Predictive analytics of Lithium-lon battery for

optimization and battery failure using

Machine learning algorithms -A Survey

Kush Ise Satyam Singh

Department of Computer Engineering Department of Computer Engineering
Pimpri Chinchwad College of Engineering Pimpri Chinchwad Coilege of Engineering
Pune, India Pune, India
[email protected] kunwarsatyam.singh1 [email protected]

Abhiraj Deshpande Roshan Bahiram

Department of Computer Engineering Department of Computer Engineering
Pimpri Chinchwad College of Engineering Pimpri Chinchwad College of Engineering
Pune, India Pune, India
abhiraj.deshpande [email protected] roshan.bahiram [email protected]

Prof. Rajesh Lomte

Department of Computer Engineering

Pimpri Chinchwad College of Engineering
Pune, India
[email protected]

correct and scalable prophetic analytics to

ABSTRACT--. In the increasing world of
create a battery that is additionally
usage of batteries the sale and use of batteries
productive and safe.
need continuous testing and analysis to live
INTRODUCTION
performance characteristics. Daily
Lithium-ion batteries power virtually each
Gigabytes of knowledge, Excel-based, very
device in our lives, together with phones and
little analysis Distributed groups, new
Laptops. They're at the guts of renewable
information area unit increasing
energy and e-mobility. For years corporations
considerations for the users. Here are a unit
have tried to predict what number charging
simply a number of crucial tests: research
cycle's battery can last before it dies. Higher
and = imaging, environmental testing,
predictions would modify a lot of correct
chemical characterization, part testing,
quality assessment and improve semie
sample preparation, mechanical
permanent designing. However that's tough, as
characterization, and _ testing. Large
a result of each battery ages otherwise, looking
demand in battery optimisation because the
on its usage and conditions throughout
battery usage is increasing to exchange fossil
producing. A recent paper, claims to possess
fuels for higher surroundings. To analyze,
found the key tounravel this drawback
report, and predict batteries and identify
by “combiningcomprehensive experimental
insights by AI and Cloud. With battery
experience embedded within the software information andartificial intelligence”.
package platform, to alter analysis and
provide engineers with access to the results Although their resultsarea unit prior ancient
they have to satisfy deadlines and strategies, the teamcentered a lot of on their
performance targets. It saves time with domain data in
engineering than on the machine learning half
Battery analytics platforms and technologies In [2] the author suggested the emergence of AI and
machine learning (ML) which has contributed
square measure transferral intelligence into the
significantly to the increase in the analysis and
battery producing method and Ist life. The development of the latest provincial measurement
intelligence generated by analytics platforms is strategies car batteries are powered by electricity.
employed to increase the lifetime of batteries, Data-driven methods, such as ML, get an additional
manufacture higher batteries, unlock new measurement of charging status (SOC) and health
status (SOH) due to the larger hand of battery
markets for end-of-life batteries and forestall information and improved computer power. This paper
safety hazards and recollects. Innovators provides a survey of battery state measurement
square measure gathering knowledge in new strategies that support cubic centimeter methods such
ways in which from battery producing, loT and as feed forward neural networks (FNNs), perennial
neural networks (RNNs), vector support devices
alternative hardware and Battery Management
(SVM), radial basis functions ( RBF), and active
System. The main business models within the networks. strategies that reflect information quality
house square measure eitherpure SaaS solutions standards, inputs and outputs, see conditions, battery
or SaaS+ hardware element. Pure SaaS types, and expressed the accuracy of giving students an
platforms faucet into existing sources of even greater picture of the study of the cubic
centimeter landscape of SOC and the scale of sol. to
knowledge, like BMS, whereas SaaS +
begin with, to generate an understanding of how best
hardware solutions will generate new to compare by comparing different neural network
knowledge. Innovative solutions square structures, Associate in Nursing FNN and long
measure targeting totally different components memory (LSTM) RNN are trained fifty times every
of the worth chain, from battery production to 3000 times. The error is completely different from all
of the trainer’s repetitions due to the initial random
in-vehicle use, together with totally different values of the training parameters, indicating that it is
vehicle sorts, from small quality to serious duty necessary to train the networks more often to get the
vehicles. Customers embrace battery makers, most effective result. In addition, it is recommended
automotive vehicle manufacturers, fleet that when comparisons between measurement
strategies similar to those provided in this review
operators and battery repurposed the worth
paper, comparable networks should have the same
proposition differs across client segments. variety of learning parameters and be trained and
evaluated with the same information. Apart from that,
it is difficult to make a general conclusion regarding
LITERATURE REVIEW the level of measurement strategy provided.
In [l] the author used to achieve reliable battery
In [3] the author proposes a system for accurately
management in electric vehicles (EVs), a
predicting the life span of sophisticated, off-line
sophisticated State-of-Charge (SOC) calculator systems such as lithium ion batteries are essential for
