IET Electrical Systems in Transportation

Special Issue: Energy Storage and Electric Power Sub-Systems for

Advanced Vehicles (Based on IEEE-VPPC 2014)

ISSN 2042-9738
Disparity in initial and lifetime parameters of Received on 1st March 2015
Revised on 28th September 2015
lithium-ion cells Accepted on 15th November 2015
doi: 10.1049/iet-est.2015.0007
www.ietdl.org

Susanne Lehner 1, 2 ✉, Thorsten Baumhöfer 1, 2, Dirk Uwe Sauer 1,2, 3

1
Electrochemical Energy Conversion and Storage Systems Group, Institute for Power Electronics and Electrical Drives (ISEA),
RWTH Aachen University, Aachen, Germany
2
Jülich Aachen Research Alliance, JARA-Energy, Aachen, Germany
3
Institute for Power Generation and Storage Systems (PGS), E.ON ERC, RWTH Aachen University, Aachen, Germany
✉ E-mail: [email protected]

Abstract: Special section ‘Energy storage and electric power sub-systems for advanced vehicles’: Nowadays automotive
battery systems consist of a vast number of lithium (Li)-ion cells in a serial connection. Next to the operation strategy
especially disparities in the performance parameters of the cells at their initial states and during lifetime have therefore
a huge influence on the overall system performance and its lifetime. Opposing that fact the disparity of cells due to
production process and different boundary conditions is mostly neglected in system design. The cells are rather
considered as even according to parameterisation measurements carried out and lifetime experiments of only a few
cells per test condition. This study will show that the diversification of the cells has to be taken into account during
system layout and packaging in order to optimise the lifetime and performance on system level. Moreover, the impact
of the operation strategy on ageing is analysed more in detail and guidelines are given. In these analyses, the study is
based on elaborate laboratory experiments with a high number of Li-ion automotive cells.

1 Introduction out [11]. By varying, for example, the mean state of charge in cyclic
ageing tests, moreover the operational mode can be optimised by
In the recent years, especially automotive applications ranging from adjusting the state of charge (SoC) window, the pack is later on used
hybrid to full electric vehicles have become a promising market for in [12]. Due to the high amount of parameters that in some cases
lithium (Li)-ion batteries [1] as the new European CO2 emission also influence each other, many different test points are needed. Per
regulations in 2015 and 2020 have to be met by the manufacturers test point usually several cells are tested in order to be able to
[2]. In America, the same development can be observed. The average the results and to exclude cells that are differing widely from
strong focus on Li-ion batteries in that sector can be explained by the rest of the tested cells. Further information on the averaging of
the specific demand. Cycle lifetimes of up to 1000 cycles at 80% the measurement behaviour can be found exemplarily in [13]. Ageing
depth-of-discharge (DoD) have to be reached on cell level for full models as presented by Schmalstieg et al. [11] are parameterised with
electric vehicles [1]. Not all battery technologies are able to fulfil the acquired data and can be used in the following to determine the
such desired deep cycles without tremendously enhancing ageing prime battery for a specific application. These models have in
as shown, for example, in [3] for lead-acid batteries. At the same common that all cells within the system are assumed to perform
time a high calendric lifetime is needed as, for example, the equally so that the cell ageing corresponds directly to the system
average vehicle lifetime is ∼12 years nowadays in Germany. lifetime. In system design this leads to a grouping of cells according
Even though the really used driving range is quite low, as for example to their initial capacity, which is often already done by the cell
95% of all single trips are shorter than 42 km in Germany [4] and the manufacturer selling different cell grades according to the respective
average daily driven mileage in the UK even only around 40 km [5], capacity [14]. However, this approach does not meet the actual
range anxiety of the customers is a big issue. This leads to huge conditions. As shown by the authors in [13, 15], the single cells’
battery systems with an energy content of ∼20 (BMW i3)–85 (Tesla ageing differs independently from the initial capacity resulting in a
Model S) kWh. Due to this, the battery will remain the most expensive premature ageing of the overall system in case of a serial connection
component in the vehicle for the next 10–20 years even though the cell of many cells. The analysis of the production variance of 48 Sanyo
prices have been decreasing tremendously in the last years. Forecasts 18650 cells showed a margin of 350 cycles between the best and the
for 2020 estimate Li-ion cell costs for automotive applications to be in worst cells on reaching the end of life (EOL) criteria of 80% of the
the range of only 77 €/kWh (USCAR)–200 €/kWh (Roland Berger) initial capacity. The average achievable cycles until EOL criterion is
[6]. Therefore, it is mandatory to further optimise the total cost of reached are almost 1100. In this around 4% of the tested cells
ownership by enhancing the battery system’s lifetime and usage in showed a considerably faster than average ageing that led to the wide
order to be competitive to internal combustion engine cars. To get the margin. Only one of these worse cells in a pack with a strictly serial
most out of the battery system it is furthermore possible to take second architecture would lead to a give-away of a wide amount of the
use concepts for grid stabilisation (see [7, 8]) into account if the ageing capacity of the rest of the cells and therefore trigger the need to
can be predicted precisely. oversize the system in order to ensure a sufficient driving range over
Due to this, there are various research activities focusing on the lifetime [12]. This effect is even further enhanced by the boundary
ageing effects itself or on developing new materials prone to a high conditions of the cells in the system, for example, by the
amount of cycles as shown exemplarily by the authors in [9, 10]. To unpreventable temperature gradient among the cells during usage.
characterise and benchmark cells time-consuming test matrices are In this paper, the effects of production tolerances are analysed
conducted considering a lot of different parameters such as mean more in detail. Doing so among other experiments with 700
state of charge, temperature and current rate. Considering the ageing automotive cells will be presented considering in particular the
analysis, calendric as well as cyclic ageing tests have to be carried initial state of the cells. In a second part, the impact of worse cells

