 provides a high-level represen-
tation of a real battery system, from the perspective of typical
reason for this is to minimize battery degradation and to prevent BMS functions and interfaces [1]–[16].
thermal instability, conditions that can lead certain chemistries Nevertheless, since electrochemical batteries are not ideal de-
to a thermal runaway. Moreover, in order to maximize battery vices and limited sensing and actuation is possible, these BMS
capacity, active or passive balancing circuits are implemented tasks are not trivial. Voltage, current, and cell surface tempera-
in BMSs. Without balancing, if multiple cells are connected in tures are the only measurable variables and may be related to the
series, the weakest cell constrains the entire pack performance. battery states through highly nonlinear functions [17]–[19]. Due
Local operational differences and uncertainties at the material to aforementioned issues, it is easy to understand the interest in
level are related to these cell-to-cell variations, which lead to study of BMSs, especially for applications with high manage-
further differences in the cell-to-cell performance over time. But ment requirements, like e-mobility. This means that developers
an advanced BMS does not only balance and keep the battery and manufacturers of sophisticated BMSs may require extensive
within a certain safe operating area (SOA), it may also manage a testing of controller hardware (HW) and software (SW), such as
analog front-end (AFE) and the performance of generated con-
trol code. In comparison with tests conducted on real batteries,
Manuscript received September 2, 2015; revised January 29, 2016 and April
27, 2016; accepted May 24, 2016. Date of publication June 29, 2016; date of tests conducted on a hardware-in-the-loop (HIL) simulator may
current version November 18, 2016. Paper 2015-TSC-0678.R2, presented at be more cost and time effective, easier to reproduce, and safer be-
the 2015 Tenth International Conference on Ecological Vehicles and Renewable yond the normal range of operation, especially at early stages in
Energies, Monte-Carlo, Monaco, Mar. 31–Apr. 2, and approved for publication
in the IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS by the Transportation the development process or during fault simulation (see Fig. 2).
Systems Committee of the IEEE Industry Applications Society. However, as discussed in the following section, it is not
J. V. Barreras, M. Swierczynski, E. Schaltz, and S. J. Andreasen are easy to provide an HIL simulator environment able to provide
with the Aalborg University, Aalborg 9000, Denmark (e-mail: jvb@et.aau.dk;
mas@et.aau.dk; esc@et.aau.dk; sja@et.aau.dk). in real-time signals for testing a wide variety of BMS func-
C. Fleischer and D. U. Sauer are with the Institute of Power Electron- tions, such as cell protection; safety validation measuring of
ics and Electrical Drives, Rheinisch-Westfälische Technische Hochschule current, voltage, and temperature; balancing of cells; analysis
(RWTH) Aachen University, Aachen 52062, Germany (e-mail: Christian.Fleis
cher@isea.rwth-aachen.de; sr@isea.rwth-aachen.de). and parameter prediction; fault tolerance through fault inser-
A. E. Christensen is with the Lithium Balance A/S, Ishoj 2635, Denmark tion testing; logging, telemetry, and external communications
(e-mail: andreas@lithiumbalance.com). (see Figs. 1 and 2).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. In this paper, an advanced HIL simulation battery model
Digital Object Identifier 10.1109/TIA.2016.2585539 with state-of-the-art capabilities is developed for the purposes
0093-9994 © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.
BARRERAS et al.: ADVANCED HIL SIMULATION BATTERY MODEL FOR BATTERY MANAGEMENT SYSTEM TESTING 5087

II. CRITICAL REVIEW OF BATTERY EMULATION

Nowadays HIL simulators are widely employed to proto-
type, design and test complex real-time systems in many ap-
plications due to a combination of safety, feasibility, time- and
cost-effectiveness factors. In the next section, the HIL simu-
lation approaches in relation with LIBs presented in scientific
and technical literatures are reviewed with a special focus on
automotive applications.
Fig. 2. BMS testing: real battery system versus HIL simulation concept.
HIL simulation uses some virtual or emulated components
and some physical components. If a real LIB system is used,
the method is so-called battery-in-the-loop (BIL) or battery-
of BMS functional testing and fault insertion on a commercial HIL. BIL simulation is proposed in automotive applications
HIL simulator. The HIL simulator running the developed battery for the purposes of battery evaluation [20], BMS validation
model is able to provide signals for testing any BMS function, [1], verification of battery models [21], or assessment of power
such as monitoring, protection, balancing, control of the current or energy management strategies [2], [22], [23], considering a
path, heating/cooling management, diagnosis or data manage- complete virtual vehicle [1], [2], [20], [21] or in combination
ment, in the frame of fault insertion and functional testing. The with other real components [22], [23].
HIL simulation battery model is based on LIB models, which HIL simulation methods with virtual LIBs, pack-based (or
are integrated in a system-level simulation with real-time in- aggregated) approaches [3], [24]–[38] are more frequent in lit-
terface (RTI) capabilities with the commercial HIL simulator erature than cell-based (or multi-cell) approaches [4]–[13]. In
setup. Then, the HIL simulation battery model is converted to general, in order to control the charging/discharging process,
C code and emulated in real time with the HIL simulator (see any BMS for a series or series–parallel-connected battery pack
Fig. 2). Finally, in order to show the capability of the HIL sim- should at least monitor the pack voltage, the pack current, the
ulator, a commercial BMS is tested through a wide variety of single cell voltages and the temperatures at cell or pack level.
exemplary cases in an e-mobility application. Moreover, cell balancing is required to avoid severe SOC unbal-
As aforementioned, the HIL simulation battery model con- ances over time due to cell-to-cell differences [14], [16], [39].
sists not only of the LIB model, but also of certain model man- Hence, it is easy to understand that, for purposes of complete
agement controllers and other models of typical components BMS validation, cell-based HIL simulation is preferred over
of an LIB system as shown in Section III. The model man- pack-based.
agement controllers are used to configure the HIL simulation HIL simulation using virtual LIB pack-based methods is pro-
battery model for different test cases. Typical components of posed in stationary [36]–[38] and automotive applications [3],
an LIB system considered in the simulation include the tem- [24]–[35], the latter with the goal of electric powertrain eval-
perature and current sensors, the switch box, the charger or the uation [24]–[28], BMS validation [3], [29], vehicle controller
load, which is modeled based on a conventional dynamic model unit testing [30], assessment of power or energy management
of a battery electric vehicle. With regard to the LIB model, strategies [31], [32] or verification of the feasibility of the HIL
an advanced electrothermal multicell equivalent circuit model concept [33], [34]. On the other hand, HIL simulation using vir-
(ECM) coupled with an aging model is proposed. It is worth not- tual LIB cell-based methods is proposed either for addressing
ing that the aging model presented is not used for any purposes BMS validation [4]–[12] or only testing balancing circuits [13].
other than generating realistic cell-to-cell parameter variations. If there is high power flow between real components and the
In this way, different realistic testing scenarios may be repro- HIL simulator, the technique is called power-HIL (PHIL) [24]–
duced. This approach, to the best of our knowledge, is a novelty [28], [31], [32], [34]–[36]. This a common feature of pack-
in the literature. based HIL simulators used for electric powertrain evaluation
The paper is organized as follows. In Section II, the con- [24]–[28]. PHIL simulation is also applied in [35] to lead-acid
cept of battery emulation is reviewed based on the literature. In technology, where a portable system using virtual 12 V SLI
Section III, the developed HIL simulation battery model is pre- battery is presented and tested during motor cranking in a car.
sented, with focus on the LIB system. The electrothermal mul- In contrast, the usual case in HIL simulation only involves a
ticell ECM coupled with an aging model is described and its low level of power, voltage, and current related with analog and
parameterization is detailed, based on experimental data from digital I/O signals or communication networks, e.g., Controller
large format 40 Ah Li-ion pouch cells with nickel manganese Area Network (CAN) bus [1]–[13], [20]–[23], [29], [30], [33],
cobalt oxide (NMC) cathode manufactured by Kokam. Mod- [37], [38]. In general, this is the case of all the cell-based HIL
els of other typical components of the battery system are also simulation approaches found in the literature [4]–[13]. In some
outlined in this section. In Section IV, the experimental setup cases, a strictly communication-based approach is proposed [4]–
for HIL simulation is presented, distinguishing between the [6], [10], [30], [37], [38], which means that BMS functions such
commercial BMS under test and the HIL simulator itself. In as measuring or balancing cannot be tested, either at cell or pack
Section V, experimental results of the interaction of the BMS level.
with the real-time simulator are provided. Finally, Section VI Regarding balancing, an isolated power interface for each
gives the conclusion. cell in series (or group of cells in parallel in case of a
5088 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 52, NO. 6, NOVEMBER/DECEMBER 2016

