Resource-Efficient Monitoring of Energy Storage Systems During

Transport and Storage: A Data-Driven Approach to Early Short

Circuit Detection

Christoph Schrade

, Theo Zschörnig

, Leonard Kropkowski

and Bogdan Franczyk

1,2

Information Systems Institute, Leipzig University, Grimmaische Str. 12, 04109 Leipzig, Germany

Institute for Applied Informatics (InfAI), Goerdelerring 9, 04109 Leipzig, Germany

Fraunhofer-Institut für Nachrichtentechnik, Heinrich-Hertz-Institut HHI, Einsteinufer 37, 10587 Berlin, Germany

Keywords: Internal Short Circuit, Energy Storage System, Battery Electric Vehicle, Battery Energy Storage System, Data

Driven Early Detection.

Abstract: Due to national and international laws and regulations, the number of energy storage systems has risen sharply

in recent years. While battery systems in operation can often be monitored by installed battery management

systems to ensure safe operation, there are still no standardized monitoring methods for batteries during

transport or storage. Consequently, this article proposes a solution for monitoring such batteries in the typical

logistic processes of storage and transport. Particular attention is paid to a resource-efficient implementation

of a data-driven algorithm that is adopted from existing literature and enables the early detection of internal

short circuits, which are the main cause of thermal runaways of battery storage systems. As the transmission

frequency of an external monitoring device is a particularly resource-critical variable, the extent to which

different data frequencies influence the detection performance is also investigated.

1 INTRODUCTION

Energy storage systems (ESS) play a crucial role in

energy transition initiatives worldwide. The main

goal of this transition is to reduce energy

consumption as well as greenhouse gas emissions but

also to increase the utilization of renewable energy

sources. In this regard, ESS enable the storage of

renewable energy, which is often generated

irregularly, therefore making electricity supply more

sustainable and flexible. Two main application areas

of ESS are in battery electric vehicles (BEVs) as well

as battery energy storage systems (BESS) for

residential and commercial applications. Their

importance in the energy transition is clearly reflected

in increasing sales figures over the last years.

However, the use of ESS is always associated with

logistical tasks such as transportation to the

application or production site, the return at the end of

life (EoL) for reuse or recycling as well as their

storage at different stages of their lifecycle. During

these periods, continuous monitoring of ESS is

necessary due to the sensitivity of the integrated

batteries, the resulting safety risks and for ongoing

quality assurance. In this context, internal short

circuits (ISCs) of battery cells in particular are a

major source of danger. Research to date already

offers promising artificial intelligence (AI)-based

approaches that can recognize these short circuits.

However, these require the constant availability of the

battery management system (BMS), which is not

reliably possible when transporting or storing ESS. In

addition, the effect of reducing the monitoring

frequency on the performance of these approaches

has not been investigated. Against this background,

we propose an approach, which can overcome this

challenge thus enabling remote early battery short

circuit detection for ESS in logistics scenarios.

The remainder of this paper is structured as

follows: Section 2 outlines the research motivation,

background, and challenges. Section 3 reviews

related work. Section 4 describes the proposed

solution. Section 5 presents experiments on inference

frequencies for early short circuit detection, with

results discussed in Section 6. Finally, Section 7

summarizes findings and suggests future research

directions.

Schrade, C., Zschörnig, T., Kropkowski, L. and Franczyk, B.

Resource-Efﬁcient Monitoring of Energy Storage Systems During Transport and Storage: A Data-Driven Approach to Early Short Circuit Detection.

DOI: 10.5220/0013291900003929

In Proceedings of the 27th International Conference on Enterprise Information Systems (ICEIS 2025) - Volume 1, pages 781-788

ISBN: 978-989-758-749-8; ISSN: 2184-4992

781

2 BACKGROUND

ESS in BEVs as well as BESS in residential and

commercial buildings are almost exclusively utilizing

a lithium-based battery chemistry (IEA, 2022; Marsh,

2023). Under certain conditions, such as

overcharging, overheating or mechanical damage,

lithium-ion based batteries can catch fire or explode.

This is due to the uncontrolled release of their energy

in a short amount of time. This phenomenon is called

thermal runaway and describes an uncontrolled and

exponential increase in the temperature inside the

battery, which may result in serious accidents.