IET Electr. Syst. Transp., 2016, Vol. 6, Iss. 1, pp. 34–40

34 & The Institution of Engineering and Technology 2016
 20429746, 2016, 1, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-est.2015.0007 by Faculdade Medicina De Lisboa, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Table 1 Cell data To further comprehend the spread in initial capacity, the
correlation of the capacity to various factors was analysed. In a
rated capacity 5 Ah
electrode materials graphite/Lithium Nickel Manganese first step, the relation of capacity to internal resistance was
Cobalt Oxide (NMC) investigated as shown in Fig. 3. The black squares mark the batch
voltage limits 2.5–4.2 V with the high capacity, whereas the cells of the lower batch are
recommended usable voltage 2.7–4.1 V represented by light grey circles. The overall linear fit has a slope
range
nominal voltage 3.6 V of −0.28 indicating that cells of the higher batch also tend to have
pulse charge current (10 s) ≤150 A a slightly lower internal resistance. This trend can be seen even
pulse discharge current (10 s) ≤220 A stronger for a longer pulse duration as shown in [12].
continuous current ≤50 A Next to the correlation between internal resistance and capacity
also the dependence on the delivery state was analysed. Therefore,
the storage voltage during transport of the cells as well as the
and an enhanced ageing of single cells are discussed showing weight of the cells was measured at inventory upon delivery. As
different options to handle it in system design. shown in [12], the storage voltage had no impact on the clustering
in two batches as the cells were all delivered in an very good
mean storage voltage of 3.61 V with the infinitesimal standard
2 Investigation of initial cell parameters deviation of 5.2 mV. As shown by the slight correlation, the
weight of the cells is smaller than 2.5% spread which is also quite
For a profound analysis of the initial cell parameters, a 5 Ah prismatic similar, yet it can be stated that the cells with a higher weight also
shaped automotive cell, designed for hybrid and plug-in hybrid tend to show a higher capacity (Fig. 4). Regarding the internal
applications was used. Detailed data of the cell provided by a German resistance again no correlation can be found.
vehicle manufacturer can be found in Table 1 giving the cell
parameters from the data sheet. As can be seen, the recommended
usable voltage range is smaller than the end-of-charge and 3 Minimising the initial spread
end-of-discharge limits. This is often done by manufacturers to
prolong the overall cell lifetime by avoiding the most stressful Having seen what a high difference can already be present right from
operation areas close to the full charge and, respectively, full discharge the beginning, it is of interest to minimise the effect on system level.
states. Those more narrow limits were therefore also used in all Most commonly done already at the cell manufacturer is a grading of
experiments explained in the following. Thus, discharging the cell the cells according to their initial capacity. This sorting can further be
from 4.1 to 2.7 V corresponds to a DoD of 100%. enhanced by initial tests at the pack manufacturer. However, a further
Capacity as well as pulse resistance tests were carried at the institute cell selection process at the pack manufacturer is very time consuming
to benchmark each cell at its initial state. All tests were run at an and costly when it relates to the capacity of the cells. Huge testing
ambient temperature of 25°C in a Binder MK 720 temperature capacities would be needed and depending on the cell capacity they
chamber and mark the reference point for all later measurements. have to be able to deliver relatively high currents in order to run the
The cycling was carried out on individual electrical circuits of a tests at an admissible C-rate and therefore speed. In the upper part
Digatron multiple cell formation and testing (MCFT) battery tester, of Fig. 5, the histogram of 80 cells of type 18650 with a nominal
