INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING, MANAGEMENT &
APPLIED SCIENCE (IJLTEMAS)
ISSN No. 2278-2540 | DOI: 10.51584/IJLTEMAS | Volume XV Issue I January 2026
Page 1
www.rsisinternational.org
Predictive Health Monitoring Systems for Electric Vehicle
Powertrains Using Edge AI and CAN Bus Data
Dhage Abhishek Yuvraj; Aniket Atresh; Prof. Urmila Burde
Department of Electronics and Telecommunication Engineering, Ajeenkya DY Patil School of
Engineering, Pune, India
Department of Electronics and Telecommunication Engineering, Ajeenkya DY Patil School of
Engineering, Pune, India
Asst. Professor, Department of Electronics and Telecommunication Engineering, Ajeenkya DY Patil
School of Engineering, Pune, India
DOI:
https://doi.org/10.51583/IJLTEMAS.2026.150100001
Received: 22 December 2025; Accepted: 27 December 2025; Published: 21 January 2026
ABSTRACT
Electric vehicles (EVs) are becoming increasingly important in the shift toward sustainable mobility. While
their adoption is accelerating, ensuring the health and reliability of EV powertrains remains a critical challenge.
Failures in subsystems such as batteries, motors, and controllers may cause unexpected breakdowns, reduced
efficiency, and safety issues. Predictive Health Monitoring (PHM) systems aim to prevent such failures by
identifying anomalies before they escalate. Traditional PHM solutions often depend on cloud platforms, but
these face drawbacks such as latency, high bandwidth requirements, and privacy risks. To address these
challenges, edge-based PHM employs embedded devices to process Controller Area Network (CAN) bus data
locally, enabling real-time diagnostics, enhanced privacy, and cost efficiency.
This paper presents a structured survey of PHM techniques for EV powertrains using Edge Artificial
Intelligence (Edge AI) and CAN bus data. The contributions include: (i) classification of PHM approaches into
model-based, data-driven, security- focused, and edge-deployed methods, (ii) comparative analysis of recent
works, (iii) a summary table of surveyed papers, (iv) discussion of hardware implementations reported in
literature, and (v) identification of future research directions.
Key Words - Electric Vehicles, Predictive Maintenance, Edge AI, CAN Bus, Vehicle Diagnostics, Anomaly
Detection.
INTRODUCTION
The transportation sector is witnessing a global transition toward electric mobility. EVs are valued for their
efficiency, reduced environmental impact, and compliance with stricter emission norms. At the heart of every
EV lies the powertrain, consisting of batteries, motors, controllers, and power electronics. Maintaining the
health of these systems is crucial for safety and performance.
Traditional diagnostic systems are reactive in nature, identifying issues only after they occur. Cloud-based
predictive approaches improve fault anticipation but face challenges such as reliance on internet connectivity,
increased latency, and data privacy concerns. Edge Artificial Intelligence (Edge AI) provides an alternative
by processing vehicle data locally using embedded devices, enabling real-time fault detection while
minimizing external dependencies.
This survey reviews recent advances in predictive health monitoring for EV powertrains using Edge AI and
CAN bus data. The contributions of this paper are as follows:
1)
A detailed classification of PHM research into four categories.
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING, MANAGEMENT &
APPLIED SCIENCE (IJLTEMAS)
ISSN No. 2278-2540 | DOI: 10.51584/IJLTEMAS | Volume XV Issue I January 2026
Page 2
www.rsisinternational.org
2)
A comparative review of literature from 2020–2025.
3)
Inclusion of hardware-oriented implementations from recent studies.
4)
Identification of open challenges and research opportunities for the future.
BACKGROUND
A.
Controller Area Network (CAN Bus)
The CAN bus is a standard in-vehicle communication protocol that enables data exchange between
electronic control units (ECUs), battery management systems, motor controllers, and sensors. It serves as
the primary data source for PHM, offering real-time information on voltage, current, temperature, and
system faults.
B.
Edge Artificial Intelligence (Edge AI)
Edge AI refers to running machine learning models on embedded devices instead of cloud servers. This
reduces latency, lowers dependence on internet connectivity, and safeguards privacy. Platforms such as
Raspberry Pi and ESP32, when combined with optimized libraries like TensorFlow Lite, can perform
anomaly detection on CAN data in real time, making them suitable for predictive monitoring applications.
LITERATURE SURVEY
I.
Patil et al. (2025) introduced an AI-based predictive maintenance framework that combines Long Short-
Term Memory (LSTM) networks with Federated Learning for secure fault detection in EVs [1]. While the
approach enhances privacy and supports predictive modelling, it imposes significant computational
requirements, limiting deployment on embedded devices.
II.
In another study, Wang et al. (2025) utilized Support Vector Machines (SVM) and K-Means clustering on
On-Board Diagnostics (OBD-II) data to monitor emissions and driving behaviour [2]. The method enables
improved vehicle
diagnostics but requires standardized OBD-II datasets for consistent integration with CAN bus analytics.
