Advanced Battery Management System using

MATLAB/Simulink

Bijender Kumar Dr. Neeta Khare Dr. P.K.Chaturvedi

Electronics Engineering Department Energy Storage & System Integrator Electronics Engineering Department
Banasthali University, Banasthali Columbus, USA SRM University, Ghaziabad
Rajasthan, India neeta_ks@yahoo.com Uttar Pradesh, India
bijender82mehandia@yahoo.co.in pkchat1944@gmail.com

Abstract—A battery management system (BMS) is a system temperature. SoH is a ‘measure’ that reflects the aging of a
that manages a rechargeable battery (cell or battery pack), by battery and its ability to deliver the specified performance
protecting the battery to operate beyond its safe limits and compared to a fresh battery.
monitoring its state of charge (SoC) & state of health (SoH). BMS
has been the essential integral part of hybrid electrical vehicles Battery Monitoring means keeping an eye on the key
(HEVs) & electrical vehicles (EVs). BMS provides safety to the operational parameters during charging and discharging.
system and user with run time monitoring of battery for any Electrical key parameters are voltage, current, battery internal
critical hazarder conditions. In the present work, design & resistance and ambient temperature during cycles [1]. SoC &
simulation of BMS for EVs is presented. The entire model of SoH indications provide protection by generating alarms or
BMS & all other functional blocks of BMS are implemented in visual indications for any normal malfunctioning in the system.
Simulink toolbox of MATLAB R2012a. The BMS presented in Smooth functioning of heavy mobile systems depends on the
this research paper includes Neural Network Controller (NNC), accurate SoC and SoH monitoring. The monitoring is
Fuzzy Logic Controller (FLC) & Statistical Model. The battery addressed by an independent system known as battery
parameters required to design and simulate the BMS are management system. A general BMS architecture is presented
extracted from the experimental results and incorporated in the and the function of each block is given in Ref. [2]. An effective
model. The Neuro-Fuzzy approach is used to model the BMS can protect the battery from damage, predict battery life,
electrochemical behavior of the Lead-acid battery (selected for and maintain battery operation for offering high precision and
case study) then used to estimate the SoC. The Statistical model is
perfection [3]. A few BMS for different battery packs which
used to address battery’s SoH. Battery cycle test results have
been used for initial model design, Neural Network training and
are now commercially available are detailed in Ref. [4-6]. A
later; it is transferred to the design & simulation of BMS using BMS for LiFePO4 battery is presented in Ref. [7]. A Modular
Simulink. The simulation results are validated by experimental Battery Management System for HEVs is presented in Ref. [8].
results and MATLAB/Simulink simulation. This model provides The BMS presented in this research paper is designed and
more than 97% accuracy in SoC and reasonably accurate SoH. simulated using the experimentally extracted parameters of the
lead acid battery. In the present work, SoC is estimated using
Keywords— EVs; SoC; SoH; BMS; NNC; FLC.
the Neuro-Fuzzy Approach, in addition multivariate linear
I. INTRODUCTION Regression Model is used for estimating SoH [9]. The NNC is
simulated & trained on the collected data set of lead acid
Batteries are widely used as the most common electrical battery using NNTOOL and transferred to Simulink. If-then
energy storage device in vehicles as a replacement of rules of fuzzy modelling are defined to create the linguistic
traditional fuel. The capacity of battery is gradually reduced results. FLC is imported in Simulink. Regression Model is
due to corrosion, passivation, out gassing, decomposition of structured in Simulink by using different mathematical blocks.
materials and morphological changes on electrode surface
during its operation, when it is subjected to a load or a source. The paper is organized as follows. The proposed advanced
Also, the battery weakening process can be enhanced when the BMS is presented in section II including adopted SoC & SoH
battery is subjected to operate beyond the specified safe algorithms and their Simulink implementation. Further the
operating conditions. The safe operation and long life of a MATLAB/Simulink simulation results are discussed in section
battery depend on the current delivery during charging and III followed by conclusion remarks in section IV.
discharging modes within the specified limit under all specified
temperature conditions. The safe operation of battery cannot be II. ADVANCED BATTERY MANAGEMENT SYSTEM
estimated only by direct measurements as it also depends on The proposed structure of advanced BMS is shown in Fig.
the present state of the battery. The battery state is used to 1. The block diagram of the BMS is facilitated with the
estimate the expected lifetime of the battery and can simply be following blocks:-
described by two parameters: SoC and SoH. SoC is interpreted
as a present charged state that mostly depends on current in & ¾ Battery Input Parameters Extraction Block
out and its initial charge condition of a battery at given ¾ NNC
 ¾ FLC
¾ Data Storage Read/Write Block
¾ Statistical Model
¾ Output Module
NNC and FLC together operate as SoC determination block of
BMS. SoH determination block of BMS is a Statistical Model.

