Battery Design based upon Life Cycle Statistics
E. Chiodo*, D. Lauria*, V. Fabrizi**, F. Ortenzi**, V. Sglavo**
*Department of Electrical Engineering and Information Technology, University of Naples, (Italy),
** ENEA National Agency for New Technologies, Energy and Sustainable Economic Development (Italy)
Keywords: Lead-acid Batteries, life cycle estimation,
electric vehicles, experimental measurements, probability
distributions.
Abstract
Battery Life Cycle modelling and estimation are key
challenges in the modern electric systems, also due to the
development of renewable energy and electrified
transportation systems exploitation: in such fields indeed
the battery has emerged as the most prominent energy
storage device, attracting a significant amount of studies.
It is however well known that battery lifetime depends
on many parameters, such as power density, specific
power, energy density, operating environment, etc., so
that an accurate methods to analyze this dependence, also
taking into account the randomness of the above
parameters in real operating conditions, is a necessary
even if difficult task. In this framework, in the paper,
after a thorough statistical data analysis, a probabilistic
method for battery design is proposed which ensures the
optimization of a suitable cost function which has the
expected lifetime as a basic input.
The method takes properly into account, in particular, the
random variations in specific power experienced by a
lead-acid battery. The statistical features of lifetime are
also explored by means of a large series of numerical
simulations based upon suitable probability distributions
which may characterize the above operating conditions,
in order to obtain an efficient estimation of the above
lifetime distribution.
1 Introduction
Batteries, conceived as a proper number of cells
connected in series or in parallel, have received since
their discovery a great attention by many researchers and
recently a further emphasized interest can be registered
[1-12], due to the their high potentialities as electrical
source for electric vehicles, as storage systems in smart
grid environment and naturally as support technology for
communication systems. The attractive characteristics of
this kind of electric power source has to satisfy the
required constraints of reliability, so avoiding premature
failure with consequent reduction of the battery cycle-
lifetime, which in the paper is also denoted as “life cycle”
(LC), as measured by the charge -discharge cycles.
A significant amount of studies have been recently
devoted find more and more reliable solution to such
problem. In previous papers [9,10] some of the authors
proposed a probabilistic approach based upon the
Rakhmatov and Virudhula battery model [7-8] with
random input for describing the battery behavior under
uncertainties. Interesting results were obtained in deriving
the probabilistic density function of the time to failure.
More specifically it was demonstrated that, under mild
assumptions, the hitting time of a prefixed barrier voltage
can be described by an Inverse Gaussian distribution. The
probabilistic approach has the intrinsic advantage to
properly handle the stochastic nature of the different
involved parameters, which in that case was dealt with
for system operation or forecasting purpose. As for
concerns battery design, one of the most crucial challenge
is the choice of the most convenient model for solving the
problem of the battery size design over a prefixed time
horizon of interest. Many approaches have been proposed
in the relevant literature, but a lot of difficulties persist,
being related to the incidence of many factors which do
not allow to derive a quite general modelling able to
describe in exhaustive way batteries performances under
different operating conditions. It is well known that
battery lifetime, or life cycle (LC, as measured by the
charge -discharge cycles) depends on many parameters,
such as power density, specific power, energy density,
operating environment, etc. Accurate methods to analyze
this dependence, also taking into account the randomness
of the above parameters in real operating conditions, are
necessary: it has been indeed shown that parameter
uncertainties may lead to significant variations in life
cycle estimation, so that large differences are observed
between the expected and experienced battery LC. In
particular, concerning electric vehicle applications, in a
significant number of cases an unexpected degradation of
battery performances has been measured with respect to
manufacturers claims on battery lifetime. In this
framework, in the paper, after a thorough statistical data
analysis, a probabilistic method for battery design is
proposed which ensures with a high degree of probability
that the LC is higher than a prefixed level provided by
standards which will be desirably issued in the near
future, following studies like the one here proposed.
The method takes properly into account, in particular, the
random variations in specific power experienced by a
lead-acid battery (with 585 Ah of capacity) during its
operation.
The input data came from an experimental campaign
done in the Italian Research Centre of Enea Casaccia with
an electrical bus (Tecnobus Gulliver U520) with a storage
module of 72 V and 585 Ah of capacity serving an on-
demand transport service. The bus has been equipped
with an automotive on-board device collecting GPS data
and vehicle parameters.
The proposed approach combines experimental data
and a Monte Carlo procedure, allowing to identify the
battery size the minimum cost corresponds to, by keeping
into account the replacement cost over a quite extended
time interval.
At this aim a first preliminary statistical analysis is
performed with respect to the experimental data acquired
and elaborated in the Italian Research Centre of Enea
Casaccia. More specifically, the probability density
functions of the maximum power and the energy
requested are derived. Hence a cost function is identified,
by taking into account the replacement problem, which is
strictly related to the lifetime. The lifetime prediction is
based upon the accelerated test which without any doubt
provide the core of the lifetime model. However, a
correction of the interpolation model needs, for taking
into account the unavoidable degradation terms related to
the ageing effects which cannot be wholly observed in
data corresponding to accelerated test. The nonlinearity
of the problem does not allow to derive a closed form of
the optimum size, but the probabilistic procedure is very
feasible to implement and could be employed in a quite
general way. It is trivial to observe that for the
implementation effectiveness an experimental
preliminary campaign is needed, but this allows to
guarantee a certain superiority with respect to a wholly
theoretical design procedure, involving many parameters
which are practically impossible to determinate or to
estimate.
The paper is organized as follows. In section II the
experimental set is discussed, putting in evidence the
various step for obtaining the measurement data. In the
section III, a new battery design methodology is
proposed. In the last section the design approach is
numerically shown with respect to the measurement data
at own disposal.
parameters such as “design” specific power, as below
described.
Within a project denoted as "QUIBUS", an experimental
acquisition campaign has been made in order to
characterize the behavior of an electric bus during an on-
demand transport service.
The equipment installed on-board was devoted to acquire
a series of information needed for our subsequent
analysis: the instantaneous speed of the vehicle, the
travelled distance, and a set of electrical quantities
regarding the actual status of the battery, as the
instantaneous ingoing/outgoing motor current, the
instantaneous battery voltage and an estimate of its
residual capacity. These vehicle parameters were
acquired on a time base of two measurements every time
second. Also geographical definitions such as
instantaneous latitude, longitude and altitude were
acquired through a GPS receiver connected to the main
hardware; these latter data were acquired on a time base
of one measurement every second. All data were
organized in a database for further analyses, for a total of
55 days present in it. However, the contents related to 23
days, due to an excessive number of incomplete/missing
data, cannot be considered useful for our analysis.
The general arrangement of the acquisition system can be
observed in fig. 1.