Following Shahid and Agelin-Chaab (2022), the three

major reasons for thermal runaway to start are

mechanical (R1), electrical (R2) and temperature

abuse (R3), all of which typically lead to an ISC.

However, the speed at which an ISC can occur varies

depending on the type of abuse and its severity (P.

Sun et al., 2020). For this reason, early detection of

an ISC is a particularly important aspect of battery

safety and a vivid research topic.

Past research has already yielded a number of

promising AI-based approaches that can detect ISCs

at an early stage. However, these approaches expect

the constant availability of the BMS. However, steady

access to the BMS cannot be guaranteed away from

their place of operation. This significantly increases

the difficulty for safely handling ESS in logistics

scenarios since the potential risk status of individual

batteries cannot be reasonably monitored during pre-

and after-sales processes by stakeholders such as

logistics service providers or freight forwarders.

One possible approach to solve this problem is the

use of an external device that can access the data from

the BMS, while the underlying ESS is not in

operation. Past research has already looked into the

design and application of such a device, but focused

solely on its utilization during the first life of an ESS

at its place of operation. This work aims to bridge this

gap, which has been acknowledged by current

research projects (Plotnikov et al., 2023), and provide

an approach that enables monitoring ESS in BEVs as

well as BESS in residential and commercial buildings

aside from their place of operation, especially during

transportation and storage.

A key challenge when using an external device to

record the BMS and environmental data of an ESS is

its power supply. Especially in logistics scenarios,

there are no fixed power sources. Moreover,

additional framework conditions have to be observed

when designing the external device and must be

considered when solving the overall problem:

● High energy usage for data transmission

(FC1) (Jayakumar et al., 2014)

● High energy usage for complex

computations (FC2) (Tekin et al., 2023).

● Continuous operational readiness (FC3)

(Callebaut et al., 2021)

● Compact design and limited battery

capacity (FC4) (Callebaut et al., 2021;

Jayakumar et al., 2014)

● Longevity and low maintenance (FC5)

(Callebaut et al., 2021; Jayakumar et al.,

2014).

In view of these conditions, it is necessary to

develop an approach that balances the need for close

monitoring with the limitations of the necessary

battery operation.

Against this background, this work presents an

approach that enables the collection of BMS data,

namely voltage readings of all battery cells, and

environmental data for ISC early detection using an

external device to supplement an ESS during

logistical processes such as transport and storage.

Moreover, the main research question of this paper is:

Is early ISC detection possible using a low

monitoring data frequency? In order to answer this, a

promising approach for early ISC detection from the

literature (Schmid & Endisch, 2022; Schmid et al.,

2022; Schmid, Kneidinger, & Endisch, 2021;

Schmid, Liebhart, et al., 2021) is adopted and the

results from respective research articles concerning

the detection time for ISCs are validated in several

experiments. In addition, the approach is compared to

similar ones. Finally, the effect of the monitoring data

frequency on the detection performance of the chosen

approach is investigated.

3 STATE OF THE ART

Battery fault detection methods can be divided into

three different classes: threshold based, model-based

and data-driven methods (Schmid, Kneidinger, &

Endisch, 2021; Shang et al., 2024). There are also

different types of features that are used; some of them

directly measurable, e.g. voltage and current, others

not, e.g. state of charge (SoC) and capacity or state of

health (SoH).

Threshold based methods follow the approach of

defining a critical threshold for directly measurable

features, which, when exceeded or undercut, indicate

a fault. Model-based approaches attempt to estimate

features that cannot be measured directly and then use

them for error detection. The idea behind data-driven

ICEIS 2025 - 27th International Conference on Enterprise Information Systems

782

methods is to use techniques from mathematical

statistics to derive regularities and patterns from

battery data recorded during operation.

In the early stages of an ISC with relatively high

short circuit resistance the effect on the

aforementioned battery features is rather small ("soft"

ISC) which makes it very difficult to detect the fault

at this early stage by directly introducing critical

threshold values (Lai et al., 2020; Schmid & Endisch,

2022). Thus, threshold-based approaches are hardly

the correct tool for early ISC detection.