series–parallel-connected pack) is desired to exchange bidirec-

tional power flow between virtual cells and real electric circuits.
But such interface is only presented in cell-based simulators de-
scribed in [11]–[13]. Results are shown, respectively, for only
1, 4, or 6 cells in series, and current levels at 200 mA, 1.45 A,
or 5 A. Moreover, it should be noted that the setup presented in
[13] is only targeted for testing balancing circuits, and not for
testing any other BMS functions.
With regards to the temperature emulation, it is claimed in
several references, but the thermal model is either not described
[3], [4], [7], [29], [33] or the temperature is artificially intro-
duced [5]. In fact, a thermal model is only presented in [34].
However, a pack-based approach is followed and the lumped
ECM described is parameterized without experimental data.
Moreover, the self-made HIL simulator setup presented can
only run discharging profiles.
The most common approach for battery modeling methods is Fig. 3. Overview of the HIL simulation battery model.
electrical models based on ECMs, e.g., impedance-based mod-
eling (abstract approach) [3], [6], [7], [11]–[13], [21], [24]–[29],
[15], [39]. Similarly, testing of control strategies for balancing
[32], [34]–[37]. Other approaches are proposed rarely in the lit-
systems can be performed using an HIL simulator equipped
erature. ECMs can provide information about the macroscopic
with a communication interface, e.g., CAN-based. However, a
quantities that are monitored by a BMS, i.e., current, voltage,
complete analysis and validation of balancing system SW and
and temperature (if a thermal model is included), offering a bet-
HW can only be achieved if the HIL simulator setup can provide
ter tradeoff between complexity and accuracy than other math-
isolated power interfaces for each cell (or group of cells in par-
ematical or physic-chemical approaches, like the ones proposed
allel) in series. Taking into account the considerable attention
in [24] and [31]. In [24], an extended modified Shepherd model
that balancing systems have enjoyed in the literature since the
(abstract approach) is proposed, and in [31], a dual-foil-based
introduction of LIBs [16], [44]–[46], it is easy to understand the
model (physic-chemical approach).
importance of this advanced feature.
However, ECMs do not allow extrapolation from one material
As a rule of thumb, it can be stated that, for purposes of general
or cell designs to another one. Therefore, for a universal BMS
functional analysis of a BMS, simpler battery models may be
validation, a library of battery models for different cell sizes,
used. However, more accurate models may be implemented
formats, chemistries, and manufacturers are desired. Such a
with the aim of ensuring that complex functionalities of a BMS,
library is not presented in the reviewed literature.
such as diagnostic algorithms for SOC, state-of-function or SOH
Moreover, the accuracy of the ECM lays on the circuit topol-
estimation, can be analyzed and validated using HIL simulation.
ogy and the component dependencies on SOC, SOH, current,
or temperature. The common approach is to use nonlinear dy-
namic ECMs with two RC elements. This circuit topology may III. HIL SIMULATION BATTERY MODEL
account for most important dynamic effects. However, simple In this section, the HIL simulation battery model is presented,
or no dependencies are usually considered in the literature. For with special focus on the LIB system model. As aforementioned,
example, only SOC dependencies in open circuit voltage are the HIL simulation battery model consists not only of the elec-
considered in [6], [12], [25], [27], [36], [37] and the dependen- trothermal multicell ECM coupled with an aging model, but
cies are not considered or are not well defined in [3]–[5], [26], also of the simulation model management controllers and other
[28], [33], and [34]. models of typical components of a battery system, like the tem-
It should be noted that even if a nonlinear dynamic ECM with perature and current sensors, the switch box, the charger, or the
two RC elements is used, the model may not reproduce accu- load as shown in Fig. 3.
rately realistic battery profiles if only a few or nondependencies In Sections III-A-E, the design and characterization of the LIB
are considered in the circuit components, as discussed thor- electrothermal multicell ECM coupled with an aging model is
oughly in the literature related to battery modeling [17]–[19], presented. Parameterization is based on a commercial 40-Ah
[40]–[43]. high-energy Kokam SLPB100216216H pouch cell. The ECM
Furthermore, in order to account for cell-to-cell differences model consists of an inductance, an ohmic resistor, and two
or emulating different testing scenarios, an aging model can ZARC elements in series. Experimental pulse power charac-
be coupled to the electrical model. This approach is followed terization, capacity check, electrochemical impedance spec-
in this paper and to the best of our knowledge is a novelty in troscopy (EIS), and anisotropic thermal diffusivity measure-
the literature. However, it is worth noting that aging prediction ment techniques are used for parameter estimation, taking into
models are out of the scope of this paper. With rising interest account aging, temperature, and SOC dependencies. Results
of BMS’s applications, like e-mobility and Smart Grid, these are presented for a Kokam cell for illustrative purposes, but
complex functionalities have become increasingly important a full library has been developed including battery cells with
BARRERAS et al.: ADVANCED HIL SIMULATION BATTERY MODEL FOR BATTERY MANAGEMENT SYSTEM TESTING 5089

Fig. 5. Approximation of ZARC element through series connection of RC

elements [34].

Fig. 4. Nyquist plot of the complex impedance of a Kokam 40-Ah NMC-

cathode cell.

different formats (cylindrical, pouch, prismatic), capacities (2,

40, 53, 100 Ah) and cathode chemistries [NMC, Lithium Ti-
tanate Oxide (LTO), Lithium Iron Phosphate (LFP)], including
other manufacturers such as Sanyo, Panasonic, or SkyEnergy. In
Section III-F, the temperature sensors model is outlined. Finally,
in Section III-G, the fuse and switch box model is presented,
taking into account all the components.