allowing to test up to 50 cells at a time. As test, routine capacity voltage of 2.05 Ah is shown. The cells are A graded cells of the
tests at different C-rates were applied, followed by a pulse test in manufacturer with an additional pre-selection process excluding the
charge and discharge directions. ones with a slightly lower initial capacity. The resulting average
A histogram of the initial capacity at 1C and the fitted normal capacity is 2084 mAh with a standard deviation of 12.97 mAh. This
distribution of the cells are shown in Fig. 1. Fig. 1a shows shows that a quite low initial spread is achievable for these cells.
the distribution among all tested cells. It can be seen that the Another option when it comes to pack manufacturing is to use
distribution shows not the expected Gaussian distribution but rather a smaller cells in a parallel connection instead of large format Li-ion
bimodal one. The highest measured capacity is 5.27 Ah, whereas the cells. This concept is, for example, favoured by Tesla using an
lowest is 4.83 Ah which is clearly below the nominal capacity of the architecture with 74 cells in parallel and 96 of these resulting
cell. The total average is 5.06 Ah with a standard deviation of 119.5 modules in series for the 85 kWh battery pack. The lower part of
mAh. The clustering could be observed also for other current rates Fig. 5 shows a similar concept of modules with 26 cells in
(0.1C, 0.5C, 4C). Regarding the different current rates, the bimodal parallel. The cells used are the same ones as in the upper part of
shape was the least for the lowest current rate, which can be the figure. No further sorting process or initial measurements have
explained by the higher influence of measurement errors at that low taken place before the module construction. As can be seen for the
excitation rates. Due to these results, it was decided to cluster the histogram of the 320 modules a normal distribution was achieved.
cells in two batches. The higher batch includes all cells with at least The average capacity of the modules is 54.06 Ah with a standard
the nominal capacity, whereas the second cluster consists only of the deviation of only 0.28 Ah. The standard deviation is therefore only
cells with a capacity below the nominal one [12]. Within the clusters 0.5% of the nominal capacity of the modules and therefore even
the mean values are 5.14 and 4.90 Ah and the standard deviation was lower than in the pre-selection process. With an even higher
reduced to 52.2 and 28.8 mAh, respectively. parallelisation, this effect can be even enhanced as the capacity of
The initial pulse resistance tests were performed with a pulse of 4C the parallel module is
and a pulse length of 20 s at steps of 10% SoC each. Having seen the
strong clustering in the initial capacity one might expect to find a

nparallel
similar behaviour also in the initial resistance. However, no clustering Qmodule = Qcell,i (1)
can be observed as shown in Fig. 2. Being a high-power cell, the i=1
mean internal resistance is with 1.89 mΩ really low and shows a
standard deviation of 0.065 mΩ. On analysing the cells separately for and therefore the spread in average capacity will statistically decline
the clusters based on capacity, only a slightly lower mean value can be and the more cells are used in parallel. Hence, for a high amount of
found for the upper cluster with 1.88 mΩ and a standard deviation of cells connected in parallel, the resulting module capacities would
0.066 mΩ. The mean value for the lower cluster remains the same converge to the mean value.
with a standard deviation of 0.060 mΩ. The overall spread between
highest and lowest resistance is 17%. In these considerations regarding
the internal resistance error values have to be kept in mind, as the 4 Influence of topology on lifetime
accuracy is 0.2% of the final value for the current and voltage
measurements and due to the small internal resistance of the cell it’s As shown in [13, 15, 16], ageing behaviour of the single cells on
contacting has a high impact on the overall resistance. pack level shows a strong deviation and cannot simply be foreseen