The review Shang et al. (2024) gives a thorough

treatment of the most recent research literature

concerning model-based and data-driven approaches.

As noted in Schmid et al. (2022), model-based

approaches mostly suffer from the drawback that the

reliability of feature estimations cannot really be

secured. Therefore, the present work focuses on data

driven methods.

In the literature many different data driven and

machine learning approaches are described: Isolation

Forest (Jiang et al., 2022), Support Vector Machines

(Yao et al., 2021) various neural network

architectures, for example LSTM and Radial Basis

Function neural networks (Ojo et al., 2021; Wang et

al., 2021) and Local Outlier Factors (Z. Sun et al.,

2022) can all be found as employed detection

methods.

A major challenge in the early detection of ISC is

the robustness against noise in the sensor data

(Schmid & Endisch, 2022; Schmid, Kneidinger, &

Endisch, 2021; Shang et al., 2024). This makes

Principal Component Analysis (PCA) a good

detection approach.

One disadvantage of PCA is that the projection

used for dimension reduction is purely linear, which

means that non-linear structures in the data may be

lost in the process. This makes PCA an unsuitable

tool for non-linear variations data such as cell level

voltage data of a battery in low SoC range (Schmid &

Endisch, 2022; Schmid et al., 2022; Schmid,

Liebhart, et al., 2021).

In Schmid and Endisch (2022), Schmid, Liebhart,

et al. (2021) and Schmid et al. (2022) the non-

linearity problem is tackled by using a non-linear

extension of PCA, the kernel PCA (KPCA) originally

introduced in Schölkopf et al. (1997). Solving both

the non-linearity and the sensor noise issue makes the

KPCA model from Schmid and Endisch (2022),

Schmid, Liebhart, et al. (2021) and Schmid et al.

(2022) a very promising approach for early ISC

detection in our use case scenario.

None of the aforementioned methods have been

applied for lower data frequencies. Moreover, the

review we conducted suggests that there is no

research that addresses the issue of whether data

frequency during the monitoring phase has any

impact on ISC detection performance.

The scientific approaches to tracking systems

developed to detect ISCs were also investigated. In

this regard the authors of González et al. (2022)

conducted a comprehensive literature review which

did not find any previous works concerning tracking

systems using IoT technology. In Haldar et al. (2024)

a real time tracking system for the SoC and SoH of

three-wheeled battery-operated vehicles is

introduced. The external device used for data

collection uses the batteries to be monitored as its

power source. The collected data is sent to a cloud-

based backend for processing. In Gupta et al. (2020)

a similar approach is presented for monitoring

batteries in BEVs with an external device that uses a

battery power supply and transmits data regarding the

SoC and SoH to a cloud backend.

However, all the reviewed approaches are

developed for operation during the first-life use of the

monitored batteries and at their place of operation.

More importantly, early detection of short circuits is

not carried out in any of the reviewed works.

4 SOLUTION PROPOSAL

In the following, we present our approach to monitor

ESS in logistics scenarios such as transport and

storage. The approach aims to address the challenges

and framework conditions described in section 2. The

central assumption when designing a solution for the

given context is that BMS and environmental data can

only be collected by an external device that has its

own power supply in the form of a battery. The main

goal of the approach is to enable continuous ISC early

detection in ESS, especially in logistics scenarios. In

this context, the FCs mentioned in section 2 lead to

several requirements (REs):

● RE1 (derived from FC1, 4 and 5): The

approach should be able to early-detect ISCs

using only low frequency BMS data.

● RE2 (FC2): Computations should be

offloaded from the external device as much

as possible.

● RE3 (FC3): The approach should allow

continuous monitoring.

Looking at the root causes for thermal runaway, it

is assumed that any form of abuse of a battery leads

to an increased risk of an ISC. The approach should

Resource-Efﬁcient Monitoring of Energy Storage Systems During Transport and Storage: A Data-Driven Approach to Early Short Circuit

Detection

783

reflect this, hence additional requirements can be

derived:

● RE4: The approach should be able to detect

mechanical (R1), electrical (R2) and thermal

(R3) abuse.