A. Impedance Spectrum Analysis

In Fig. 4, EIS measurements conducted on the Kokam 40-Ah Fig. 6. Comparison of different model approximations.
NMC-cathode cell are presented in a Nyquist plot. At high fre-
quencies (> 0.8 kHz), the cell shows inductive behavior caused
by inductive reactance of metallic elements in the cell and wires.
B. Approximation of ZARC Elements
This is modeled as an inductor in series in the ECM. At a fre-
quency of 375 Hz, pure resistive behavior is observed, related ZARC elements are able to accurately reproduce the
with the sum of the ohmic resistances of current collectors, ac- impedance spectra of LIBs, but it is impossible to transfer with-
tive material, electrolyte, and separator. This effect is modeled out approximations their parameters from the frequency domain
as a resistor in the ECM. At lower frequencies, one depressed into the time domain. A common method to approximate ZARC
semicircle is observed in the spectrum. This is associated to an elements is using a variable number of RC-elements (see Fig. 5).
Solid Electrolyte Interphase (SEI) effect superimposed with the Each of the individual RC elements represents a semicircle in
double layer capacity and the charge-transfer resistance at the the Nyquist plot. To represent the ZARC exactly, infinite RC
electrodes. All these effects can be modeled by using a ZARC elements are needed. Nevertheless, taking into account the com-
element in the ECM. At frequencies below 0.375 Hz, a linear putational cost versus accuracy, a few RC elements are sufficient
slope with an angle of 45◦ is observed. This is associated with for a good approximation.
the diffusion process and can be also approximated by another As observed in Fig. 6, the error for five RC elements in
ZARC element. comparison to the approximation with only one RC element
A ZARC element is made up of a resistor in parallel with a is reduced to a satisfactory result. The overall approximation
constant-phase-element (CPE): equation is derived as follows [17]:
  ξ1 R1
1 −ξ Z2ZARC (jω) = Rser +
ω0 = ⇐⇒ Z = A · (jω) (1) 1 + (jω)−ξ · R1 · C1
A·R
R2
1 + . (4)
ω0,ZARC = ω0,RC =
CRC · R
(2) 1 + (jω)−ξ · R2 · C2
1
⇐⇒ CRC = . (3) C. ECM Parameter Identification
ω0,RC · R
Some parameters of the ECM can be directly derived from
A CPE consists of a generalized capacity and depression the impedance spectrum. For example, the series resistance at
factor. If the depression factor is equal to 1, the ZARC element high frequencies can be estimated from the intersection of the
corresponds to a simple RC element. If it is equal to 0, it just spectrum with the real part axis. Other parameters, like the
represents a resistor in series. inductor in series, can be neglected in some cases [18], for
5090 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 52, NO. 6, NOVEMBER/DECEMBER 2016

TABLE I
THERMAL PARAMETERS EXTRACTED BY ANISOTROPIC THERMAL DIFFUSIVITY
MEASUREMENT

Heat conductivity in-plane 33 W/(m · K)

Heat conductivity through-plane 0.61 W/(m · K)
Density 1.982 kg/(dm3 )
Heat capacity 860 J/(kg · K)
Mass 1.18 kg

Fig. 7. Employed equivalent circuit for thermal modeling.

Fig. 8. Active thermography test bench.

example, if electric vehicle applications are simulated, due to
their characteristic slow dynamics. Moreover, it does not make
sense to simulate phenomena that cannot be measured due to
At battery pack level the following heat transfer phenomena are
limited sampling rate of the BMS or the HIL simulator.
considered:
An in-house tool based on MATLAB Easyfit function is ap-
1) transfer within a cell (mostly heat conduction);
plied to fit the experimental EIS data in order to parameter-
2) transfer to the environment (convection and radiation);
ize the ECM. The Nelder–Mead simplex search algorithm is
3) transfer to neighboring cells (heat conduction, similar
used, which allows an accurate and precise emulation of the
to a.).
impedance spectra based on the modeling function. This func-
The heat conductivity is defined by
tion is crucial, because it defines the quality and robustness of
the parameter identification process. It should be noted that it dT 1 
= · (−T − Ti /Ri ) (10)
is always useful to define starting parameters derived directly dt C
from the impedance spectrum [19]. with index i describing the thermal resistance Ri and conductiv-
ity of the neighbors. The heat exchange with the environment is
D. Thermal Model based mainly on the convection and radiation and described as
 
Thermal effects modeled at the cell level using an ECM are dT 1  −dQconv dQrad
illustrated in Fig. 7. The underlying processes such as heat flow = · − (11)
dt C dt dt
dQ/dt or temperature are converted to current and voltage.
Therefore, radiation, convection, and heat conduction are rep- with
resented by a voltage drop over nonlinear thermal heat transfer dQconv
resistances as = αcon · A · (T − Ti ) (12)
dt
Rradiation = T1 − T2 /(A · α1,2 · (T14 − T24 )) (5) dQrad
= αcon · A · (T 4 − Ti4 ). (13)
dt
Rconvection = 1/A · α (6)
The thermal parameters presented in Table I were determined
Rconduction = d/λ · A (7) by nondestructive experimental tests. The heat capacity was just
with d representing the thickness, λ the specific heat conductiv- calculated from cooling curves under controlled ambient tem-
ity, and α the convection coefficient. peratures. With regard to thermal conductivity, it was calculated
The temperature of a volume element is calculated using from the product of cells density, specific heat capacity, and ther-
the current temperature and its change. The temperature of an mal diffusivity as described in the American Society for Testing
element is increased by the temperature difference T , once a and Materials (ASTM) standards. The tests were conducted on
heat Q is applied. The heat is proportional to T : a self-made active thermography test bench (see Fig. 8). Its de-
sign, measurement principle and data analysis methods were
 Q = C · T. (8) directly based on laser flash method, but using different heat
power sources, following ASTM standards.
C represents the heat capacity of the element and depends on
A flash lamp method was used for calculation of in-plane
the material and mass. T is calculated by
  thermal diffusivity from temperature variation measurements.
dT 1 dQreceived dQdelivered After exciting the front plane of the cell uniformly with a flash
= − . (9)
dt C dt dt lamp, the rise of temperature over time in the rear face plane was
BARRERAS et al.: ADVANCED HIL SIMULATION BATTERY MODEL FOR BATTERY MANAGEMENT SYSTEM TESTING 5091

Fig. 9. Thermograms from in-plane thermal diffusivity measurements.

Fig. 11. Online integration of individual cell aging simulation with elec-
trothermal model.

internal resistance and loss of capacity are monitored every 100

Fig. 10. Thermal diffusivity: experimental results versus model fitting. driving cycles conducting checkup or reference performance
tests.
recorded by a commercial thermal imaging camera provided by The checkup tests consist of a pulse power characterization
Thermosensorik (see Fig. 9). test for current dependence investigation and a capacity check.
A special fixture was designed to keep the pouch cell stable The pulse power characterization test is intended to determine
during the tests. The frame consisted of electrically and ther- the internal resistance over the cells usable voltage range us-
mally insulated Pertinax base plates blanked out in the exact ing a test profile that includes both discharge and charge/regen
dimensions of the cell. Then the cell was clamped between two pulses. The test profile is applied to the cell under test for differ-
plates. Any gaps between the cell and the base plates were cov- ent temperatures in consecutive steps of 10% SOC. The SOC is
ered with thermally insulated tape in order to prevent unwanted adjusted based on the accumulated Ah referring to the nominal
transmission of infrared radiation to the opposite side of the cell. capacity. The resistance is calculated as the difference between
The battery cell surface was cleaned and sprayed with graphite the voltage before the discharge pulse and Δt after the beginning
for uniformly distributing the flash light energy over the entire of the discharge pulse, divided by the amplitude of the current.
surface. On the other hand, a liquid-cooled single-winded in- The capacity check tests consist of a typical CCCV charge and
duction coil supplied by a high-frequency generator was used CC discharge cycle conducted at 1 C-rate and different tem-
as heat power source for through-plane measurements. Due to peratures. It should be noted that during the checkup tests the
the cell surface-to-thickness ratio, based on the laboratory expe- cells are operated always inside an SOA limited by certain tem-
rience, an induction coil placed close to the cell tabs is preferred peratures, voltages, and currents specified in the manufacturer’s
over the flash lamp for through-plane experiments on LIB pouch data sheet.
cells. Thermal diffusivity was finally obtained from fitting the Finally, the ZARC element parameters are identified based on
whole experimental data using an in-house tool based on MAT- EIS measurements. The impedance spectra are also measured at
LAB (see Fig. 10). It is worth noting that in this paper heat different SOC and temperatures for using frequencies between
generation due to entropic or phase change, changes in the heat 10 mHz and 5 kHz.
capacity and mixing have all been neglected. With regard to the To represent the individual aging states of the cells, three parts
contribution of reversible or entropic heat, it should be noted are integrated as shown in Fig. 11:
that it is neglected in many studies when the focus is not in bat- 1) impedance-based electric model;
tery performance at low C-rates. Regarding the heat generation 2) thermal model;
due to phase change, changes in the heat capacity and mixing, 3) aging simulation.
they are commonly neglected [14], [40], [47], [48]. Every aging parameter was extracted separately and can be
changed offline or on-the-fly while running an HIL simulation.
E. Aging Model for HIL Simulation Hence, it is possible to represent realistic operating conditions
of multicell battery packs by applying a statistical distribution
In order to emulate the variation of cell-to-cell characteristics, of the individual aging states. Therefore, using the three part
the impedance-based ECM is coupled to an aging model. This LIB model proposed here, a full range of realistic scenarios can
model maps the results of the parameterization from accelerated be addressed at the LIB battery system-level HIL simulation.
aging tests of battery cells cycled with a real driving cycle.
A physical-based function is extracted from the experimental
F. Temperature Sensor Model
data in order to describe the aging state of the cell. All the
cycling tests were carried out at different temperatures based on A BMS may monitor temperature sensors throughout the
the average yearly temperatures of Germany. The increase of whole battery system, although battery systems may differ in
5092 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 52, NO. 6, NOVEMBER/DECEMBER 2016