2 Battery Life cycle

Several tests have been performed in ENEA to calculate
the life-cycle of lead-acid batteries [4,5,6], applied to the
evaluation of the benefits in the adoption of the Fig. 1 A general outline of the data acquisition system
supercapacitors in parallel with the batteries.
In [5] a battery pack installed on a city-car prototype, has In fig. 1, we can note the on-board system, able to collect
been tested executing the type approval driving cycle (the the vehicle instantaneous parameters and to transmit them
ECE15) in order to calculate the number of maximum to a server through a GSM/UMTS channel; moreover,
cycles of the battery. A comparison between a simple also the geographical data are available through a GPS
battery storage with a hybrid (battery plus receiver.
supercapacitors) has been made and the results was that Fig. 2 reports a typical recording, the corresponding
with the supercapacitors, capable to limit the maximum instantaneous speed profile (km/h versus time) during a
current supplied from batteries, the life of the lead-acid working day.
batteries increases. The same comparison has been made
also for a forklift [6], with the same result. Also the
economical evaluation demonstrate that the
supercapacitors can decrease the maintenance costs of the
vehicles. Such evaluation demonstrates that there is a
reduction of costs also taking account for the higher
purchase cost due to the Supercapacitors subsystem. In
[7], a life-cycle curve has been developed in order to
describe the probabilistic laws governing the number of
cycles, L, i.e. the battery lifetime, in terms of a design