● RE5: The approach should be able to change

data collection frequency according to

predefined conditions.

Looking at the requirements, RE 1 is most

important for the feasibility of the proposed approach.

Since the transmission of BMS data from the external

device to a receiver has a high energy demand, the

frequency at which the device sends its data to a

receiver is a critical variable. This must be set as low

as possible without impairing the monitoring of the

ESS. This requirement also corresponds with the

main research question of the present article, if early

ISC detection is possible with low monitoring data

frequency. In this context, the literature review

suggests using a data driven AI detection approach.

Against this background, a series of experiments are

conducted in Section 5 to examine which approaches

are suitable and what effects different data

frequencies have on the performance of the selected

approaches.

RE 2 is addressed in the presented approach by

introducing 2 different layers for data processing:

● The edge layer that comprises the external

device and enables the collection of BMS

and environmental data.

● The cloud layer, which receives and

processes the collected data from the edge

layer.

In this context, we propose a variable data

collection frequency in the edge layer, addressing RE3

and RE5. Under normal conditions, a baseline

frequency minimizes power consumption from

processor load and data transmission. However,

external factors like movement, vibration, or

temperature changes trigger an adaptive increase in

detection frequency. This requires sensors to monitor

environmental variables such as speed, rotational

movement, and temperature. Rule-based logic enables

the detection of potential abuse, addressing RE4.

The cloud layer is horizontally scalable for data

ingestion, processing, and distribution, leveraging

established IoT and big data technologies.

Communication between the edge and cloud layer

uses mobile standards (GSM, LTE, 5G) and

lightweight protocols (e.g., MQTT, CoAP, AMQP).

https://kafka.apache.org/

https://hadoop.apache.org/

Data processing involves two components: model

training and inference. To monitor batteries

throughout their lifecycle, training data must capture

normal behavior across the cycle. As lab data for new

battery types may be limited, models require

continuous retraining to improve predictions.

Inference occurs in real time to identify critical

batteries promptly and notify stakeholders. Therefore,

the approach employs a lambda architecture, with

periodic training in the batch layer and real-time

inference in the streaming layer. Example

technologies to implement this are Apache Kafka

Hadoop

, Apache Spark

or Apache Storm

5 EXPERIMENTS

In order to evaluate the feasibility of our approach, a

series of experiments were conducted to examine the

performance of early detection approaches with

reduced data frequencies. In this context, a real

battery with 6 cells connected in series was artificially

short-circuited in a laboratory environment. This was

carried out with three different resistors as the short

circuit resistor (10Ω, 1kΩ, 10kΩ). Moreover, several

different data frequencies were implemented for the

inference phase, i.e., the phase after the induction of

the short circuit (c. f. section 5.2). Each experiment

was started with an initial phase in which the properly

functioning cells were cycled over a period of time to

generate data for training the model. After this, an

external short circuit (ESC) was induced for one of

the cells by implementing a load resistor and the time

required by the detection algorithm to detect the short

circuit was measured. In this regard, an ESC was

triggered instead of an ISC for reasons of

practicability. Although the behavior of an ESC is not

identical to that of an ISC, Zhang et al. (2017)

nevertheless states that an ESC can mimic the early

phase of an ISC.

In Schmid and Endisch (2022), Schmid et al.

(2022), Schmid, Kneidinger, and Endisch (2021) and

Schmid, Liebhart, et al. (2021) the principal idea for

ISC detection is to look at relations between the single

cell voltages of a system. As discussed in section 3,

the mathematical method that forms the basis of

Schmid and Endisch (2022), Schmid et al. (2022),

Schmid, Kneidinger, and Endisch (2021) and Schmid,

Liebhart, et al. (2021) is principal component

analysis (PCA), which is optimized and exploited in

many different ways to perfectly suit the problem. In

https://spark.apache.org/

https://storm.apache.org/

ICEIS 2025 - 27th International Conference on Enterprise Information Systems

784

particular in Schmid and Endisch (2022), Schmid,

Liebhart, et al. (2021) and Schmid et al. (2022) the

authors utilize the non-linear extension kernel

principal component analysis (KPCA) which makes

it possible to address non-linear relationships in the

data. The line of attack for fault detection is to check

how close a given cell’s voltages vector lies, after a

transformation, to the principal component space that

was computed using training data. More precisely the

so called T

- and Q-test statistics values of the vector

of cell voltages are computed while monitoring. The

general principle is that low T

- and Q-values

correspond to data which is similar to the training data

while high values display anomalous behavior.