Fig. 2. Activity chart for data collection

Our previous research group also implemented the model

into an instrumentation board for measuring & scaling of all
the five parameters of battery as shown in Fig. 3. The inbuilt
microcontroller on the instrumentation board runs the NNC &
FLC algorithms.

Fig. 1. Proposed structure of Battery Management System

The detailed descriptions of each block of proposed BMS

are discussed below:-
A. Input Parameters Extraction
The selection of parameters and their data collection are big
challenges in any kind of BMS design. We are referring Fig. 3. Instrumentation Board for extracting battery parameters
pervious work of our research group where, our colleagues
have collected data through laboratory based circuits- Bitrode B. Neural Network Controller
battery test system and Agilent impedance measurement The NNC is facilitated with Artificial Neural Networks
system at Exide Industries Ltd, R&D Lab, Kolkata, India for
(ANN) model which gives Specific Gravity (SG) output in
extracting the following five electrical parameters of a 12V
response to stimuli of battery parameters. The NNC algorithm
lead acid battery MF40sv:- 1) Consumption Time 2) Current
Drawn 3) Terminal Voltage 4) Temperature 5) Internal and its Simulink implementation are discussed in the
Resistance. Results given in Ref. [10] proves that NNC design following sections.
with preferred above five input parameters gives greater than
99% accuracy for Specific Gravity (SG) measurement, whereas 1) Algorithm for SG determination
the accuracy reduces to less than 90% with four parameters
(Current Drawn, Terminal Voltage, Temperature, Internal One of the key parameters of battery operation is the
Resistance) and falls down to less than 60% when only three specific gravity of the electrolyte. ANN is used in this research
parameters (Current Drawn, Terminal Voltage, Temperature) to simulate the discharging process of a battery and to predict
are employed, leading to the decision that all five parameters the specific gravity of electrolyte using all five input
are essential for accurate BMS design. The measurements were parameters [10].
carried out after simulation of operation of car in the lab
separately for Slow Discharge (SD) and Real Cranking (RC) at 2) Simulink Implementation
different temperatures (00C, 270C & 470C), different state of
charge (100%, 75% & 50%) and for different age of battery In the present research, the NNC block of proposed BMS
(fresh battery, one year old, two year old), as shown in activity in Fig. 1 is facilitated with the algorithm of SG
chart Fig. 2. determination/ANN architecture. The detailed Simulink
Model of NNC is shown in Fig. 4.
 relationship between input Specific Gravity (NNC output) and
the battery temperature with the SoC to a linguistic term
depending on the 35 rules of fuzzy modeling. If-then rules of
Fuzzy Modeling are defined in table 1 to specify the SoC of
the battery.
Table 1: The Fuzzy Rule Base for %SoC