Fig. 2 A typical recording
3 A New Battery Design Methodology
As previously mentioned, the rationale behind the
design procedure is the knowledge of data measurements
which are the milestone of the procedure explained in the
following, i.e. the so-called "Lauria method" .
The core of the procedure is based upon the
exploitation of a convenient cycle lifetime model which
is obtained by experimental test. It has been proven that
with respect to experiences described in the section II, the
following exponential curve is very adequate for
providing the expected number cycles as function of the
“design” specific power P and mass M.
N N0eOP / M (1)
with Oa positive constant, in our application estimated
as O = 0.031 , according to laboratory test statistics.
Even if a large variety of models could be employed
for predicting lifetime, in our opinion the exploitation of
the exponential model, by proper modifications, could be
more feasible, since the battery behavior is described in
a compact and global way. Two difficulties have to be
overcome, as described in the following.
The first relies in the need to generalize the
tailored. On the basis of the measurement data at our
disposal, the authors propose to assume as input
distribution for both the peak power and the energy
requested during a single battery cycle the Weibull
distribution [13-14] which has the well known following
cumulative probability distribution (cdf) with positive
parameters (T,J):
F(x) = 1- exp[-(x/T)J] (3)
The mean value, variance and the p-quantile -i.e. the
value xp for which F(xp)=p - can be expressed as function
of the Euler Gamma function *(z) by:
E[X] =P = T*(1 + 1/J) (4)
V2 T2 [ *(1 + 2/J) -*(1 + 1/J)2] (5)
1
xp T >  ln(1  p)@J (6)
Besides, the Weibull distribution is widely chosen as a
reference model capable of describing a large variety of
shapes for the pdf of a given rv. Furthermore, it can be
indeed characterized either by positive "skewness
coefficient" (i.e. with large right tails), either negative
ones. The “Skewness Coefficient”, SC, defined in terms
of the 3rd central moment and variance [13], is reported
in the following
SC = E[(X - P)3]/ (Var[X])3/2 (7)
More specifically, pdf curves are characterized by SC > 0
(if 0 < J < 3.6) and SC < 0 (if J > 3.6). A symmetric pdf,
quite similar to Gaussian pdf, is obtained for J = 3.6.
A Montecarlo simulation has been performed in order to
determine the mean value cost of the total number of the
various batteries over the entire time interval, related to
the system mission, by taking into account the fact that
the replacement interval, i.e. the lifetime of the battery
L, is a stochastic variable. The mean value of the cost is
regarded as a function of the total mass of the battery.
The mean cost is provided by the following relationship:
experimental model, over a restricted time horizon, to a ª 1  r  º»
L
larger one, corresponding to the actual system under E > C@ E « k M M (8)
1  r   1»¼
L
consideration. The authors propose to modify the «¬
expected lifetime L obtained by starting from the
experimental test, i.e. the above variable N from now on where kM is the specific cost and r is the interest rate,
denoted as L0, in the following manner: assumed equal to 0.07.
The lifetime L, expressed in year, can be evaluated on
L L0 e DL0 (2) the basis of 1), where:

N 0 0.031P / M
The coefficient α in the present paper has been chosen
equal to 0.03. The simple expression allows to express in
an effective way the various degradation effects which
occur along the entire battery lifetime, which cannot be
manifested during the reduced time of the experimental
tests.
The latter is related to the stochastic nature of the load
current. At this aim a probabilistic framework has to be
L0 e 0.03 * M * D / E E (9)
365
D is the discharge depth, E the energy over a single cycle,
expressed in kWh, and β is a constant which has the
meaning corresponding to the well-known Peukert’s law
[7-8]. It is easy to argue that the Monte Carlo procedure
is indispensable since a closed form for the probability
density function of the battery cost is practically
34
impossible to determine.
Samples for P and E are generated according the 33
Weibull distribution, already mentioned, which has been
proven the best one in fitting the numerical data. 32