We compared the detection performance of the

KPCA (A

) approach with respect to short circuit

detection time with all other approaches found in the

scientific literature, which also only require the

voltage values of individual battery cells for the

detection. More precisely, plain PCA (Schmid,

Kneidinger, & Endisch, 2021) (A

) and a very simple

method that just tracks the voltage difference between

the cells (Lai et al., 2020) (A

) were evaluated as well.

5.1 Experimental Setup

The batteries utilized in this study are Samsung

INR18650-32E cells, each with a nominal capacity of

3.2Ah. These cells feature a lithium-nickel- cobalt-

aluminum oxide cathode paired with a graphite

anode. The cells have been arranged in a series

configuration using 3D-printed cell holders.

Individual monitoring of the cell voltage has been

implemented using a Gantner Q.bloxx XL A107

The cycling of the cells is managed using an EA-PSB

10080-120 power supply in conjunction with custom

software tailored to administer the dynamic cycling

protocol. To minimize the impact of external

temperature fluctuations on short circuit detection,

the entire experimental setup was housed within a

thermal chamber maintained at 35°C.

5.2 Training

The detection algorithm was tested on ESC experiment

datasets using 10Ω (R1), 1kΩ (R2), and 10kΩ (R3)

resistors to induce short circuits. The 10kΩ resistor

caused such a slow short circuit that the experiment

was halted after several hours without detections,

leading to the 10kΩ setting being discarded.

While the authors in Schmid and Endisch (2022)

and Schmid et al. (2022) evaluated their approach

https://www.gantner-instruments.com/de/produkte/bloxx/

with a fixed inference data frequency of 10Hz

(Schmid & Endisch, 2022) resp. 0.1Hz (Schmid et al.,

2022), in this work the detection time for all resistor

values was evaluated for different inference data

frequencies of 1Hz (F





Hz (F





Hz (F





) and





Hz (F

Since all approaches (A

-A

) were evaluated

using the data sets from two different resistor settings

and R

) while applying five different monitoring

frequencies (F

-F

) 30 different experimental settings

were studied.

To ensure the training dataset adequately

represented the entire initial cycling phase without

being overly large, 1000 evenly distributed points

were selected. The data was first downsampled from

10Hz to 1Hz by averaging. Data points used for

inference were excluded from the training set. For R

and R

, the starting inference data point (T

) was

chosen about 30 minutes before inducing the short

circuit.

The critical threshold for T²- and Q-values in the

KPCA approach was set at the 0.999-quantile of

training data values. To address fluctuations in testing

data, especially at higher monitoring frequencies, the

detection logic required unusually high T²- and Q-

values to persist for 10 minutes. For the highest

frequencies (1 Hz and





Hz), the approach was

further refined by requiring at least 20% of values in

the last 10 minutes to be critical to detect a short

circuit. For lower frequencies, this adjustment was

unnecessary, as only one data point is transmitted

every 5, 10, or 15 minutes.

5.3 Results

With the aforementioned detection logic

implemented, the following detection times shown in

Table 1 were achieved using the KPCA approach.

Table 1: Detection times of the KPCA approach using

experimental settings A

1,2

1-5

1Hz

𝟏

𝟔𝟎

𝟏

𝟑𝟎𝟎

𝟏

𝟔𝟎𝟎

𝟏

𝟗𝟎𝟎

10Ω 2min

7sec

1min 6min 1min 1min

1kΩ 151

min

133

min

103

min

253

min

Resource-Efﬁcient Monitoring of Energy Storage Systems During Transport and Storage: A Data-Driven Approach to Early Short Circuit

Detection

785

The experiments were carried out again with the

same detection logic for short circuit detection, but

this time with PCA as the basis of the algorithm. In

this context, identical detection times were observed

for the 10Ω resistor (c. f. Table 2). However, the PCA

approach did not work for the 1kΩ resistor. A short

circuit could not be detected using any of the

monitoring frequencies.

