Battery

Management
System
대체에너지기관

Nadya Novarizka Mawuntu

기계설계공학
 a. Background

Increasing Demand of LiB

Introduction

Li-Ion Battery Application

BATTERY
Source: IEA (International Energy Agency) MANAGEME
NT
01
 b. BMS Functionality

VOLTAGE

CURRENT

TEMPERATURE
Introduction

KEY FUNCTIONS Accurate modeling and state

Battery Modeling and
estimation can ensure reliable
02 State Estimation operation and optimize the battery
system.
 c. Battery Modeling
Battery modeling plays an important role in estimating battery states, which
include state of charge (SOC), state of health (SOH) and state of power (SOP).

Electrochemical Models (EM) Data-driven Models (Black Box)

• EM describes physical and electrochemical • Black box models such as neural network
processes inside the battery. models require a large amount of
+ High accuracy, but experimental data to train models.
- Heavy computation + Good modeling capability
Introduction

Electrical Equivalent Circuit Models

(ECM)
• Facilitates estimation schemes with electrical parameters.
+ Simple structure and easy to be identified.
- Susceptible to aging process
03
 Equivalent Circuit Model (ECM)

Equivalent Circuit Model Battery Polarization

Models the battery by building up the battery behavior Polarization : Any departure of the
using common circuit elements. cell's terminal voltage away from OCV
due to a passage of current.
1. Open-circuit voltage (OCV) and SOC
Equivalent Circuit Model

2. Instantaneous Response

Diffusion voltage
Relaxation
3. Diffusion Voltage

01
Instantaneous Response
04 Voltage Response to Discharge Current Pulse
 Equivalent Circuit Model (ECM)

1st Order ECM 2nd Order ECM

Equivalent Circuit Model

4 Parameters 6 Parameters

1. Voltage Source 1. Voltage Source

2. Series Resistor (R0) 2. Series Resistor (R0)
Models instantaneous response to a change Models instantaneous response to a change
in input current. in input current.

01
3. 1 Pair of Resistor- Capacitor 3. 2 Pairs of Resistor- Capacitor
05
Mimic the diffusion voltage process. Mimic the diffusion voltage process.
 Parameter Estimation

?
How to estimate the model parameters?

Experiment Simulation
Parameter Estimation

compared
Model

At every iteration, Error Minimized

06
01
Parameters Updated
 Parameter Estimation

? How to estimate the model parameters?

Incremental OCV Test

Experiment
Cell was fully discharged
18,650 LiNiMnCoO2/Graphite
Parameter Estimation

using a negative pulse

lithium-ion cells
current for 12 minutes, with
relaxation duration of 2
hours at every 10% SOC.
“Focus on middle-range SOC”

01
07

Source: Center for Advanced Life Cycle Engineering (CALCE), University of Maryland.
 Parameter Estimation

? How to estimate the model parameters?

Simulation
Model
Parameter Estimation

Inside the battery block:

1-RC Pair 2-RC Pairs

Table-based
4 parameters 6 parameters
parameters:
• Voltage Source = 9 parameters • Voltage Source = 9 parameters
• Internal Resistance = 9 parameters • Internal Resistance = 9 parameters
01
08
• Single RC-Pair = 18 parameters • Double RC-Pair = 36 parameters

36 parameters 54 parameters
 Parameter Estimation

?How to estimate the model parameters?

The dynamics of the nth order ECM battery
Estimation Method:

cells can be described by Eq. (1).

A cost function is used to minimize the square-error
sum (SSE) between the experimental data and
N= Number of RC pair; i= Current ; VOC= State of Charge dependent OCV
simulations of the battery voltage.
Vn= Voltage across the RC branch ; Vbat= Voltage output
Parameter Estimation

And discretized to Eq. (2).

[ ][ ( )
][ ]
𝑇𝑆
𝑉 1 [𝑘+1] e xp − 0 𝑉 1 [𝑘]
¿ 𝑅1 𝐶1 ¿
=
¿ ¿ Nonlinear Least Squares
𝑉 𝑁 [𝑘+ 1] ¿0
(
e xp −
𝑇𝑆
𝑅1 𝐶1 ) 𝑉 𝑁 [𝑘]
Minimizes the squares of the residuals.

[ ( ( ))
)]
𝑇𝑆
𝑅 1 1− 𝑒𝑥𝑝 −
𝑅1 𝐶 1
+ 𝑖[𝑘] Unknown circuit parameters:

01
(
¿ 𝑅 𝑁 1 −𝑒𝑥𝑝 −
( 𝑇𝑆
𝑅𝑁 𝐶 𝑁 ) (CN, RN, R0 and Voc)
Where f1(x), f2(x),…, fn(x) represent residuals, and n is
09
the number of samples.
 Result and Analysis

Vout
Result and Analysis

Second-order ECM (Improved Thevenin Model)

07
describes the slow polarization better than the
10 first-order ECM.
 Result and Analysis

Estimated Parameters
First-order ECM
Second-order
  First-order ECM
ECM
Parameters 36 54
Iterations 11 9

Estimation Time 3 hours 18 minutes 4 hours

Result and Analysis

Sum-squared Error 0.0176 0.0053

Second-order ECM In comparison, second-order ECM converged in

longer estimation time but has lower SSE value.

07
11
 B

SOC
ESTIMATION
 State of Charge Estimation
Percentage of the remaining capacity to the
battery nominal capacity.

CATEGORIES
State of Charge Estimation

1. Conventional methods
 Open circuit voltage, coulomb counting.
2. Observer-based methods.
Contribution of different algorithm in the past
 Luenberger observer, PI observer.
studies (2009-2018) on online SOC estimation
3. Filter-based methods
(Source: IEEE, 2018 database).
 EKF, UKF, HIF, CKF, AUKF, etc.
4. Data-driven
Most Popular!
01  NN method, GA, SVM, FNN.
13
 Coulomb Counting Method
Integrating the current that enters and leaves the
cell, and divides it with the total capacity to keep
track of the state of charge over time.

Advantages:
State of Charge Estimation

+ Simplicity and low computational cost

Disadvantages:
- The accumulation of current sensor error
and its inability to recover from a wrong
initial condition because of the lack of
feedback from voltage measurement.
01
14
 Extended Kalman Filter
EKF uses partial derivatives and fi rst order
Taylor series expansion to linearize the battery
model.

How it works
State of Charge Estimation

The state-space model is linearized at each time

instance, which compares the predicted value
with its measured terminal voltage to correct
the estimation parameters for SOC. The
Jacobian matrix is required.

Limitation

Firs order Taylor series suffers from a lack of

01
15 accuracy in highly non-linear condition.
 Unscented Kalman Filter
UKF is used to address the EKF signifi cant error
in high order nonlinear model based on the
unscented transformation.

A set of points called sigma points is used to

State of Charge Estimation

represent the mean and covariance of the state

distribution.
• Jacobian matrix is not required.
• Better accuracy than EKF.

Limitation

Highly affected by the specifi ed initial value.

01 High measurement noise can leads to
16
divergence.
 Result and Analysis

Comparison between EKF and UKF

Comparison
Result and Analysis

UKF p erforms a be tter

estima tion fo r b attery SO C
07
17 with high er accuracy.
 Future Work

? How to improve the estimation?

Prominent Advantages:
Data Driven 1. Adaptive and able to fi t well
SOC Estimation nonlinear systems as the battery
behavior.
Data-driven methods based 2. Relatively simple structure
Future Work

on machine learning have 3. Rapid response

rapidly developed. 4. Easily applied in different type of
battery chemistries

05
18