Fig. 12. Abstraction of an NTC thermistor-based temperature measurement

of a BMS.

Fig. 13. Abstraction of HIL simulation setup for an NTC thermistor. Fig. 14. Schematic of exemplary fuse and switch box topology and its inter-
faces.

the number, location, and type of devices used to measure the commonly integrated in a so-called fuse and switch box. More-
temperature. The commercial BMS under test in this paper uses over, the system may require: a precharge circuit, for energiz-
negative temperature coefficient (NTC) thermistors. One tem- ing the dc-link capacitor; a current measurement device, for
perature sensor per cell, located in a central position, is assumed instance, a shunt resistor or a hall effect sensor; and external
for emulation purposes. NTC thermistors are thermally sensitive communication with a BMS, which may control the relays and
semiconductor resistors that exhibit a nonlinear decrease in ef- measure the pack voltage and current. The design of the fuse
fective resistance as temperature increases. This electrothermal and switch box, particularly the device selection and location,
behavior can be modeled analytically, by means of the Stein- may differ widely depending on the specific characteristics of
hartHart equation [49], [50]. In practice, NTC thermistors are the LIB system and its application.
usually characterized in manufacturers datasheets by a B pa- For illustrative purposes, an exemplary topology of a fuse
rameter, which appears in the so-called B parameter equation. and switch box and its interfaces is shown in Fig. 14. This
The B parameter equation presented in the following is derived fuse and switch box is made up of four relays and two fuses,
from the well-known Steinhart–Hart equation using a certain a precharge resistor Rprecharge and a shunt resistor Rshunt . Note
change of variables: that a BMS with a master–slave topology is represented, since
  it is the case of the commercial BMS tested in this paper, as
1 1 1 Rs
= + ln (14) explained in the next section. The relays allow the system to
Ts Ts,0 B Rs,0
operate in three modes: stand-by or idle, discharge, and charge.
where all the temperatures are in Kelvin and all the resistances Furthermore, it should be noted that, since the dc-link may
in Ohm, with Ts representing the current semiconductor sensing be connected to a capacitive load at the load or the charger
device absolute temperature, Rs the current resistance, and Rs,0 side, i.e., a dc-link capacitor, inrush currents have to be limited
the device resistance at temperature Ts,0 = 25 °C. Solving for whenever the operating mode changes from stand-by to charge
Rs results in or discharge. For this purpose, a precharge circuit, which may
  consist of a precharge resistor in series with a precharge relay,
−B 1
T s,0 − T1s .
Rs = Rs,0 e (15) could be implemented (see Fig. 14). The BMS is behind all the
relays logic and controllers. In the HIL simulator presented in
Knowing the characteristics of the sensor, represented by B,
this paper, both the relays control signals and LIB pack voltage
Rs,0 , and Ts,0 , and the parameters of the sensor excitation cir-
vpack signals (see Fig. 14) are transmitted from/to the BMS via
cuit, represented by Vs,aux and Rs,aux , Ts may be estimated
CAN communication. Fuses are not considered, but a complex
analytically in the BMS from a simple measure of the NTC
fuse model may be included in future extension of this work.
thermistor voltage vs . Another approach, illustrated in Fig. 12,
Regarding the relays, they are represented as ideal switches in
is to generate and implement a NTC temperature-voltage look-
series with a static resistor. The precharge resistor Rprecharge (see
up table. Based on the previous, a resistorless emulation of an
Fig. 14) is rated according to a desired precharging time of
NTC thermistor is possible by means of a controlled voltage
1 s, taking into account the system nominal voltage, the dc link
source, as sketched in Fig. 13. A dedicated analog output board
capacity and the power dissipation.
of the HIL simulator is used in this paper for this purpose.
On the other hand, the BMS is configured assuming that a
shunt resistor Rshunt is used for pack current ipack measurement
G. Fuse and Switch Box Model using analog circuitry (see Fig. 14). By using Ohm’s law, a
Due to safety issues, LIB systems include installation of simple resistorless emulation is possible by means of a con-
dc circuit breakers in the current path. In high-voltage LIB trolled voltage source. Since a dual voltage analog output is not
packs, used, e.g., in e-mobility applications, fuses and relays are available in the HIL simulator setup, an external single-to-dual
BARRERAS et al.: ADVANCED HIL SIMULATION BATTERY MODEL FOR BATTERY MANAGEMENT SYSTEM TESTING 5093