In the successive section, the proposed approach is

explained with respect to the experimental data. 31

C [k€]
4 Numerical Application 30

As already outlined in the previous section, a statistical 29

analysis with respect to the measurement data is

preliminary performed. The peak power and the 28

requested energy are optimally described by a suitable

27
Weibull distribution. Specifically: 700 800 900 1000 1100
m [kg]
1200 1300 1400 1500

-The parameters of the peak power distribution are

T=18362 and J=30.36. Fig.4 Total cost versus battery mass
-The parameters of the energy demand pdf are T=3.35
and J=29.5, the curve being depicted in Fig.3.
0.25

0.3

0.2
0.25

0.15
0.2
Density
Density

0.15 0.1

0.1
0.05

0.05
0
0 2 4 6 8 10 12
Lifetime [year]
0
0 1 2 3 4 5 6 7 8
Energy [kWh] Fig.5: Lifetime distribution
Fig. 3 the Weibull pdf describing energy distribution

The correlation coefficient of the above two RV has been 5 Conclusion

estimated equal to 0.3. Thus, in the numerical application,
for sake of simplicity, these ones are supposed to be
statistically independent.
For sake of battery design the energy demand, by
maintaining the same distribution, is doubled for taking
into account an expected increased service along the time.
N0 and D have been assumed equal to 450 and 0.8
respectively.
Monte Carlo procedure has been implemented by
generating 3000 samples for both the peak power and the
energy.
The expected cost versus the battery mass is reported in
Fig. 4. The optimal value of battery mass was
numerically computed as equal to 1200 kg, as also
roughly apparent from Fig. 4.
The corresponding histogram of the lifetime L is shown
in Fig. 5, which is optimally fitted by a lognormal
distribution. The mean value of the lifetime is equal to
4.82 year which is aligned to the one substantially
observed in actual field.
In the paper, the battery lifetime prediction in terms of
basic design or random parameters, and its proper design
are based upon extensive data elaboration and suitable
lifetime model, under the so called "Lauria method".
According to such method: first, battery lifetime
prediction is based upon the accelerated test which
without any doubt provide the core of the lifetime model.
Moreover, a correction of the interpolation model is
added for taking into account the unavoidable
degradation terms related to ageing effects occurring in
real operation, which cannot be wholly observed in data
corresponding to accelerated test, not only for the length
of such tests but also because it is not possible under such
tests to keep all random factors affecting under control.
Then, the optimal design of battery is dealt with in such
framework, assuming a Weibull model for battery
random load and a proper cost function expressed in
terms of basic battery features, also taking into account
replacement intervals. A numerical application illustrated
in the second part of the paper has shown the feasibility
of the approach. The method is meant to complete to a
general probabilistic methodology for deriving the above Smart Grids Under an Inverse Gaussian Model, IREE,
International Review of Electrical Engineering, August
lifetime distribution, as illustrated in [9,10] following the
2013, pp.1253-1266
Rakhmatov and Virudhula battery model [7-8]. Further [13] A. Papoulis, S. U. Pillai, Probability, Random Variables
steps are worth to be developed towards the aspect of and Stochastic Processes, 4th ed., McGraw-Hill, 2001
parameter estimation of the model [14-16]. In such field, [14] N.L. Johnson, S. Kotz, N. Balakrishnan, Continuous
Univariate Distributions, J. Wiley, Vol. 1 and 2, 1995
the opportunity of Bayesian methods comes to mind, for
[15] W.Q. Meeker, L.A. Escobar, Statistical Methods for