Fig. 4. Simulink Model of Neural Network Controller 2) Simulink Implementation

NNC is multilayered architecture that consists of three In the present research, the implementation of FLC is done
cascading layers. The selected five battery parameters are fed using MATLAB’s Fuzzy Logic Toolbox and then translated to
as inputs to the controller using constant value parameter Simulink as shown in Fig. 5.
block (Battery input parameters). The inputs of each layer are
initially processed through the dotprod block which is used to
evaluate dotprod function between the layer’s inputs and
layer’s weights. The weights of layer1, layer2 & layer3 are
defined inside the blocks IW{1,1}, LW{2,1} & LW{3,2}
respectively. The constant value parameter blocks b{1}, b{2} Fig. 5. Simulink Model of Fuzzy Logic Controller
& b{3}are incorporated in controller to hold the biases values
for layer1, layer2 & layer3 respectively. The blocks netsum is The specific gravity and temperature online inputs are
a net input function block and used to predict the net inputs for applied to FLC, algorithm runs and predict the SoC.
layer1, layer2 & layer3 respectively by combining their Defuzzification block is used within the FLC for interpreting
weighted inputs & biases. The blocks logsig & logsig1 of the membership degrees of the fuzzy sets into crisp value. Fig.
controller are representing the Log-sigmoid transfer function 5 shows a combined block of fuzzification & defuzzyfication
block and block purelin is representing the linear transfer to provide crisp values of SoC by superimposing If-then rules.
function block. These blocks are used to calculate the required Fuzzy Inference System (FIS) incorporated in FLC is shown
output of each respective layer from their net inputs. The in Fig. 6.
parameter blocks a{1} & a{2} are the vectors of eleven
constant values for the eleven neurons of Layer1 & Layer2
respectively. The predicted output of NNC that is specific
gravity, is observed by the constant value parameter block
a{3}.
The model in Fig. 4 calculates the specific gravity of 12V
Lead-acid battery equals to 1.242 for 150A discharge for 3.16
minutes at 40.50C temperature. The battery terminal voltage
drops to 10.88 volts from its initial voltage and internal
resistance increases to 6.11 ohms due to discharge.
C. Fuzzy Logic Controller
In the present research, FLC is designed to estimate the
SoC of lead acid battery. The FLC algorithm and its Simulink
implementation are discussed in the following sections.
Fig. 6. Fuzzy Inference indicating full charged status of battery
1) Algorithm for SoC determination
Here, it is noticed that SoC results of Simulink model of FLC
The Neuro-Fuzzy Approach is adopted for SoC estimation
in Fig. 5 and fuzzy inference system in Fig. 6 are similar.
in which the nonlinear adoptive-learning capability of ANN is
used to simulate the discharging process of a battery which is
D. Statistical Model
translated linguistically using fuzzy logic to indicate the
charged state of the battery [11]. In the reference paper [12], In the present research, the statistical model of SoH is
FLC is employed for the purpose of transforming the implemented by Simulink. The cause of battery health
deterioration is the effect of aging on the grid, electrodes, determine the required slops m, m’ & m’’ using the real time
contacts, corrosion and charging /discharging cycles. The data of SG, OCV & IR. For testing purposes, the required data
algorithm for SoH estimation and its Simulink implementation for slops calculation and the value of TT are saved in memory
are discussed in the following sections. and called during the implementation of statistical model to
provide SOH of a lead-acid battery as per equation (1). The
1) Algorithm for SoH determination final output of Statistical Model comes out in the percentage
using equation (2).
The SoH determination technique using multivariate linear
regression [11] on the aging effect and the on-time consump- III. MATLAB/SIMULINK SIMULATION
tion of the battery is selected for implementation. The formula Initially, the ANN is designed, trained and simulated using
obtained for estimating SoH after applying the multiple nntool of MATLAB. Then NNC is design in Simulink in which
regression technique is given as: the ANN algorithm is implemented. The comparison between
the output of NNC (i.e SG_Observed_ANN) and the
SoH = 1.0043 + 0.0088(TT × C) + 3.8925 m(SG) + experimentally measured SG of lead-acid battery (i.e.
0.2444m’(OCV) – 0.0863m’’(IR) (1) SG_experimental_Exide) is shown in Fig. 8. The value of the
sum of squares of error (SSE) is 0.0004989 and R-square is
Where TT is the run-time of the battery and C is the discharge 0.999. Results in the Fig. 8 verify the values of SSE & residual.
rate. TT × C gives the ampere-hour consumption of the battery
and m (SG), m’ (OCV) and m’’ (IR) are slopes of specific
gravity, terminal voltage and internal resistance respectively
whose values are calculated in Ref. [12]. SoH is representing
the used life of battery. The expected battery life is estimated
by the following formula:

Expected Battery Life (%) = 100 – Used Life (SoH) (2)

2) Simulink Implementation

The multivariate linear Regression Model described in Fig. 8. SG of ANN versus SG experimental
equations (1) & (2) for battery SoH estimation is implemented
in statistical Model block of proposed BMS in Fig. 1 and its This linear plot proves that NNC Simulink model is
detailed structure is shown in Fig. 7. accurately designed as it measured the expected values of
experimental SG and values of SSE & R-square also proves
99.9% accuracy of our model. The variations in the estimated
SoC of advanced BMS with the increase in SG &
temperatures compensated at 00C, 270C & 400C are shown in
Fig. 9.

Fig. 7. Simulink Implementation of statistical model Fig. 9. SoC versus SG & Temperature

The constant parameters are fed to the Statistical model by It is observed that SoC is high at higher Specific Gravity
using six constant parameters blocks. The value of the and also increases with increase in temperature (verified from
discharge current ‘C’ is fetched from the data input as a real [13], [14]). The relation between SoC & SG is not linear at
time data from the proposed structure of BMS as shown in different temperatures due to the effect of increase in
Fig. 1. The sum blocks and product blocks are used to temperature on electrochemical reactions and out gassing or
hydrogen gas generations. In addition, Fig. 9 does not show Expected Battery Life vs Temperature for two year old battery
100% of SoC even at 1.24 service specific gravity that means 96 100% RC
battery suffers from capacity loss and its operation varies with

Expected Battery Life (%)