It can provide a continuously controllable voltage in a range of

0–6 V 1.5 mV for each of the outputs. This relatively wide
voltage range allows not only emulation of both overvoltage
and under voltage scenarios, but also emulation of short circuit
faults between cells. Furthermore, broken wire faults could be
also reproduced by controlling a series of switches placed in an
interface board between the connection board and the EV1077
boards.
The EV1077 voltage outputs are galvanically isolated, allow-
ing series connection. In the experiments presented in this paper
8 EV1077 boards are used to emulate 32 cells in series, but
the setup could be extended with up to 24 EV1077 boards in
order to allow emulation of 128 cells (or group of cells in par-
allel) in series. Regarding the dynamic response, any voltage
reference value step can be followed in less than 500 s. This
Fig. 15. Setup of hardware-in-the-loop test bench.
feature combined with high computational power of the DSP
and a fast data transmission enables a maximum update rate
for all the cell’s voltages of 1 kHz. Furthermore, a bidirectional
voltage conversion board is implemented. Note that the polarity
balancing current of maximum 1 A is allowed, which enables
of the shunt voltage drop depends on the direction of the LIB
testing of both active and passive cell balancing systems. A
pack current.
number of other I/O cards are used to provide communication
interfaces (e.g., CAN Bus) or emulation of temperature sensing
IV. EXPERIMENTAL SETUP FOR HIL SIMULATION or battery pack current sensing-based continuously controllable
A. Commercial BMS Under Test voltage outputs.
The LIB system simulation model, illustrated in Fig. 16, is
The commercial BMS under test is the sBMS v6 provided by
developed offline on MATLAB/Simulink SW from Mathworks.
Lithium Balance A/S (see Figs. 15 and 16). The BMS uses a
An implementation SW provided by dSPACE, the so-called RTI,
master–slave configuration, where each slave unit can be con-
provides blocks that link the I/O capabilities of the dSPACE HIL
figured to supervise eight cells and two temperatures. The slave,
simulator to the MATLAB/Simulink models. Furthermore, RTI
so-called local monitoring unit, is capable of passive balancing
extends the functionalities of the MATLAB/Simulink C code
with up to 1 A simultaneously on all eight cells and will au-
generator, enabling automatic generation of the C code from the
tomatically reduce the balancing power if the heat cannot be
offline LIB system model. Once compiled, this code is ready to
dissipated at high enough rates. Four slave units are used in the
be downloaded to the dSPACE HIL simulator HW from a host
experiments presented in this paper. However, a total of 32 slave
PC, which is connected to the real-time HW via a link board.
units can be used for each master, giving a total of 256 super-
An optical interface is used for purposes of galvanic isolation. A
vised cells and 64 temperature inputs. The master, the so-called
universal modular experiment and instrumentation SW provided
battery management central unit, is capable of measuring LIB
by dSPACE, the so-called ControlDesk, is installed in the host
pack current, LIB pack voltage, isolation faults, and contactor
PC in order to set up and control experiments. It is also used
faults. Communication to external systems is provided through
to download the model code on the DSP (DS1006 processor
a configurable CAN bus interface, where any parameter can
boards), the core of the real-time HIL simulator.
be broadcasted. The control of contactors, switches, and fans
are all implemented in the BMS and can have a multitude of
V. EXPERIMENTAL RESULTS
configurations.
In this section, the LIB model validation and experimen-
B. HIL Simulator tal results are presented. The BMS tests performed on the
HIL simulator setup comprises exemplary functional and fault
The commercial HIL simulator test bench is provided by
insertion tests.
dSPACE GmbH (see Fig. 15) and interfaces with the BMS un-
der test and a host PC either via electrical, optical, or CAN bus
interfaces (see Fig. 16). It is a very complex modular system A. LIB Model Validation
built according to customer-specific requirements. It consists As a first step, the electrothermal LIB model is validated in
of multiple racks and integrates many off-the-shelf components simulations. Fig. 17 shows the simulated voltage response in
like digital signal processors (DSP), I/O cards, cell voltage em- comparison to the real measured voltage. The dynamic load
ulation boards, power supplies, or signal conditioning boards. profile applied was logged in a real EV and processed for sim-
The voltage emulation at the cell level is performed with ulation. It can be noticed that the model is able to follow the
several EV1077 battery cell voltage emulation boards. Each of measured voltage response. The same can be concluded from
these controllable buffer amplifier boards can emulate four cells the temperature comparison. The starting temperature is initial-
and be configured in order to match specific test requirements. ized at 25 °C and stays constant due to lower dynamical driving.
5094 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 52, NO. 6, NOVEMBER/DECEMBER 2016

Fig. 16. Overview of main functions and interfaces between the host PC, the HIL simulator, and the BMS.

TABLE II
OVERVIEW OF FEASIBLE BMS TESTING SCENARIOS USING DEVELOPED HIL
SIMULATION PLATFORM

Functional testing Dynamic discharging.

CC-CV charging.
Overtemperature during discharge.
Charging test at low temperatures.
Overcurrent during charge.
Short circuit(s) at cell, pack or system level.
Overvoltage during regenerative braking.
Diagnostic algorithms.
Isolation monitor.
Idle or stand-by.
Global power consumption test.
Non-functional testing Current, voltage, temperature accuracy.
Fig. 17. Electrothermal battery model validation between measurement data Baseline, compliance, endurance, load,
and model simulation. Performance, recovery, scalability, stress,
Usability and volume testing.
Fault insertion testing Open circuit in interface(s).
Short circuit in interface(s).
As soon as the load current increases, a rise in temperature is Noise in interface(s).
observed due to higher power losses. The model is also able to Increase of interface(s) line resistance.
Sensor(s) malfunction.
follow this trend accurately. Actuator(s) malfunction.
System-level component(s) malfunction.
B. BMS Testing on HIL Simulator
In the next step, the BMS is tested on the HIL simulator
running the developed system model, in the frame of functional as a form of stress testing, which, in turn, can be classified as a
and fault insertion testing. During functional testing, all the type of nonfunctional testing.
low-level functions of a BMS may be tested. Additionally, each In this paper, only exemplary functional and fault insertion
of these low-level functions may be related to a particular SW tests are presented. Although, using the same HIL simulator
and/or HW of the BMS, which, in turn, may be related to a platform, a wider range of testing scenarios can be explored
certain electric or electronic HW interface between the BMS (see Table II). A comprehensive collection of functional, non-
and the rest of the system. On the other hand, during fault functional, and fault insertion tests will be presented in a future
insertion tests, the behavior of the system may be studied when extension of this work. For instance, with regard to fault inser-
faults are inserted either in the BMS SW or HW or its HW tion testing, next scenarios will be considered: HW-based fault
interfaces. insertion in the HW interface, due to open circuit, short circuit,
The importance of fault insertion testing lies in the ability to noise or increase of line resistance; malfunction of sensors and
evaluate the reliability and robustness of a BMS, key parameters actuators, like temperature or current sensors; or malfunction of
for the industry today [15], [39], [48]. Additionally, it may be system-level components, like the dc-link capacitor or fuse and
of relevance to note that fault insertion testing can be classified switch box components.
BARRERAS et al.: ADVANCED HIL SIMULATION BATTERY MODEL FOR BATTERY MANAGEMENT SYSTEM TESTING 5095

Fig. 18. HIL battery cell emulation using dynamic driving cycle.

Fig. 20. CC-CV charging with balancing.

Fig. 19. Contactor logic of the BMS. Due to integration limitations, the system
voltage output is only activated when contactor sequence has completed.

Last but not least, since the HIL simulator platform allows Fig. 21. System voltage and current during CC-CV charging.
different battery pack topologies, it is also possible to test a mul-
titude of serial, parallel, or mixed serial–parallel connections of
cells. This may be important for testing battery packs consisting
of multiple modules and how a BMS handles scenarios related
to modules failing or being replaced.
1) Exemplary Functional Tests Results: An exemplary dis-
charging scenario in an e-mobility application is shown in
Fig. 18. Voltages of 32 cells in series during a dynamic dis-
charge are presented. Simulation parameters such as initial cell
temperature and ambient temperature are selected assuming that
the underlying distribution is normal. Initial cells SOC are the
result of a previous discharge/charge cycle starting from a ran-
dom initial SOC distribution. Moreover, an intentionally wide
random distribution of aging parameters is considered, in order
to make cell-to-cell variations in capacity and impedance more
marked. BMS functions related with monitoring, switch box
control or protection can be checked with this functional test. Fig. 22. Maximum and minimum cell SOC for the balancing sequence.
With regard to the switch box control, Fig. 19 shows the contac-
tor logic when the system operating mode changes from idle or
stand-by to discharge, illustrating a precharge sequence. Note are considered here. A random distribution of the initial SOCs
that the system voltage is not correctly modeled, but instead ap- of the cells within a certain window is used. The initial SOC
plied when the contactor on the positive leg of the battery pack differences are equalized by the BMS balancing circuits during
is closed. In future work this will be added. the charging process, in order to maximize usable battery ca-
An exemplary CCCV charging scenario, considering gross pacity. Note that the commercial BMS under test has a terminal
cells SOC unbalance, is illustrated in Figs. 20–23. The same voltage control algorithm for a passive balancing system imple-
cells characteristics used for the previous discharging scenario mented. The parameters for the balancing window are defined
5096 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 52, NO. 6, NOVEMBER/DECEMBER 2016

Fig. 25. Model-based fault insertion: Critical zero volt mode.

