Content uploaded by Alexander Reiter
Author content
All content in this area was uploaded by Alexander Reiter on May 08, 2023
Content may be subject to copyright.
Model-based predictive maintenance for
Li-ion battery systems
1,2,3Alexander Reiter, 2Stefan Brand, 2Christian Rosenm¨uller, 1Susanne Lehner, 2Oliver Bohlen,
3Dirk Uwe Sauer
Motivation
Methods of predictive maintenance for large-scale bat-
tery systems allow the early detection of fault poten-
tials and the consequent replacement or repair of faulty
components before severe compromises to system
health, safety or performance must be made. While
numerable methods exist for the detection of short-
term faults such as thermal runaways or low-ohmic
short circuits, predicting long-term errors, which de-
velop and intensify over the scope of multiple months
or years still proves difficult. This work presents a
battery system model suited for the model-based pre-
diction of such long-term faults. It incorporates the
effects of cell-to-cell variations (CtCVs) and there-
fore allows the calculation of voltage and temperature
distributions in the system. By comparing the mea-
sured temperature and voltage spread (∆Vcells(t) =
max(Vcells(t)) −min(Vcells(t))) of the potentially
faulty system to their simulated equivalents, long-term
errors such as internal micro short circuits or accel-
erated aging can be detected. In addition to the
model design, the development of methods for predic-
tive maintenance also requires synthetic testing and
validation data, since faults in battery systems usually
only occur sporadically and do not allow a method-
ical investigation. For this purpose, a modular test
bench, which allows the insertion of faults with a de-
fined intensity was used to generate experimental data
showing the influence of high-ohmic short-circuits, in-
creased cell connection resistances and accelerated ag-
ing processes on the system. By using this data new
methods for predicative maintenance can be methodi-
cally developed and evaluated.
Application &
operator
Algorithms
Physical systemLoad profile
Parallel model
Operation
strategy
Measures of predictive maintenance
Measured
signals
Simulated
signals
Comparison
Trend detection
Prognosis
< > = ~
Figure 1: Concept of model-based predictive maintenance
Authors and Contact
1Technology Development, MAN Energy Solutions
SE, 86153 Augsburg, Stadtbachstraße 1, Germany
2Institute for Sustainable Energy Systems (ISES),
University of Applied Sciences Munich, 80335 Munich,
Lothstraße 64, Germany
3Chair for Electrochemical Energy Conversion and
Storage Systems, Institute for Power Electronics and
Electrical Drives (ISEA), RWTH Aachen University,
52074 Aachen, Campus-Boulevard 89, Germany
Corresponding author:
Alexander Reiter: alexander.reiter@man-es.com
Electro-thermal battery system modeling
The battery system model allows the calculation of the average voltage
and temperature as well as the distributions (∆Vcells and ∆Tcells) within
the system caused by cell-to-cell variations. A detailed description of the
model is given in [1]. In the following, the key aspects are highlighted:
Electrical cell model with Rs+ 2 RC impedance string and
zeroth-order hysteresis model.
Electrical system model with individual ∆R,∆Cand SoCinit for
every cell + statistical model simplification using parameter
distributions for the calculation of confidence intervals.
0D thermal cell model with Joules and entropy heat generation.
Thermal system model for a commercial Li-ion battery module.
Electro-thermal battery model
Parameter
distribution
System
structure
Figure 2: Electro-thermal battery system model
Modular battery system test bench
Resistance-optimized
cell connectors
Li-ion cells (21700-type)
6s3p configuration
Easy cell swap
Figure 3: Modular battery system test bench
Developing reliable methods for predictive maintenance requires
extensive testing and validation against experimental data of sys-
tems in faulty condition. Since this data basis cannot be provided
by productive battery systems in the field, a modular battery sys-
tem test bench was designed. The test bench consists of 21700-
type cylindrical battery cells in 6s3p configuration with individual
voltage, current and temperature measurement. The system al-
lows the fast and simple exchange of individual cells as well as
the insertion of parallel and serial resistances. By this, the ef-
fects of accelerated aging of individual cells, increased contact
resistances and high-ohmic internal or external short-circuits can
be evaluated. Also, the fault intensities (e.g. the resistance of
the short-circuit) can be systematically varied. This allows the
generation of a data basis for the development of algorithms for
predictive maintenance.
Experimental data generation
During the experimental data generation, accelerated cell aging, increased connection resistances and high-ohmic short
circuits were considered as typical long-term fault scenarios [2]. The faults were replicated via additional serial resistors
(connection resistances), parallel resistors (external short circuit) or by exchanging one cell of the module by an aged cell.
The fault scenarios comprising different fault intensities (different resistances and SoHs) were evaluated for a 24 h load
profile representing stationary grid applications [3].
0 5 10 15 20
Time in h
-8
-4
0
4
8
System current in A
Load profile
Figure 4: Experimental data generated from modular test bench. Top left: Load profile used during data generation, Top right: Voltage
distribution for external short circuit at cell 3, Bottom left: Voltage distribution for increased connection resistance at cell 3, Bottom right:
Voltage distribution for accelerated aging at cell 3.
Further Work
In following work, the developed test bench will be used to create an exhaustive data matrix for the named faults in various
intensities. Also, the integration of additional faults will be considered. In the next step, the collected data will be used
to parameterize and validate a battery system model representing the faults. This would allow to replace the still rather
demanding measurements on the test bench by a simple simulation. For this approach, different positions of faulty cells as
well as cross-combinations of multiple faults at once shall also be investigated. In the last step, the measured and simulated
data sets will be used for the development of algorithms for the predictive maintenance of large-scale battery systems.
[1] A. Reiter, S. Lehner, O. Bohlen, and D. U. Sauer, “Electrical cell-to-cell variations within large-scale battery systems — a novel
characterization and modeling approach,” Journal of Energy Storage, vol. 57, p. 106 152, 2023.
[2] M.-K. Tran and M. Fowler, “A review of lithium-ion battery fault diagnostic algorithms: Current progress and future challenges,”
Algorithms, vol. 13, no. 3, 2020.
[3] D. Kucevic, B. Tepe, S. Englberger, et al., “Standard battery energy storage system profiles: Analysis of various applications for stationary
energy storage systems using a holistic simulation framework,” Journal of Energy Storage, vol. 28, p. 101 077, 2020.