IET Electr. Syst. Transp., 2016, Vol. 6, Iss. 1, pp. 34–40

& The Institution of Engineering and Technology 2016 35
 20429746, 2016, 1, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-est.2015.0007 by Faculdade Medicina De Lisboa, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Fig. 1 Histogram of the initial capacity [12]
a Complete cell batch
b Cells with an initial capacity greater than or equal to nominal capacity
c Cells with an initial capacity less than nominal capacity

by the initial capacity of the cells. Hence, grading of the cells by their The shape parameter β describes the failure mode of the battery.
initial capacity as currently done by the manufacturers has only a A shape parameter of 0.5 accounts for infant mortality, for
limited positive influence on the overall lifetime of the system. example, due to poor production. A shape parameter of 1 leads to
This leads nowadays to oversized packs in order to ensure a high a time independent probability (exponential probability) and
enough capacity over the complete product lifetime. Therefore, in therefore represents random failures. Higher shape parameters are
that section the influence of the topology on the statistical product caused by ageing and are therefore of interest in this analysis. The
lifetime shall be analysed. The reliability of Li-ion cells can be scale parameter η accounts for the compression of the time axis
modelled best by a Weibull distribution as proposed in [17, 18]. and gives at the point in time t = η the characteristic lifetime of a
The two-parameter Weibull distribution is a generalisation of the product, which means that 63.2% of the items have failed.
exponential function that enables the description of ageing A typically scaled Weibull plot is given in Fig. 5. An ageing
processes with time variant ageing rates what makes it a good fit behaviour prone to Weibull fitting would result in a linear shape
for batteries. The corresponding failure rate l(t) and the of the measurement values in it, as the y-axis is scaled by ln(ln(1/
cumulative distribution function F(t) can be described by the two 1−y(t)) and the x-axis is plotted logarithmically as well to linearise
formulas it. In the figure, the results of an ageing test with 48 equal cells of
type 18650 are shown. The cells were aged in a temperature
  chamber at 25°C and a current rate of 2C with a cycle depth of
b t b−1 60% with a mean SoC of 50%. The ageing experiment resulted in
l(t) = (2)
h h a spread in cyclic lifetime between best and worst cells of 33% of
the mean cyclic lifetime which indicates how strong the deviation
b
is. The black crosses in the figure mark the failure probability
F (t ) = 1 − e−(t/h) (3) calculated of the measurement values by the Bernard’s

Fig. 2 Histogram of the initial internal resistance after a 2 s pulse

IET Electr. Syst. Transp., 2016, Vol. 6, Iss. 1, pp. 34–40

36 & The Institution of Engineering and Technology 2016
 20429746, 2016, 1, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-est.2015.0007 by Faculdade Medicina De Lisboa, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
to the different ageing effects [20]. This is backed up by the
characteristic shape of the failure probability with clearly visible
knees. Each knee can be interpreted as a change of the major
contribution to the overall failure probability.
To reach a better fit, a mixed Weibull distribution consisting of
several independent Weibull distributions [21] is proposed
modelling the different ageing effects. This mixed Weibull
distribution