Fig. 23. Temperatures recorded during initial CC-CV charge sequence.

applied and within a short time the voltage of cell 1 (blue curve)
tends to diverge from the average. Additionally, due to higher
resistance in cell 1, the temperature is rising to a higher level
in comparison to the healthy cells. Such a scenario may be de-
tected by a BMS and counter measures may be performed, either
by BMS or the energy management of the vehicle supervising
unit. These measures are part of the so-called fault tolerance
routines. Complex algorithms computed on a BMS master unit
may be able to detect the problem. However, simpler algorithms
(e.g., cell voltage average) implemented either at the master
or slave level may also be able to detect the fault, providing
a higher degree of redundancy. In the case of the BMS under
test, fault tolerance routines against this kind of fault are not
implemented. Although, the BMS under test was able to detect
Fig. 24. Model-based fault insertion: Higher series resistance for cell 1. cell over voltage and even increase line resistance. However, the
latter functionality is only effective during charging.
in the GUI of the commercial BMS. After the BMS starts the Another typical fault test scenario considers the interconnec-
charging process, the voltage increases gradually. As soon as tion error between the BMS slave board and the terminals of the
the cells approach the cell target voltage window, the balancing cell. Loose connectors, noise, shorts, severed wires, or blown
is activated as illustrated in Fig. 20. Due to balancing, the over- fuses may lead to erroneous voltage sensor. In Fig. 25, a criti-
shoot is reduced within the allowable window. Since the value cal zero volt mode is presented. Since the controller is not able
of the pack charging current is much higher than the value of to measure the cell voltage, the BMS goes into a critical fault
the cell balancing current, at some point, the charging process mode. Depending on the configuration settings of the BMS con-
has to be stopped, in order to extend time for balancing. All troller, a further charge/discharge the pack may not be allowed.
the cells, even though having different characteristics and initial This open-circuit fault is inserted by opening a physical relay
voltages, are kept in a narrow voltage window within the cell that exists in the output of the cell voltage emulation boards of
target voltage window at the end of charge (see Figs. 20 and the HIL simulator. Fig. 25 shows this scenario for a load driving
21), which means that a reduced cell-to-cell variation in SOC is profile with 32 cells in series. After 950 s, the cell voltage of
achieved (see Fig. 22). BMS functions such monitoring, switch cell 12 (red signal) suddenly drops to zero volts due to fault
box control, charge current control, balancing or protection are insertion. The same kind of test can easily be performed for the
checked with this functional test. temperature sensors. However, an open circuit in temperature
2) Exemplary Fault Insertion Tests Results: Fig. 24 illus- sensors does not lead to a critical fault mode with the BMS
trates an exemplary failure caused by a higher internal resis- under test.
tance and self-discharge or cell interconnection problems. In Another exemplary fault insertion scenario, based on an actu-
order to emulate this fault, line resistances are added to the cell ator malfunction, is shown in Fig. 26. In this case, a malfunction
single model during operation. The figure shows the dynamic of the charging current controller is considered. Although the
response of the system in an e-mobility application. With the charging current setpoint is set to 40 A, the charger is giving
aim of representing realistic operating conditions, a relatively 60 A, i.e., a 50 % increased current. This is easily achieved in
narrow normal parameter distribution of the individual aging simulation by manipulation of the control signal received from
states is applied to the 32 cells in series of the LIB consid- the BMS. In real life, errors in the charger or BMS settings may
ered in the setup. At high levels of discharge or charge/regen lead to this kind of fault. As illustrated in Fig. 26, after un-
power, the effects of this cell-to-cell parameter variation are successful attempts to control the charger current during a few
evident. After a simulation time of 1000 s, the failure mode is seconds, the BMS stops the charging process. A few seconds
BARRERAS et al.: ADVANCED HIL SIMULATION BATTERY MODEL FOR BATTERY MANAGEMENT SYSTEM TESTING 5097

and fault insertion tests will be presented in a future extension

of this work.
ACKNOWLEDGMENT
The authors would like to thank The Danish Council for
Strategic Research for sponsoring the following projects: the
Advanced Lifetime Predictions of Battery Energy Storage and
ReLiable.
REFERENCES
[1] A. Ebner, F. V. Conte, and F. Pirker, “Rapid validation of battery man-
agement system with a dymola hardware-in-the-loop simulation energy
storage test bench,” World Elect. Vehicle Assoc. J., vol. 1, pp. 205–207,
2007.
Fig. 26. System voltage, cell voltages, and system current during fault injec-
[2] Y. Yuan, X. Wei, and Z. Sun, “Assessment of power consumption con-
tion with 50 % increased charge current.
trol strategy for battery management system using hardware-in-the-loop
simulation,” in Proc. IEEE Vehicle Power Propulsion Conf., 2008, pp.
1–6.
later, the BMS tries to start the process again, but fails again. [3] Y. Li, Z. Sun, and J. Wang, “Design for battery management system
The same cycle is repeated several times with similar results. hardware-in-loop test platform,” in Proc. 9th Int. Conf. Electron. Meas.
The BMS could be configured to allow only a certain number Instruments, 2009, pp. 3-339–3-402.
[4] H. Wu, “Hardware-in-loop verification of battery management system,”
of retries before coming into a critical fault mode. in Proc. 4th Int. Conf. Power Electron. Syst. Appl., 2011, pp. 1–3.
[5] R. Subramanian, P. Venhovens, and B. P. Keane, “Accelerated design and
optimization of battery management systems using HIL simulation and
VI. CONCLUSION rapid control prototyping,” in Proc. IEEE Int., Elect. Vehicle Conf., 2012,
pp. 1–5.
For purposes of general functional analysis of a BMS, simple [6] H. Rathmann, C. Weber, W. Benecke, and D. Kahler, “Sophisticated esti-
LIB models are commonly used in HIL simulation. However, mation of hardly measurable conditions of lithium-ion batteries,” in Proc.
complex LIB system models are needed to ensure that all the 39th Annu. Conf. IEEE Ind. Electron. Soc., 2013, pp. 1862–1866.
[7] S. Wu, Y. Zou, X. Peng, and H. Li, “Hardware-in-loop verification of
functionalities of an advanced BMS can be analyzed and vali- battery management system with RT-LAB,” in Proc. IEEE Conf. Expo,
dated, including, e.g., monitoring, protection, balancing, diag- Transp. Electrification Asia-Pacific, 2014, pp. 1–4.
nostics, or fault tolerance. [8] Q. Wang, X. Z. Wei, and H. F. Dai, “Hardware-in-loop test platform for
electric vehicle cell battery management system,” Appl. Mech. Mater.,
This paper presents an advanced HIL simulation battery vol. 29–32, pp. 2398–2403, Aug. 2010.
model for testing state-of-the-art BMS on a commercial HIL [9] G. L. Plett, R. Billings, and M. J. Klein, “Desktop and HIL validation
simulator. In comparison with the test conducted on real bat- of hybrid-electric vehicle battery-management-system algorithms,” SAE
2007 World Congr., 2007.
teries, BMS validation tests conducted on this platform may be [10] J. Zeng, J. J. Sun, and Y. Ma, “The system architecture design about
more cost and time effective, easier to reproduce, and safer be- test platform of battery management system,” Adv. Mater. Res., vol. 645,
yond the normal range of operation, especially at early stages in pp. 217–220, Jan. 2013.
[11] A. Collet, J. Crebier, and A. Chureau, “Multi-cell battery emulator for
the development process or during fault simulation. advanced battery management system benchmarking,” in Proc. IEEE Int.
The methodology followed in this work comprises following Symp. Ind. Electron., 2011, pp. 1093–1099.
stages: [12] H. Dai, X. Zhang, X. Wei, Z. Sun, J. Wang, and F. Hu, “Cell-BMS valida-
tion with a hardware-in-the-loop simulation of lithium-ion battery cells for
1) comprehensive review of battery simulation approaches; electric vehicles,” Int. J. Electr. Power Energy Syst., vol. 52, pp. 174–184,
2) development of an advanced HIL simulation battery Nov. 2013.
model; [13] W. C. Lee and D. Drury, “Development of a hardware-in-the-loop sim-
ulation system for testing cell balancing circuits,” IEEE Trans., Power
3) LIB model validation in simulations; Electron., vol. 28, no. 12, pp. 5949–5959, Dec. 2013.
4) BMS testing on HIL simulator, comprising exemplary [14] J. V. Barreras, C. Pinto, R. de Castro, E. Schaltz, S. J. Andreasen, and R. E.
functional, and fault insertion tests. Araujo, “Multi-objective control of balancing systems for Li-ion battery
packs: A paradigm shift?” in Proc. IEEE Vehicle Power Propulsion Conf.,
Regarding the LIB model, a nonlinear dynamic electrother- 2014, pp. 1–7.
mal multicell ECM coupled with an aging model is presented, [15] W. Waag, C. Fleischer, and D. U. Sauer, “Critical review of the methods
taking into account temperature, aging, and SOC dependen- for monitoring of lithium-ion batteries in electric and hybrid vehicles,”
J. Power Sources, vol. 258, pp. 321–339, 2014.
cies. Parameterization is based on experimental data from [16] M. Einhorn, W. Roessler, and J. Fleig, “Improved performance of serially
tests conducted on a commercial 40-Ah high-energy Kokam connected li-ion batteries with active cell balancing in electric vehicles,”
SLPB100216216H pouch cell with NMC cathode. System-level IEEE Trans. Veh. Technol., vol. 60, no. 6, pp. 2448–2457, Jul. 2011.
[17] S. Buller, Impedance Based Simulation Models for Energy Storage De-
components of the battery system are also considered, like tem- vices in Advanced Automotive Power Systems (Aachener Beiträge des
perature sensors or the switch box. ISEA). Maastricht, Germany: Shaker Verlag GmbH, 2003.
With regard to the experimental results, first the LIB model is [18] D. Andre, M. Meiler, K. Steiner, H. Walz, T. Soczka-Guth, and D. U. Sauer,
“Characterization of high-power lithium-ion batteries by electrochemi-
validated using a dynamic load profile from a real battery electric cal impedance spectroscopy. II: Modelling,” J. Power Sources, vol. 196,
vehicle. Then, in order to prove the advanced capabilities of the no. 12, pp. 5349–5356, Jun. 2011.
HIL simulator platform, a set of exemplary functional and fault [19] D. Andre, M. Meiler, K. Steiner, C. Wimmer, T. Soczka-Guth, and
D. U. Sauer, “Characterization of high-power lithium-ion batteries by
insertion tests were performed on a commercial BMS. As a final electrochemical impedance spectroscopy. I. Experimental investigation,”
remark, a comprehensive collection of functional, nonfunctional J. Power Sources, vol. 196, no. 12, pp. 5334–5341, Jun. 2011.
5098 IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 52, NO. 6, NOVEMBER/DECEMBER 2016