75% RC
50% RC
temperature. The capacity loss should be compensated for
temperature variations. 90
In Fig. 10, the graphical representations of the expected
percentage battery life with temperature at different ages of
battery (fresh, one year old & two year old) for RC & SD 84

operations of a car are shown. The Fig. 10 (A), (C) & (E)
represent the RC operation of a car and Fig. 10 (B), (D) & (F)
represents the SD operation of a car. 0 10 20
o
30 40

Temperature ( C)
Expected Battery Life vs Temperature for Fresh Battery (E)
Expected Battery Life (%)

100% RC
75% RC
Expected Battery Life vs Temperature for two year old battery
90 50% RC 95

100% SD

Expected Battery Life (%)

75% SD
50% SD
84

90

78

0 20 40
o 85
Tem perature ( C)

(A)
80
Expected Battery Life vs Temperature for Fresh Battery
0 20 40
95
o
Temperature ( C)
Expected Battery Life (%)

100% SD
75% SD
50% SD

90 (F)
Fig. 10. Expected Battery life versus Temperature for different aging of
85 battery for both RC & SD operations

It is observed here that battery life is reducing with increase

80
0 20
o
40 in temperature (verified from [14]) and also battery is
Temperature ( C)
degrading as it is being old.
(B)
IV. CONCLUSION
Expected Battery Life vs Temperature for one year old battery
100
In this research paper, it is concluded that
Expected Battery Life (%)

MATLAB/Simulink model implementing the Battery

96
100% RC
Management System is very efficient approach to indicate the
75% RC
50% RC
battery state SoC & SoH. There are lots of methods existing in
92
market for SoC & SoH determination. In the presented paper,
88
the Neuro-Fuzzy and Statistical model are re-simulated with
Simulink to compare the two MATLAB tools and for
84 connecting the Simulink model to HDL coder for chip design
0 20 40
of BMS. It can be seen by MATLAB/Simulink simulation
Temperature ( C)
o
results that a Neuro-Fuzzy Approach & Statistical Model have
high degree of confidence for control strategy &
(C) implementation of a BMS. The designed Simulink model will
Expected Battery Life vs Temperature for one year old Battery
be convenient for developing FPGA based BMS. FPGA based
100% SD
BMS has advantages over other existing BMS because of low
75% SD NRE cost, low power consumption, high speed of operation,
Expected Battery Life (%)

50% SD
92
large reconfigurable logic, large data storage capacity and
hence better performance.
88

REFERENCES
84
[1] Vinal, G.W., “Storage Battery”, John Wiley and Sons, New York, IVth
edition, pp.130-336, 1924-1955.
0 20
Temperature ( C)
o
40
[2] Valer Pop, HJ Bergveld “Battery Management System Accurate State-of-
Charge Indication for Battery Powered Applications” Philips Research
Book Series Volume 9 Chapter 1, Springer Nederland 2008.
(D)
[3] C. Chen, K.L. Man “Design and Realization of a Smart Battery
Management System” International multiconference of engineers and
computer scientists volume II, Hong Kong 2012.
[4] EV Hybrid Vehicle Electronic, Extend Computer and Instrument,
http://www.extendcomputer.com/html/ev_hybrid.html.
[5] Smart Guard Battery Control System, Aerovironment, Inc.,
http://www.aerovironment.com.
[6] BatOpt Battery management System, AC Propulsion Inc.,
http://www.acpropulsion.com/products/Battery_mgt.html.
[7] B.P.DIVAKAR, H.J.WU, J. XU “Battery Management System and
Control Strategy for Hybrid and Electric Vehicle” 3rd International
Conference on Power Electronics Systems and Applications 2009.
[8] Thomas Stuart, Fang Fang “A Modular Battery Management System
for HEVs” University of Toledo 2002.
[9] Neeta Khare, Rekha Govil “Modeling Automotive Battery
Diagnostics” Power Electronics Technology, March 2008.
[10] Neeta Khare, Rekha Govil “A Battery Storage System for Fault
Tolerance” IEEE, 2007.
[11] Neeta Khare, Rekha Govil, and Surendra Kumar Mittal, “A Process
of determining State of Charge and State of Health of a Battery”,
Indian Patent, 813/KOL/2005.
[12] Neeta Khare, Shalini Chandra and Rekha Govil “Statistical modeling
of SoH of an automotive battery for online Indication” IEEE, 2008.
[13] Rolls Battery Engineering, North America http://www.rolls-
battery.com/content/technical-downloads?q=node/51.
[14] BDI Inc.
http://cms.cdnrg.com/images/uploads/downloads/files/Temperature_E
ffec ts.pdf.