n

n  
FM (t ) = mi F i ( t ) = mi 1 − e−(t/hi )bi (5)
i=1 i=1

is a sum of the weighed subdistributions. In Fig. 6, the dark blue line

shows the fitted mixed Weibull distribution consisting of three
separate Weibull functions. Obviously, it matches the actual failure
probability derived from the measurements quite good which
shows that a mixed Weibull distribution is a good approach to
further improve reliability models for batteries such as in [17, 22].
The derived fit can be used for further reliability and lifetime
Fig. 3 Correlation of initial capacity to internal resistance modelling for the overall system.
To explore the influence of cell failures in different topologies,
simulations for a small electric vehicle were carried out. Doing so,
approximation for median ranks an energy content of 17 kWh was estimated based on the average
battery size of currently available vehicles such as the Nissan Leaf,
  p − 0.3 the BMW i3, the Mitsubishi iMiEV, the Smart ED and the VW
Fp tp = (4) e-up!. With an average driving range of 100 km and an average
N + 0.4
daily mileage below 42 km in Germany, a full cycle per 2.2 days
can be estimated. The left hand side of Fig. 7 shows the EOL
with the failure p out of N units. The Bernard’s approximation is probability for a completely serial connection of cells to build
especially proposed for test sets with <50 specimens. At higher such a pack. The energy content of all topologies is the same. In
numbers p/N + 1 can be used instead. However, the black crosses that cell sizes between 20 and 70 Ah with a step size of 5 Ah were
show a behaviour clearly diverging from a linear line. A fitted used and the same ageing rates and spread on cell level as derived
Weibull distribution as explained above with the parameters η = beforehand in the experiment for the 18650 cell (introduced in
1190.62 and β = 11.2922 leads to the fitting curve in light blue Sections 3 and 4 with Fig. 6) were assumed. The used cell sizes
(dashed line). It can be seen that the behaviour differs heavily to correspond also to cell sizes used in currently available battery
the points Fp(tp) especially in the probability of early failures. electric vehicles (BEVs) and lead to a nominal system voltage
This can be explained by the complex ageing behaviour of cells. between 266 and 925 V in order to achieve the desired energy
In general, there are a lot of different ageing and degradation content. At this point it has to be clarified that the failure rate on
mechanisms taking place in a cell. Those processes are:

† Degradation of the active materials at positive and negative

electrodes.
† Degradation of the electrolyte and formation of passive layers.
† Changes to the electrode structure.

Detailed information on these processes can be found among

others in [1, 19]. These processes are highly dependent on a lot of
external parameters such as charging rate, temperature, cycle depth
and mean state of charge and there are moreover interdependencies
among these factors. This leads to the conclusion that the overall
failure probability can be divided into different subpopulations due

Fig. 5 Comparison of spread on cell level for pre-sorted cells and spread in
Fig. 4 Correlation of initial capacity to weight of the cells modules of 26 cells (all type 18650)

IET Electr. Syst. Transp., 2016, Vol. 6, Iss. 1, pp. 34–40

& The Institution of Engineering and Technology 2016 37
 20429746, 2016, 1, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-est.2015.0007 by Faculdade Medicina De Lisboa, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
cell level will not be the same for all cell types but strongly depend
on the manufacturer, the cell geometry as well as the usage scenario.
Therefore, in system design, a detailed benchmark of different
possible cells has to be carried out in order to identify the most
promising cell for an application. Moreover, a higher cycle