III.
Ahmed and Patel (2025) explored the use of Quantum Artificial Intelligence in combination with Federated
Learning for EV fault diagnosis [3]. Their method achieved high accuracy in detecting system faults; however,
it depends on experimental quantum hardware, which is not yet widely available for practical deployment.
IV.
Roy et al. (2025) proposed a privacy-preserving predictive maintenance system using Homomorphic
Encryption integrated with Edge AI [4]. This technique ensures secure CAN diagnostics by safeguarding
sensitive data, but the computational overhead places a heavy load on edge devices.
V.
Zhang et al. (2024) conducted a survey of anomaly detection methods in in-vehicle networks, focusing on
deep learning techniques and CAN bus analysis [5]. While their review provides a strong foundation for
machine learning-based diagnostics, it lacks real- world validation.
VI.
Lee et al. (2024) investigated battery health prediction using Deep Belief Networks (DBN) and Recurrent
Neural Networks (RNN) [6]. Their model achieved accurate State of Health (SoH) estimation but remains
unsuitable for resource-constrained edge devices, highlighting the need for lightweight alternatives.
VII.
Singh et al. (2024) adopted Reinforcement Learning to optimize energy management in hybrid vehicles
[7]. Although the approach is not explicitly health-focused, it shows potential for enhancing motor control
fault tolerance.
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING, MANAGEMENT &
APPLIED SCIENCE (IJLTEMAS)
ISSN No. 2278-2540 | DOI: 10.51584/IJLTEMAS | Volume XV Issue I January 2026
Page 3
www.rsisinternational.org
VIII.
Fernandez et al. (2024) presented a digital twin framework to simulate EV powertrain operations and
evaluate faults virtually [8]. The framework enables safe prototyping and model validation but requires a
complex and resource-intensive setup.
IX.
Gupta and Sharma (2023) employed statistical models and machine learning techniques to estimate the
Remaining Useful Life (RUL) of automotive components [9]. While their models align with industry
standards, they face challenges in real-time applications.
X.
Hossain et al. (2023) applied Convolutional Neural Networks (CNN), SVM, and time-series analysis for
lifecycle-based predictive maintenance in industrial equipment [10]. Although not EV-specific, the approach
demonstrates transferable concepts applicable to EV systems.
XI.
Kumar and Singh (2023) reviewed cybersecurity measures for electric vehicles, focusing on CAN bus
security protocols and Intrusion Detection Systems (IDS) [11]. While their study strengthens the
understanding of EV cybersecurity, it does not address diagnostic capabilities directly
XII.
Zhang et al. (2023) provided a broad review of AI integration in EV technologies [12]. Their work
highlights the role of AI in EV development but lacks depth in machine learning applications for predictive
monitoring.
XIII.
Das et al. (2023) applied DBN and LSTM models for accurate RUL estimation of EV batteries [13].
Although effective, the implementation faces limitations due to edge device constraints.
XIV.
Wiki Contributors (2023) proposed a conceptual framework for integrated vehicle health management
[14]. The framework outlines a system-wide monitoring structure but does not include practical hardware
considerations.
XV.
Another contribution by Wiki Contributors (2023) outlined strategies for Intelligent Maintenance Systems
(IMS) [15]. While useful for designing high- level predictive systems, the study is not AI-specific.
XVI.
Further, Wiki Contributors (2023) discussed Industrial Internet of Things (IIoT) architectures for
predictive maintenance [16]. Though not designed specifically for EVs, these architectures can facilitate sensor
integration in vehicular networks.
XVII.
Zhao et al. (2020) described an edge AI framework for vehicular applications [17]. Their work
demonstrates scalable deployment of AI at the edge, though it is not directly tailored to EVs.
XVIII.
Mishra and Rao (2020) explored distributed vehicular computing through fog and AI technologies [18].
Their approach supports distributed EV monitoring but suffers from scalability challenges.
XIX.
Chen et al. (2020) analyzed vulnerabilities in CAN bus systems and proposed IDS-based countermeasures
[19]. While the study is vital for vehicle security, it does not integrate predictive maintenance aspects.
XX.
Finally, Yadav et al. (2020) investigated machine learning-based IDS techniques for real-time fault
detection in CAN networks [20]. Although effective in detecting attacks, the work does not directly address
health diagnostics.
Summary:
The existing studies on Predictive Health Monitoring (PHM) in Electric Vehicles (EVs) demonstrate the use
of a variety of modern techniques designed to enhance vehicle reliability, safety, and performance. Researchers
have implemented multiple approaches such as deep learning, anomaly detection, reinforcement learning,
and digital twin systems to improve fault prediction and maintenance efficiency.