Table 2: Detection times of the PCA approach using

experimental settings A

1-5

1Hz

𝟏

𝟔𝟎

𝟏

𝟑𝟎𝟎

𝟏

𝟔𝟎𝟎

𝟏

𝟗𝟎𝟎

10Ω 2min

7sec

1min 6min 1min 1min

Finally, another approach was utilized in which a

short circuit was considered to have been detected if

the maximum difference between the individual cell

voltages was greater than 0.5 volts (Lai et al., 2020).

In this regard, the detection times for the 10Ω

experiment were around 20 minutes for all

frequencies. The approach did not detect anything for

higher short circuit resistances.

6 DISCUSSION

Looking at the results of the conducted experiments,

we conclude that the employed data driven methods

can be utilized to enable early ISC detection with low

monitoring frequencies. Specifically, the KPCA

approach (A

) yielded promising detection times

compared to the other two approaches. Additionally,

the results of detection times for the experimental

settings A

1,2

1-5

are of the same order of magnitude

as those in Schmid and Endisch (2022) and Schmid et

al. (2022), which supports the validity of our results

regarding the detection times. The unsuitability of the

PCA approach (A

) for high short circuit resistances

) has already been discussed in Schmid et al.

(2022). The voltage-difference approach (A

) does

not yield successful results for (R

) either.

The results also show that there is no clear linear

trend in the relationship between monitoring

frequency and detection time in any of the

experimental settings. This suggests that the detection

times strongly depend on the distribution of selected

data points, meaning they could vary with different

data selections. In order to reinforce this assumption,

the experimental settings A

1,2

1-5

were used again

for detection with an initial time offset of T

minutes. The results show that detection times for

lower frequencies (<=





Hz) increase up to 13 times,

which further highlights that detection latency for low

frequencies is highly influenced by the reception time

of the data points.

Table 3: Detection times of the KPCA approach using

experimental settings A

,2F

1-5

with T

+7.

1Hz

𝟏

𝟔𝟎

𝟏

𝟑𝟎𝟎

𝟏

𝟔𝟎𝟎

𝟏

𝟗𝟎𝟎

10Ω 2min

7sec

1min 8min 8min 13

min

1kΩ 151

min

199

min

144

min

Concerning the results of experimental settings

1-2

1,2

1-5

we recognize that higher monitoring

frequencies were not always faster in detecting ISCs

compared to lower frequencies. This is in contrast to

the naive expectation that a higher frequency enables

faster detection. However, the detection logic seems

to be a factor here - at higher frequencies, statistical

value (T²- and Q-values) fluctuations are more

pronounced, whereas these fluctuations are less

pronounced at lower frequencies. Therefore, the

choice of a suitable detection logic is strongly

frequency-dependent. In this study, a similar

detection logic was used across all frequencies F

1-5

enhance the comparability of the results. For practical

applications, however, this means that different,

frequency-optimized detection logics should be

implemented.

In conclusion, our experiments show that using a

data-driven approach such as KPCA is important to

enable our solution proposal for low-frequency

monitoring, energy-efficient, scalable battery

monitoring. However, the precise settings of

monitoring frequencies for different conditions must

be determined for each battery type individually as

the time from damage or abuse to thermal runaway

depends on various factors, e.g. capacity (Zhao et al.,

2016) or environmental influences (Ji et al., 2021).

Therefore, the generalization from the conducted

experiments to other battery configurations or real-

world scenarios is not given and needs to be

investigated in future research. Finally, the ISC early

detection model must be trained anew for each

different battery type.

ICEIS 2025 - 27th International Conference on Enterprise Information Systems

786

7 CONCLUSION & OUTLOOK

This paper proposes a method for remote early

detection of battery short circuits in ESS for logistics.

It outlines the research background, motivation, and

related work. While current ISC detection using BMS

data works in controlled settings, it lacks support

during transport and storage. Key requirements

include low-frequency data collection to save battery

life, cloud-based computation, and adaptive

monitoring based on factors like temperature or

movement.