lifetime of automotive cells than for the considered 18650 cell is
usually found. However, to compare solely the influence of the
topology on the statistical EOL probability, these influences can be
neglected. Next to the failure probability of the cells also the
influence of the battery management system (BMS) as well as the
interconnection of cells and the connection to the monitoring unit
were calculated during the simulation. The failure probability of
the single monitoring channels was calculated by the exponential
probability distribution and the layout of the Nissan Leaf BMS
was used as a reference in order to model the monitored cells per
cell controller chip, which is four in that case. In the
interconnection of cells and connection to the monitoring unit, a
welded connection was estimated due to its high penetration in
market for mass production. Data to model failure rates of the
electronics and interconnections were taken from MIL-217.
Fig. 7 Comparison of EOL probability for different string lengths (cell
In the figure, it can be seen that especially for higher voltages
capacity 20–70 Ah) assuming the same failure probability on cell level. In
already after a relative moderate usage time of 2 years (≃17.500 h)
the right hand side the influence of the cells, the connections and the
the probability of EOL is already above 15% for the packs
monitoring unit on the overall failure probability are shown
consisting of 20 Ah cells. This is mainly caused by the impact of
the BMS as well as the connections as can be seen in the right
hand side of the image for the 20 and 70 Ah cell designs. At the
same time these systems also have in general a lower lifetime demand can be fulfilled. This has the advantage that no additional
expectancy as it becomes also more likely to have a bad cell parts are required but comes at the drawback of a higher weight
within the string limiting the overall system performance as can be and also need for additional space. Instead of oversizing the cells
seen by the failure probabilities of the cells in the right subfigure. it is also possible to shift the energy between the cells via an
Therefore, the strong disparity in ageing results also in the need to active converter [23]. This so called active balancing evens out the
take it into account in system design. Without that, a huge amount of spread between the cells at the drawback of higher costs and
the cells’ capacity within a serial connection could not be used over a complexity than passive balancing.
long period of the vehicle lifetime as the worst cell in the string A promising method proposed by Baumhoefer et al. [13] is to
determines the usable capacity of the whole string. There are select the cells by several initial performance parameters. It was
several possibilities to deal with that matter which shall be shown for an 18650 Sanyo cell that this works fine needing only a
addressed in the following. limited number of attributes which could be determined relatively
In the easiest approach, the used cells are simply oversized for the easily during a later on mass production process. As extensive
specific application so that also with an ageing spread the energy measurements and a big data analysis are necessary to identify
these criteria and they cannot simply be transferred to another cell
type, it is yet a time-consuming method.
Finally, the already explained parallel connection on cell level can
be considered. In Section 3, it was already shown that the initial
spread in the string can be successfully limited by these parallel
modules instead of using larger cells with a comparable capacity
in the string. This in general is also true for the ageing in the
parallel connection. As shown by Gogona et al. [24] it is in that
case mandatory at least for Li iron phosphate cells to go for even
initial internal resistances due to their flat voltage curve. In case of
an internal resistance mismatch, the less resistive cell would