was developed in the form of machine learning. the development of fast technologies. However, the
The process of compiling a_ standardized many ways of aging, the diversity of essential
algorithm designed for fuzzy C-means (FCM) resources and the dynamic working conditions remain
major challenges. We usually produce a
begins to separate the sample training information
comprehensive database containing 124 phosphate /
within the drive cycle of the lithium-ion battery. graphite cell cycling cells under rapidly changing
The bunch effect is used to be told by the topology conditions, a variety of life cycles ranging from one
parameters and the antecedent model. The hundred and fifty to two, 300 cycles. Using the
extraction of voltage curves from early cycles to
algorithmic-square-squares formula is used to
indicate energy damage, we often use machine
subtract its following parameters. To ensure the learning tools in each to predict and separate cells in
accuracy and robustness of the intellect, the back cycle life. Our larger models reach the 1% test error to
reading distribution formula was finally adopted obtain a predictor of cycles of life cycle cycles per 100
simultaneously with the opposite and subsequent cycles (showing an increase between zero. 2% from
the first dose) and 4.9% of the cycles of the five major
increase. The test results confirm that the
cycles of life cycle divisions.
calculated calculator shows partial accuracy and
exceeds that for standard modeling methods.
In [4] the author has tried in the case of shopper large-scale study of cases with battery data obtained
physical and energy storage systems, the market from Ames of National Physical Science and Space
demand for lithium-ion batteries is rapidly Administration (NASA).Details of the Prognostics
increasing to increase the capacity of lithium-ion, Center of Excellence (PCoE). The proposed DNN
it is necessary to analyze the influence of battery rule has been compared with different machine
parameters present in the battery. In this paper, learning algorithms, namely, Support Vector
data details} mining technology is used to Machine (SVM), k-Nearest Neighbors (k-NN),
evaluate and analyze lithium-ion _ battery Artificial Neural Networks (ANN), and simple
parameter data, intended to test the relationship regression (LR). Test results reveal that DNN
between multiple parameters and power in battery algorithm performance is possible compare or break
charging and output processes, and Python different rules for machine learning. In addition, the
language is used to understand this conclusion. presented results may serve as a benchmark for the
Lithium-ion battery data processing technology learning process of the RUL abuse predictor directly
includes cleaning and selection of lithium-ion in the application of lithium-ion batteries.
battery data, analysis and adjustment rules for
battery organization rules for services violating In [7] the author attempted a State-of-Charge (SOC)
the Apriori law algorithmic law. period balance plays a necessary role in energy
efficiency. This paper proposes the hire of a man-
In [5] the author suggested that prediction and made Neural Network (ANN) to install the Graphite
health management could be a promising way to / LiCoO2 pack of computer-charged lithium-ion
deal with the risks of previous failures and be pack. The MATLAB software package has been used
enforced in many well-known systems and battery to develop and monitor multiple network
systems. important for safe operation and configurations for better hardware of ANN. The
determining, battery capacity must be calculated results show that the Mean Square (MSE) error
exactly. During this time, we often use energy detected provides ANN as an effective procedure.
measurable details such as voltage, current, and Therefore, it predicts SOC battery bank rates with the
temperature profiles from the battery management accuracy of only the power supply, current, and
system have different patterns in cycles like aging. charging / discharge time.
According to this information, the link between
the power and charging profiles are read through ALGORITHM
neural networks. Specifically, accurately measure
health status tends to use feed forward neural Recurrent Neural Networks (RNN) [1] [2]:
network, convolutional neural network ,and a long Since weather data may be based on previous
STM. Our results show that process used for most data, the RNN model is better suited because
channels is based on voltage, current, and the Recurrent Neural Network is intended to
temperature profiles. surpasses the traditional manage sequential dependencies. The output
method that uses only a profile profile of up to from the previous state is transferred to the
25% -58% depending on the absolute absolute current state, which is why the hidden state in
error. the RNN remembers the sequence [1].

In [6] the author has tested Prognostic and health K-Nearest Neighbor classifier [1] [2]: One of
management (PHM) will ensure that the lithitum- the methods used for partitioning and
ion battery operates safely and retrieval. The item is split by a majority vote
honestly. The method used for PHM battery of its neighbors, the object is given the most
analysis is to use State of Health (SoH) and common category among the bits of the
Remaining useful Life (RUL) battery. Advances nearest neighbors. It's a lot used in real-life
in computer tools and large-scale knowledge situations because it is an abstract definition
algorithms have moved into a brand new era of does not make any basic assumptions about
data-driven data analysis methods, aggressive distribution data.
machine learning algorithms .This paper presents
the first development of data-driven predictions,
using Deep Neural Networks (DNN) predictive
method as well as RUL of lithium-ion battery. The
operation of the planned method was applied to a
 hidden layer hidden layer