The proposed solution involves a battery-powered

external device for remote monitoring, collecting

battery and environmental data, and a two-layer

architecture: an edge layer for IoT-based data

collection and a cloud layer for scalable analysis and

real-time ISC alerts. The approach presented extends

the state of the art by demonstrating a way of

recognizing ISCs of ESS at an early stage in situations

that were not previously considered in scientific

literature.

In order to evaluate the approach, experiments

with a 6-cell battery setup and artificially induced

short circuits (10Ω, 1kΩ, 10kΩ resistors) to simulate

early ISC phases were carried out. Voltage data was

analyzed using KPCA for anomaly detection, which

proved effective across different frequencies and

outperformed both standard PCA and simpler voltage

tracking methods.

Future research needs to investigate how

analyzing data across the entire lifecycle of a battery

could refine detection logic and improve accuracy.

Moreover, the experiments should be carried out in

more realistic environments and with different battery

types so that the transferability of the approach can be

evaluated. Finally, the computational costs of

different detection approaches should be compared

with each other.

ACKNOWLEDGEMENTS

The work presented in this paper is partly funded by

the German Federal Ministry for Economic Affairs

and Climate Action (BMWK 16TNW0016D) as well

as by the German Federal Ministry of Education and

Research (BMBF 02J21E022).

REFERENCES

Callebaut, G., Leenders, G., van Mulders, J., Ottoy, G.,

Strycker, L. de, & van der Perre, L. (2021). The Art of

Designing Remote IoT Devices-Technologies and

Strategies for a Long Battery Life. Sensors (Basel,

Switzerland), 21(3).

González, I., Calderón, A. J., & Folgado, F. J. (2022). IoT

real time system for monitoring lithium-ion battery

long-term operation in microgrids. Journal of Energy

Storage, 51, 104596.

Gupta, V., Sharma, N., Maram, D., & Priyadarshi, H.

(2020). IOT enabled data acquisition system for electric

vehicle. In AIP Conference Proceedings, A Two-Day

Conference on Flexible Electronics For Electric

Vehicles (p. 40002). AIP Publishing.

Haldar, S., Gol, S., Mondal, A., & Banerjee, R. (2024). IoT-

enabled advanced monitoring system for tubular

batteries: Enhancing efficiency and reliability. E-Prime

- Advances in Electrical Engineering, Electronics and

Energy, 9, 100709.

IEA. (2022). Global Supply Chains of EV Batteries.

https://www.iea.org/reports/global-supply-chains-of-

ev-batteries

Jayakumar, H., Lee, K., Lee, W. S., Raha, A., Kim, Y

[Younghyun], & Raghunathan, V. (2014). Powering the

internet of things. In Y. Xie, T. Karnik, M. M. Khellah,

& R. Mehra (Eds.), ISLPED'14: Proceedings of the

2014 ACM Conference on the International Symposium

on Low Power Electronics and Design : August 11-13,

2014, La Jolla, CA, USA (pp. 375–380). ACM.

Ji, H., Chung, Y.‑H., Pan, X.‑H., Hua, M., Shu, C.‑M., &

Zhang, L.‑J. (2021). Study of lithium-ion battery

module’s external short circuit under different

temperatures. Journal of Thermal Analysis and

Calorimetry, 144(3), 1065–1072.

Jiang, J., Li, T., Chang, C., Yang, C., & Liao, L. (2022).

Fault diagnosis method for lithium-ion batteries in

electric vehicles based on isolated forest algorithm.

Journal of Energy Storage, 50, 104177.

Lai, X., Yi, W., Kong, X., Han, X., Zhou, L., Sun, T., &

Zheng, Y. (2020). Online detection of early stage

internal short circuits in series-connected lithium-ion

battery packs based on state-of-charge correlation.

Journal of Energy Storage, 30, 101514.

Marsh, J. (2023). Comparing lithium-ion battery

chemistries. https://www.energysage.com/energy-

storage/types-of-batteries/comparing-lithium-ion-

battery-chemistries/

Ojo, O., Lang, H., Kim, Y [Youngki], Hu, X [Xiaosong],

Mu, B., & Lin, X. (2021). A Neural Network Based

Method for Thermal Fault Detection in Lithium-Ion

Batteries. IEEE Transactions on Industrial Electronics,

68(5), 4068–4078.