Fig. 8 Lifetime probability for the considered 17 kWh battery system with a
Fig. 6 Cumulative probability of a cyclic ageing experiment with cells of string length of 103 cells for different parallel-series architectures on module
one batch at same test conditions level

IET Electr. Syst. Transp., 2016, Vol. 6, Iss. 1, pp. 34–40

38 & The Institution of Engineering and Technology 2016
 20429746, 2016, 1, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-est.2015.0007 by Faculdade Medicina De Lisboa, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Fig. 9 Influence of a single cell fault in a module of parallel cells

experience a higher current during charging. By this it would serial cells. It results in a failure rate of 1.55 · 10−15 for the above
approach the end of charge state sooner than the higher resistive discussed module and therefore almost negligible. Yet, one has to
one resulting ultimately in an increase in current in the cell with bear in mind at the same time that the string capacity is reduced
the higher internal resistance and the higher currents to a further by the capacity of the failed cell. Hence, with each cell failure the
enhancement of the ageing process in all cells. capacity and overall lifetime is reduced tremendously and therefore
To highlight the influence of the parallel-series topology, the a failure of several cells within a module should not happen
simulation for one of the series connections considered in Fig. 7 during real operation. This can also be seen in Fig. 9 showing
was carried out for different parallelisation rates on cell level. As exemplarily the ageing behaviour of one of the modules. At a
most of the common vehicles feature a nominal voltage around remaining capacity of 90% a single cell fault is assumed. This
400 V resulting from a serial connection of around 100 cells, it corresponds to a decline of the module capacity of almost 2%. The
was used as reference scenario (1p103s). The original cell size is further ageing can therefore be assumed to approach as if
50 Ah. Therefore, modules consisting of 2 (25 Ah), 4 (12.5 Ah), 5 the module is already further aged, which is shown by shifting the
(10 Ah), 10 (5 Ah) and 20 (2.5 Ah) cells in parallel were ageing graph further to the left.
considered resulting in the scenarios 2p103s (206 cells with 25 Ah
each), 4p103s (412 cells with 10 Ah each), 5p103s (515 cells with
10 Ah each), 10p103s (1030 cells with 5 Ah each) and 20p103s
(2060 cells with 2.5 Ah each). In Fig. 8, the resulting lifetime
5 Conclusion and outlook
probabilities are given. It is obvious that the higher the
In this paper, the initial electrical tests of a large cell batch were
parallelisation the higher the expectable lifetime will be. While the
statistically evaluated and correlations between internal resistance
completely serial module has a probability of 20% to reach its
and capacity as well as the cell weight and storage voltage were
EOL criteria after ∼4.5 years this is true for the 20p103s module
investigated. This statistical evaluation was enabled by the high
only after 6.3 years. This is caused by the averaging among the
amount of 700 cells in the experiment, which is usually not the
ageing of the single cells within the module making it less likely
case in battery analysis. A slight trend to lower internal resistances
to have a distinctly worse module than the average within the
for cells with higher capacity was found. However, by grouping
string. This can also be seen by the higher steepness of the
the cells in the two clusters it could be shown that there is no
probability curves for a high parallelisation rate.
significant dependence between the two values within the clusters.
Additionally, these parallel modules offer superior reliability as a
It was analysed how this initial spread can be minimised and
single cell fault must not directly lead to a fault of the overall
furthermore the additionally diversification of ageing speed be
module. The impact of these spontaneous failures was so far
minimised on system level. For this, several different options such
neglected in the introduced results and would further add to the
as active balancing or modules of parallel cells were discussed and
likeliness of an early EOL. For single cell failures rates of p = 1
the statistical representation of serial connections was given. It was
ppm (parts per million) can be taken as a realistic assumption
shown that for a reliable system with a satisfying lifetime more
already for faulty production batches as can be seen, for example,
complex topologies as pure series connections are worth a further
by the product recall of Dell in 2006. About 4.1 million notebooks
consideration.
were recalled due to 30 battery incidents [25]. Other sources, such
In the following it is important to widen the data basis for ageing
as [26], claim the failure rate for 18650 cells in general to be in
experiments on system level both on experimental as well as field
the range of 0.2 ppm. Considering the modules of Section 3, a
tests. For different usage scenarios, the impact of the single cell
failure probability on module level can be calculated with this
ageing process should be further investigated and modelled more
failure rate for a single cell failure. The modules are designed for a
profound using distribution functions that were introduced in this
vehicle application with a peak discharge rate of 5.8C on cell
paper. Doing so, especially the more complex topologies
level. However, the cells are able to bear a peak current of 7C for
averaging cells behaviour over several cells in a parallel
5 s. In the application always two of the shown modules are
connection needs to be statistically investigated and further
connected in parallel leading to 52 cells in parallel. However, 44
components and effects need to be included in the considerations.
cells would already be sufficient for the peak pack current.
Therefore, the probability of a total failure on module level can be
calculated by the binomial distribution
6 Acknowledgments
nt 