[20] N. Shidore et al., “Battery in the loop: Battery evaluation in a systems [45] “Comparison and evaluation of charge equalization technique for series
context,” in Proc. IEEE Trans. Electrification Conf. Expo., 2014, pp. 1–9. connected batteries,” in Proc. IEEE 37th Power Electron., Specialists
[21] H.-S. Song et al., “Verification of battery system model for environmen- Conf., Jun. 2006, pp. 1–6.
tally friendly vehicles using a battery hardware-in-the-loop simulation,” [46] J. Cao, N. Schofield, and A. Emadi, “Battery balancing methods: A
IET Power Electron., vol. 6, no. 2, pp. 417–424, Feb. 2013. comprehensive review,” in Proc. IEEE Vehicle Power Propulsion Conf.,
[22] S. Kermani, R. Trigui, S. Delprat, B. Jeanneret, and T. M. Guerra, Sep. 2008, pp. 1–6.
“PHIL Implementation of energy management optimization for a parallel [47] A. A. Pesaran, “Battery thermal models for hybrid vehicle simulations,”
HEV on a predefined route,” IEEE Trans. Veh. Technol., vol. 60, no. 3, J. Power Sources, vol. 110, no. 2, pp. 377–382, 2002.
pp. 782–792, Mar. 2011. [48] S. Abada, G. Marlair, A. Lecocq, M. Petit, V. Sauvant-Moynot, and
[23] Y. Xiao-kun, H. Hong-wen, P. Lian-yun, and Z. Xiaolin, “Hardware-in- F. Huet, “Safety focused modeling of lithium-ion batteries: A review,”
the-loop simulation on a hybrid power system,” in Proc. 4th Int. Conf. J. Power Sources, vol. 306, pp. 178–192, 2016.
Power Electron. Syst. Appl., 2011, pp. 1–5. [49] T. Veijola, “Electrothermal simulation models for NTC and PTC thermis-
[24] C. Seitl, J. Kathan, G. Lauss, and F. Lehfuss, “Power hardware-in-the-loop tors,” Proc. CSC, vol. 2, pp. 950–955, 1998.
implementation and verification of a real time capable battery model,” in [50] D. Stanković and M. Zlatanović, “A versatile computer controlled mea-
Proc. IEEE 23rd Int. Symp. Ind. Electron., 2014, pp. 2285–2290. suring system for recording voltage-current characteristics of various re-
[25] T. Mesbahi, N. Rizoug, P. Bartholomeus, and P. Le Moigne, “Li-ion battery sistance sensors,” Sens. Actuators A, Phys., vol. 42, no. 1, pp. 612–616,
emulator for electric vehicle applications,” in Proc. IEEE Vehicle Power 1994.
Propulsion Conf., 2013, pp. 1–8.
[26] O. Konig, C. Hametner, G. Prochart, and S. Jakubek, “Battery emulation
for power-HIL using local model networks and robust impedance control,”
IEEE Trans. Ind. Electron., vol. 61, no. 2, pp. 943–955, Feb. 2014.
[27] A. Thanheiser, W. Meyer, D. Buecherl, and H. Herzog, “Design and
Jorge Varela Barreras was born in Vigo, Spain, in
investigation of a modular battery simulator system,” in Proc. IEEE Vehicle
1982. He received the M.Sc. degree in industrial en-
Power Propulsion Conf., 2009, pp. 1525–1528.
gineering from the University of Vigo, Vigo, Spain,
[28] O. Konig, S. Jakubek, and G. Prochart, “Battery impedance emulation
and the M.Sc. degree in enery engineering and the
for hybrid and electric powertrain testing,” in Proc. IEEE Vehicle Power
Ph.D. degree in battery management from Aalborg
Propulsion Conf., 2012, pp. 627–632.
University, Aalborg, Denmark, in 2009 and 2016,
[29] G. Pebriyanti, “A lithium-ion battery modeling for a HIL-battery sim-
respectively.
ulator,” in Proc. Int. Conf. Comput., Control, Informat. Appl., 2013,
In 2010, he co-founded a photovoltaic consulting
pp. 185–190.
company in Spain, serving as a Business Develop-
[30] L. f. Xu, J. q. Li, J. f. Hua, X. j. Li, and M. g. Ouyang, “Hardware in the
ment Manager. Since 2011, he has been a Researcher
loop simulation of vehicle controller unit for fuel cell/battery hybrid bus,”
in Aalborg. In June 2016, he became a Postdoctoral
in Proc. IEEE Vehicle Power Propulsion Conf., 2009, pp. 1777–1782.
Researcher at the University of Oxford, Oxford, U.K. His research interests
[31] C. Park, J. Liu, and P. H. Chou, “B#: A battery emulator and power-
include novel electric vehicle concepts and Li-ion battery modeling, simulation,
profiling instrument,” in Proc. Int. Symp. Low Power Electron. Des., 2003,
emulation, diagnostics, balancing and management systems.
pp. 288–293.
[32] A. Florescu, S. Bacha, A. Rumeau, I. Munteanu, and A. I. Bratcu, “PHIL
simulation for validating power management strategies in all-electric ve-
hicles,” in Proc. 15th Eur. Conf. Power Electron. Appl., 2013, pp. 1–6.
[33] A. Rousseau et al., “Electric drive vehicle development and evalua-
tion using system simulation,” in Proc. 19th IFAC World Congr., 2014, Christian Fleischer studied M.Sc. degree in electri-
pp. 7886–7891. cal and computer engineering with a focus on statis-
[34] A. Thanheiser, T. P. Kohler, C. Bertram, and H. Herzog, “Battery emula- tical signal processing and high performance com-
tion considering thermal behavior,” in Proc. IEEE Vehicle Power Propul- puting from Ohio State University, Columbus, OH,
sion Conf., 2011, pp. 1–5. USA, in 2009. He is currently working toward the
[35] T. Baumhofer, W. Waag, and D. U. Sauer, “Specialized battery emulator Ph.D. degree at Rheinisch - Westfálische Technische
for automotive electrical systems,” in Proc. IEEE Vehicle Power Propul- Hochschule (RWTH) Aachen University, Aachen,
sion Conf., 2010, pp. 1–4. Germany.
[36] D. Bazargan and S. Filizadeh, “Hardware-in-loop real-time simulation of Since 2009, he has been working as a Research
a battery storage system in a wind generation scheme,” in Proc. 3rd Int. Scientist at the Institute of Power Electronics and
Conf. Elect. Power Energy Convers. Syst., 2013, pp. 1–6. Electrical Drives in the Electrochemical Energy Con-
[37] D. Kawashima, M. Ihara, T. Shen, and H. Nishi, “Real-time simulation of version and Storage Systems Group. His research interests include integration of
cooperative demand control method with batteries,” in Proc. 38th Annu. battery management systems, specializing on developing novel adaptive on-line
Conf. IEEE Ind. Electron. Soc., 2012, pp. 3588–3593. state and parameter estimation methods based on probabilistic inference and
[38] W. W. Weaver and G. G. Parker, “Real-time hardware-in-the-loop simu- machine learning algorithms, creation of the software architecture, and integra-
lation for optimal dc microgrid control development,” in Proc. IEEE 15th tion of AUTOSAR components.
Workshop Control Model. Power Electron., 2014, pp. 1–6.
[39] H. Rahimi-Eichi, U. Ojha, F. Baronti, and M. Y. Chow, “Battery manage-
ment system: An overview of its application in the smart grid and electric
vehicles,” IEEE Ind. Electron. Mag., vol. 7, no. 2, pp. 4–16, Jun. 2013.
[40] L. Gao, S. Liu, and R. A. Dougal, “Dynamic lithium-ion battery model Andreas Elkjær Christensen received the M.Sc. de-
for system simulation,” IEEE Trans. Compon. Packag. Technol., vol. 25, gree in engineering from the Technical University of
no. 3, pp. 495–505, Sep. 2002. Denmark, Lyngby, Denmark, in 2009, and the indus-
[41] M. Chen and G. A. Rincon-Mora, “Accurate electrical battery model trial Ph.D. degree, which was a collaboration between
capable of predicting runtime and i-v performance,” IEEE Trans. Energy Lithium Balance A/S, Denmark, and DTU Energy,
Convers., vol. 21, no. 2, pp. 504–511, Jun. 2006. from the Technical University of Denmark in 2016.
[42] M. A. Roscher and D. U. Sauer, “Dynamic electric behavior and open- During his Ph.D., he researched new battery tech-
circuit-voltage modeling of LiFePo4-based lithium ion secondary batter- nology in form of Lithium-air batteries, with special
ies,” J. Power Sources, vol. 196, no. 1, pp. 331–336, 2011. focus on applications of electrochemical impedance
[43] J. V. Barreras, E. Schaltz, S. J. Andreasen, and T. Minko, “Datasheet-based spectroscopy for battery management. This work has
modeling of li-ion batteries,” in Proc. IEEE Vehicle Power Propulsion resulted in two pending patents. His work at Lithium
Conf., Oct. 2012, pp. 830–835. Balance A/S is focused on project management, embedded software develop-
[44] W. F. Bentley, “Cell balancing considerations for lithium-ion battery ment, functional safety in road vehicles, and supervising patent applications and
systems,” in Proc. 12th Annu. Battery Conf. Appl. Adv., Jan. 1997, grants.
pp. 223–226. Dr. Christensen is a Board Member at the Danish Battery Society.
BARRERAS et al.: ADVANCED HIL SIMULATION BATTERY MODEL FOR BATTERY MANAGEMENT SYSTEM TESTING 5099