Plotnikov, M., Schreynemackers, P., Joachimsthaler, C.,

Jarmer, J.‑P., Kuhlmann, G., Karlstedt, F., Kropkowski,

L., Kürpick, C., Müller, S., & Schrade, C. (2023).

Batterie-Logistik. Whitepaper.

Schmid, M., & Endisch, C. (2022). Online diagnosis of soft

internal short circuits in series-connected battery packs

Resource-Efﬁcient Monitoring of Energy Storage Systems During Transport and Storage: A Data-Driven Approach to Early Short Circuit

Detection

787

using modified kernel principal component analysis.

Journal of Energy Storage, 53, 104815.

Schmid, M., Kleiner, J., & Endisch, C. (2022). Early

detection of Internal Short Circuits in series-connected

battery packs based on nonlinear process monitoring.

Journal of Energy Storage, 48, 103732.

Schmid, M., Kneidinger, H.‑G., & Endisch, C. (2021).

Data-Driven Fault Diagnosis in Battery Systems

Through Cross-Cell Monitoring. IEEE Sensors

Journal, 21(2), 1829–1837.

Schmid, M., Liebhart, B., Kleiner, J., Endisch, C., &

Kennel, R. (2021). Online Detection of Soft Internal

Short Circuits in Lithium-Ion Battery Packs by Data-

Driven Cell Voltage Monitoring. In 2021 IEEE 12th

Energy Conversion Congress & Exposition - Asia

(ECCE-Asia) (pp. 1711–1718). IEEE.

Schölkopf, B., Smola, A., & Müller, K.‑R. (1997). Kernel

principal component analysis. In G. Goos, J. Hartmanis,

J. van Leeuwen, W. Gerstner, A. Germond, M. Hasler,

& J.-D. Nicoud (Eds.), Lecture Notes in Computer

Science. Artificial Neural Networks — ICANN'97 (Vol.

1327, pp. 583–588). Springer Berlin Heidelberg.

Shahid, S., & Agelin-Chaab, M. (2022). A review of

thermal runaway prevention and mitigation strategies

for lithium-ion batteries. Energy Conversion and

Management: X, 16, 100310.

Shang, Y., Wang, S., Tang, N., Fu, Y., & Wang, K. (2024).

Research progress in fault detection of battery systems:

A review. Journal of Energy Storage, 98, 113079.

Sun, P., Bisschop, R., Niu, H., & Huang, X. (2020). A

Review of Battery Fires in Electric Vehicles. Fire

Technology, 56(4), 1361–1410.

Sun, Z., Wang, Z., Liu, P., Qin, Z., Chen, Y., Han, Y.,

Wang, P., & Bauer, P. (2022). An Online Data-Driven

Fault Diagnosis and Thermal Runaway Early Warning

for Electric Vehicle Batteries. IEEE Transactions on

Power Electronics, 37(10), 12636–12646.

Tekin, N., Acar, A., Aris, A., Uluagac, A. S., & Gungor, V.

C. (2023). Energy consumption of on-device machine

learning models for IoT intrusion detection. Internet of

Things, 21, 100670.

Wang, J., Zhang, S., & Hu, X [Xia] (2021). A Fault

Diagnosis Method for Lithium-Ion Battery Packs Using

Improved RBF Neural Network. Frontiers in Energy

Research, 9, Article 702139.

Yao, L., Fang, Z., Xiao, Y., Hou, J., & Fu, Z. (2021). An

Intelligent Fault Diagnosis Method for Lithium Battery

Systems Based on Grid Search Support Vector

Machine. Energy, 214, 118866.

Zhang, M., Ouyang, M., Lu, L., He, X., Feng, X., Liu, L.,

& Xie, X. (2017). Battery Internal Short Circuit

Detection.

ECS Transactions, 77(11), 217–223.

Zhao, R., Liu, J., & Gu, J. (2016). Simulation and

experimental study on lithium ion battery short circuit.

Applied Energy, 173, 29–39.

ICEIS 2025 - 27th International Conference on Enterprise Information Systems

788