n  n −k
P(s) = 1 − p
pk 1 − p p (6) The basis of this work has been performed in the framework of the
k
k=0 research initiative ‘Drive Battery 2015’ funded by the German
Federal Ministry for Economic Affairs and Energy. Responsibility
with nt being the number of tolerable cell faults and np the amount of for the content of this publication lies with the authors.

IET Electr. Syst. Transp., 2016, Vol. 6, Iss. 1, pp. 34–40

& The Institution of Engineering and Technology 2016 39
 20429746, 2016, 1, Downloaded from https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-est.2015.0007 by Faculdade Medicina De Lisboa, Wiley Online Library on [05/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
7 References 13 Baumhoefer, T., Bruehl, M., Rothgang, S., et al.: ‘Production caused variation in
capacity aging trend and correlation to initial cell performance’, J. Power
Sources, 2014, 247, pp. 332–338
1 Vetter, J., Novák, P., Wagner, M.R., et al.: ‘Ageing mechanisms in lithium-ion
14 Rothgang, S., Baumhöfer, T., Van Hoek, H., et al.: ‘Modular battery design for
batteries’, J. Power Sources, 2005, 147, pp. 269–281
reliable, flexible and multi-technology energy storage systems’, Appl. Energy,
2 Commission of the European Communities: ‘Results of the review of the
community strategy to reduce CO2 emissions from passenger cars and 2015, 137, pp. 931–937
light-commercial vehicles’, 7 July 2007 15 Paul, S., Diegelmann, Ch., Kabza, H., et al.: ‘Analysis of ageing inhomogeneities
3 Schaeck, S., Stoermer, A.O., Koehler, L., et al.: ‘Lead-acid batteries in in lithium-ion battery systems’, J. Power Sources, 2013, 239, pp. 642–650
micro-hybrid applications. Part I. Selected key parameters’, J. Power Sources, 16 Lih, W.-C., Jieh-Hwang, Y., Fa-Hwa, S., et al.: ‘Second use of retired lithium-ion
2011, 196, pp. 1541–1554 battery packs from electric vehicles: technological challenges, cost analysis and
4 Follmer, R., Gruschwitz, D., Jesske, B., et al.: ‘Mobilität in deutschland: struktur – optimal business model’. Int. Symp. on Computer, Consumer and Control,
aufkommen – emissionen – trends’. DLR, INFAS, Berlin, Bonn, Ergebnisbericht, Taichung, Taiwan, 2012, pp. 381–384
2010 17 Li, R., Wu, J.-F., Wang, H.-Y., et al.: ‘Reliability and failure analysis of lithium
5 United Kingdom Department for Transport: ‘Collection – Transport Statistics Great iron phosphate batteries’, Inf. Sci., 2014, 259, pp. 359–368
Britain’, July 2014. Available at https://www.gov.uk/government/collections/ 18 Eom, S.-W., Kim, M.-K., Kim, I.-J., et al.: ‘Life prediction and reliability
transport-statistics-great-britain assessment of lithium secondary batteries’, J.Power Sources, 2007, 174,
6 Schein, H., Holl, K.: ‘Vorstellungen und Erwartungen aus Sicht der Zellhersteller’. pp. 954–958
Batterieforum Deutschland, Berlin, Germany, 2014 19 Methekar, R.N., Paul, W.C.N., Chen, K., et al.: ‘Kinetic Monte Carlo simulation of
7 Zhou, C., Qian, K., Allan, M., et al.: ‘Modelling of the cost of EV battery wear due surface heterogeneity in graphite anodes for lithium-ion batteries: passive layer
to V2G application in power systems’, IEEE Trans. Energy Convers., 2011, 26, formation’, J. Electrochem. Soc., 2011, 158, pp. 363–370
pp. 1041–1050 20 Nelson, W.B.: ‘Applied life data analysis’ (John Wiley & Sons, New Jersey,
8 Peterson, S.B., Apt, J.F., Withacre, J.: ‘The economics of using plug-in hybrid USA, 2004)
electric vehicle battery packs for grid storage’, J. Power Sources, 2010, 195, 21 Elmahdy, E.E., Aboutahoun, A.W.: ‘A new approach for parameter estimation of
pp. 2377–2384 finite Weibull mixture distributions for reliability modeling’, Appl. Math. Model.,
9 Dippel, Ch., Krüger, S., Kraft, V., et al.: ‘Aging stability of Li2FeSiO4 polymorphs 2013, 37, pp. 1800–1810
in LiPF6 containing organic electrolyte for lithium-ion batteries’, Electrochim. 22 Suresh Kumar, E., Sarkar, B.: ‘Improvement of life time and reliability of battery’,
Acta, 2013, 105, pp. 542–546 Int. J. Eng. Sci. Adv. Technol., 2012, 2, pp. 1210–1217
10 Ecker, M., Nieto, N., Käbitz, S., et al.: ‘Calendar and cycle life study of Li 23 Gallardo-Lozano, J., Romero-Cadava, E., Milanes-Montero, M.I., et al.: ‘Battery
(NiMnCo)O2-based 18650 lithium-ion batteries’, J. Power Sources, 2014, 248, equalization active methods’, J. Power Sources, 2014, 246, pp. 934–949
pp. 839–851 24 Gogona, R., Pinson, M.B., Bazant, M.Z., et al.: ‘Internal resistance matching for
11 Schmalstieg, J., Kaebitz, S., Ecker, M., et al.: ‘A holistic aging model for Li parallel-connected lithium-ion cells and impacts on battery pack cycle life’, J.
(NiMnCo)O2-based lithium-ion batteries’, J. Power Sources, 2014, 257, Power Sources, 2014, 252, pp. 8–13
pp. 325–334 25 DELL: ‘Product Bulletin: Sony Battery Recall’, February 2015. Available at https
12 Rothgang, S., Baumhöfer, T., Sauer, D.U.: ‘Diversion of aging of battery cells in ://www.dellproduct.com/Program.aspx?PI=GNbVEXZPZu8=
automotive systems’. IEEE Vehicle Power and Propulsion Conf. (VPPC), 26 Yount, L.: ‘BMS-centered battery safety and reliability’. Battery Safety, Boston,
Coimbra, Portugal, 2014 USA, 2010, pp. 236–268

IET Electr. Syst. Transp., 2016, Vol. 6, Iss. 1, pp. 34–40

40 & The Institution of Engineering and Technology 2016