input layer a1 ky 1

Time - _ of 2 2 oilgalie

Current, \/ x 3 3 - \ SoC

0.8 + = cluster 1
g cluster 2
t 0.6 QO cluster 3
§ cluster 4
£04 cluster 5
8 cluster 6
a 0.2
The ANN regression algorithm consists of
one pass and one pass. In advanced
transmission, the output of neurons in each
oN 4 06
Average current (normalized) 4 04 layer is calculated layer by sequence, from the
0.2 Average voltage (normalized)
00 lowest layer with the vector of data as input.
The results of the final layer are then
compared to the output vector you want to
Vector Machine Support [2] [3]: The support vector
detect the error of each neuron in the final
machine is a Modeled Machine Learning Model
layer. The findings of those errors are then
that uses separation problems. After providing
distributed backwards by weights to make the
SVM model data sets for training data with a two-
error detection of neurons in the previous
stage label, they are able to differentiate new
layer. When a fault detection is inserted into
models. It works based on decision planes that
a neuron the metals that lead to that neuron
define boundaries.The decision plane separates one
can be replaced so that the error can be
item from another object of different classes. The
reduced. While backtracking works well for
data points are very close to the hyper plane called
some problems it works poorly for them
support vectors [9]. Kernel function is used for
many others where they fail to get better
partitioning non-linear data by converting input to
results with them many layers. Spreading
maximum magnitude Space .The Gaussian radial
back to most layered ANNs is also common
base function kernel is used in the model K (Xi, Xj)
adhere to local integrity at the same time as
= e- Iki, xj | [2/2 omega 2 when K (Xi, X |) = input
the level of learning it's slow. Although the
vectors in input space, | Xi, Xj | 2 = maximum
above-mentioned benefits of deep
space size X and Y Addition and a is a free
architecture, they have only recently been due
parameter.
to the lack of a better training method that has
not been widely discussed in the_in textbook.

Artificial Neural Networks (ANN) can be a process

CONCLUSION
[1] [6] [7]: a model made up of a collection of
individual process units, artificial reality neurons,
Hence after discussion of many research
linked to weights that can be adjusted by standard
papers and learning various algorithms used
parameters that check proximity within the
for Lithium ion battery failure prediction, and
prescribed response so those obtained by ANNs
best fitted predictive data analytics, we
approximators worldwide can mimic any nonlinear
conclude that there are various algorithms
partner precision required Networks are arranged in
with good efficiency based on different data
layers, with the main layer inserting the input and
sets used accordingly.
therefore the last layer production results. The
middle layers have no reference to the outside
world, which is why they are known as hidden
layers. true neurons are an integral part of AN AN
because of are charged for connecting all layers.
Model for somatic cell is shown in Figure
 REFERENCES

[1] Hu, Xiaosong & Li, Shengbo & Yang, Yalian.

(2015). Advanced Machine Learning Approach for
Lithium-lon Battery State Estimation in Electric
Vehicles. IEEE Transactions on Transportation
Electrification. 2. 1-1.10.1109/TTE.2015.2512237.

[2] C. Vidal, P. Malysz, P. Kollmeyer and A.

Emadi, "Machine Learning Applied to Electrified
Vehicle Battery State of Charge and State of Health
Estimation: State-ofthe-Art,”" in EEE Access, vol. 8,
pp. 52796-52814, 2020,
doi: 10.1109/ACCESS.2020.2980961.

[3] Severson, K.A., Attia, P.M., Jin, N. etal. Data-

driven prediction of battery cycle life before
capacity degradation. Nat Energy 4, 383-391
(2019).
https: //doi.org/10.1038/s4 1560-01 9-0356-8

[4] Zhi Yang and Jianhua Wu 2020 J. Phys.: Conf.

Ser. 1631 012144

[5] Choi, Yohwan & Ryu, Seunghyuong & Park,

Kyungnam & Kim, Hongseok. (2019). Machine
Learning-Based Lithiumlon Battery Capacity
Estimation Exploiting Multi-Channel Charging
Profiles. IEEE Access. PP. 1-1.
10.1109/ACCESS.2019.2920932.

[6] Knumprom, P. Yodo, N. A _ Data-

DrivenPredictive Prognostic Model for Lithtumion
Batteries based on a Deep Learning Algorithm.
Energies 2019, 12(4), 660;
https: //doi.org/10.3390/en12040660.
https:/A vww.mdpi.com/1996-1073/12/4/660

[7] Almeida, G.; Souza, A.; Ribeiro, P. A Neural

Network Application for a Lithiumlon Battery Pack
State-of-Charge Estimator with Enhanced
Accuracy. Proceedings 2020, 58(1), 33;
https://doi.org/10.3390A WEF-06915.
https:/Avww. mdpi.com/2504-3900/58/1/33