the lack of historical which is typical with high-reliability Reliability Data, J. Wiley & Sons, 1998
components, and the efficiency and robustness of the [16] H.F. Martz and R.A.Waller, Bayesian Reliability
Bayesian method, which are evident in the case of small Analysis, Malabar, FL: Krieger Publishing, 1991
[17] S. Chen, T. Ho, B.H. Mao, Component Reliability
sample sizes [16-18]. Another possible exploitation of
Estimations without Field Data, Transactions of Hong
Bayesian methods for battery design is the framework of Kong Institution of Engineers, vol. 14, n. 3, 2007 pp. 10-
Multicriteria Analysis with random inputs, as recently 17
illustrated in [19]. [18] E. Chiodo, Model Robustness Analysis of a Bayes Stress-
Strength Reliability Estimation with Limited Data, Proc.
of IEEE International Symposium on Power Electronics,
Electrical Drives, Automation and Motion (SPEEDAM
2014), Ischia (Italy), 14-16 June 2014
References [19] E. Chiodo, L. P. Di Noia, R. Rizzo, The Application of
Bayes Inference in Multicriteria Analysis to Design
[1] Agarwal, V., Uthaichana, K. , De Carlo, R.A., Tsoukalas, Energy Storage Systems in Renewable Power Generation,
L.H.: Development and Validation of a Battery Model Proc. IEEE International Conference on Clean Electrical
Useful for Discharging and Charging Power Control and Power Renewable Energy Resources Impact (ICCEP
Lifetime Estimation, IEEE Tr. on Energy Conversion, 2013), Alghero, Italy, June 11-13, 2013
Vol. 25, No. 3, September 2010, pp: 821-835.
[2] Lidula, N.W.A.; Rajapakse, A.D.: Microgrids research:
A review of experimental microgrids and test systems,
Renewable and Sustainable Energy Reviews, Vol.15,
Index Terms— electrical energy storage systems,
renewable energy, statistical methods.
[3] Conte M., Pede G., Sglavo V,, Macerata D., “Assessment
of high power HEV lead-acid battery advancements by
comparative benchmarking with a European test
procedure” Journal of power sources, Elsevier, Vol.16
No. pp:118-127, 2003
[4] Sglavo,V., Vellucci, F., “Prove Vita di batterie piombo-
acido con e senza livellamento del carico”, Report
RdS/2012/ 121
[5] Ortenzi, F., Pede, G., and Rossi, E., Cycle Life Cost
Assessment of a Hybrid Lead Acid Battery-
Supercapacitor Storage for an Electric Microcar SAE
Technical Paper 2014-01-1816, 2014, doi:10.4271/2014-
01-1816
[6] Conte M., Genovese A., Ortenzi F., Vellucci F.,Hybrid
battery-supercapacitor storage for an electric forklift: a
life-cycle cost assessment, Journal of Applied
ElectrochemistryApril 2014, Volume 44, Issue 4, pp 523-
532
[7] E. Chiodo, D. Lauria, M. Pagano, G. Pede, F. Vellucci,
Experimental performances and life cycle estimation of
hybrid electric storage systems, Proc. IEEE International
Conference on Clean Electrical Power Renewable Energy
Resources Impact (ICCEP 2013) , Alghero, Italy, June
11-13, 2013.
[8] Laldin, O., Moshirvaziri, M., Trescases, O.: Predictive
Algorithm for Optimizing Power Flow in Hybrid
Ultracapacitor/Battery Storage Systems for Light Electric
Vehicles, IEEE Transactions on Power Electronics,
vol.28,No.8, 2013, pp.3882-3895.
[9] Li W.: Evaluating Mean Life of Power System Equipment
with Limited End-of-Life Failure Data”, IEEE
Transactions on Power Systems, Vol. 19, No. 1, 2004, pp.
236-242.
[10] Rakhmatov, D., Vrudhula, S.: An analytical high-level
battery model for use in energy management of portable
electronic systems, Proc. of Int. Conf. on Computer
Aided Design (IC-CAD01), 2001, pp: 488-493.
[11] Rakhmatov, D., Vrudhula, S., Wallach D.A.: A Model
for Battery Lifetime Analysis for Organizing Applications
on a Pocket Computer, IEEE Tr. on Very Large scale
integration (VLSI) Systems, Vol. 11, No. 6, December
2003, pp: 1019-1030.
[12] E. Chiodo , A. Del Pizzo, L. P Di Noia, D. Lauria,
Modeling and Bayes Estimation of Battery Lifetime for