Maciej Swierczynski received the B.Tech. degree in Søren Juhl Andreasen was born in Aalborg,
computer engineering for industrial applications from Denmark, in 1981. He received the M.Sc.M.E. de-
AGH University of Science and Technology, Krakow, gree with specialization in electromechanical system
Poland, in 2005; the M.Tech degree in computer engi- design, and the Ph.D. degree in design and control
neering for industrial applications and in power elec- of high temperature proton exchange membrane fuel
tronics and drives from Aalborg University, Aalborg, cell system from Aalborg University, Aalborg, Den-
Denmark, in 2007 and 2009, respectively; and the mark, in 2005 and 2009, respectively.
Ph.D. degree with a thesis on Lithium-ion battery en- Since 2014, he has been working as a Senior Sys-
ergy storage system for augmented wind power plants tem Engineer with the Development Department at
from Aalborg University. Serenergy A/S. His work includes research and devel-
He is working currently as an Associate Professor opment related to methanol reformer and high tem-
with Aalborg University. His research interests include energy storage tech- perature PEM fuel cell systems. Products include uninterruptible power supply,
nologies for wind and automotive applications, battery testing, modeling, and auxiliary power unit, and hybrid electric systems for mobile power. In 2012, he
lifetime analyses. worked as an Associate Professor at Aalborg University. His involvements in
research concerned fuel cell and battery power systems.

Dirk Uwe Sauer received the Dr.-rer.-nat. degree

Erik Schaltz was born in Viborg, Denmark, in 1981. from the Faculty of Natural Sciences, University of
He received the M.Sc. and Ph.D. degrees in electrical Ulm, Ulm, Germany, in 2003.
engineering from the Department of Energy Technol- He is currently the Head of the Group for
ogy, Aalborg University, Aalborg, Denmark, in 2005 Electrochemical Energy Conversion and Storage
and 2010, respectively. Systems with Rheinisch - Westfálische Technische
From 2009 to 2012, he was an Assistant Professor Hochschule (RWTH) Aachen University, Aachen,
in the Department of Energy Technology, Aalborg Germany. In 2003, he was appointed as a Junior
University, where he is currently an Associate Pro- Professor and in 2009 he became a Full Professor.
fessor and the Programme Leader of the research Beforehand, he worked for nine years at the Fraun-
programme in E-mobility and Industrial Drives. His hofer Institute for Solar Energy Systems, Freiburg,
research interests include analysis, modeling, design, Germany. His focus is on the integration of renewable energies into grids with
and control of power electronics; electric machines; energy storage devices the support of energy storage systems and on batteries for electromobility. He
including batteries and ultracapacitors; fuel cells; hybrid electric vehicles; ther- is also a Co-founder of the companies P3 Energy Storage GmbH, BatterieInge-
moelectric generators; and inductive power transfer systems. nieure GmbH, and eBusplan GmbH.