Deep learning-based models process large amounts of vehicle data—particularly CAN bus signals—to
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING, MANAGEMENT &
APPLIED SCIENCE (IJLTEMAS)
ISSN No. 2278-2540 | DOI: 10.51584/IJLTEMAS | Volume XV Issue I January 2026
Page 4
www.rsisinternational.org
estimate parameters like the State of Health (SoH) or Remaining Useful Life (RUL) of critical components
including batteries, motors, and inverters. These models can automatically detect patterns in data and offer
accurate health predictions under varying driving conditions.
Anomaly detection methods are applied to identify abnormal behaviours or irregular sensor readings in
real-time. Early identification of such deviations enables preventive maintenance, reduces breakdown risks,
and ensures vehicle safety.
Reinforcement learning allows systems to continuously learn from operational feedback and make data-
driven decisions to improve energy efficiency, fault recovery, and system stability. This adaptive nature helps
maintain optimal performance over time.
Digital twin technology involves building a virtual model of vehicle components that mirrors real-time
operations. This virtual replica supports continuous tracking, simulation, and prediction of component wear
and degradation, leading to better diagnostic and maintenance planning.
However, despite these technological advancements, several limitations still remain. One major issue is the
lack of standardized and open-access datasets, which restricts fair comparison and validation of different
PHM techniques. Additionally, most existing systems are designed for cloud platforms rather than edge
computing, which limits their ability to operate directly on low-power embedded devices inside the vehicle.
Another key challenge is the insufficient real-world testing of proposed methods. Many models are
validated only in controlled or simulated environments, which may not fully capture the complexities of real
driving conditions.
In conclusion, although notable progress has been made in predictive health monitoring for electric vehicles,
the next step is to develop scalable, reliable, and secure PHM frameworks that are practical for deployment
in real-world automotive systems.
Author(s), Year
Method / Technique
Key Contribution
Limitation / Future
Scope
I. Patil et al., 2025
LSTM, Federated
Learning
Secure EV fault
prediction
High computational
demand; needs
optimization
II. Wang et al., 2025
SVM, K-Means
OBD-II based emission &
behaviour diagnostics
Requires standardized
OBD- II data
III. Ahmed & Patel, 2025
Quantum AI + Federated
Learning
High accuracy in EV fault
detection
Relies on experimental
quantum hardware
IV. Roy et al., 2025
Homomorphic Encryption
+ Edge AI
Privacy-preserving
CAN
diagnostics
High processing load on
devices
V. Zhang et al., 2024
Deep Learning + CAN
Bus
Survey on anomaly
detection
Lacks real-world testing
VI. Lee et al., 2024
DBN, RNN
Accurate battery State of
Health (SoH) prediction
Not optimized for
embedded devices
VII. Singh et al., 2024
Reinforcement Learning
Hybrid vehicle energy
optimization
Not directly health-
focused
VIII. Fernandez et al.,
2024
Digital Twin Framework
Virtual testing of EV
powertrains
Complex to implement
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING, MANAGEMENT &
APPLIED SCIENCE (IJLTEMAS)
ISSN No. 2278-2540 | DOI: 10.51584/IJLTEMAS | Volume XV Issue I January 2026
Page 5
www.rsisinternational.org
IX. Gupta & Sharma,
2023
Statistical + ML Models
RUL estimation from
sensor data
Limited real-time
applicability
X. Hossain et al., 2023
CNN, SVM, Time Series
Industrial predictive
maintenance methods
Not specific to EV
systems
XI. Kumar & Singh, 2023
Security Protocols, IDS
Review of CAN bus
cybersecurity
Does not address health
monitoring
XII. Zhang et al., 2023
AI Technology Review
Overview of AI in EVs
Lacks ML-specific detail
XIII. Das et al., 2023
DBN + LSTM
Accurate EV battery RUL
prediction
Limited by edge hardware
constraints
XIV. Wiki Contributors,
2023
Conceptual Framework
Integrated vehicle health
monitoring
No hardware
implementation
XV. Wiki Contributors,
2023
Predictive Strategies
Intelligent maintenance
system (IMS)
General strategies, not AI-
specific
XVI. Wiki Contributors,
2023
IIoT Architecture
Predictive maintenance
framework
Not EV-specific
XVII. Zhao et al., 2022
Edge AI Framework
Vehicular edge computing
for AI
Not tailored to EVs
XVIII. Mishra & Rao,
2022
Distributed AI + Fog
Computing
Distributed EV
monitoring
Scalability challenges
XIX. Chen et al., 2022
IDS-based Security
Analysis of CAN
vulnerabilities
No predictive health
monitoring integration
XX. Yadav et al., 2022
ML-based IDS
Real-time
CAN
anomaly
detection
Not combined with health
diagnostics
TAXONOMY OF APPROACHES
From the surveyed literature, PHM approaches can be broadly classified into:
1.
Model-Based Approaches: Simulation environments and digital twins for predicting potential
failures.
2.
Data-Driven Approaches: ML and DL techniques such as LSTM, CNN, and DBN applied to CAN
bus data.
3.
Security-Focused Approaches: IDS and encryption-based frameworks for secure data transfer.
4.
Edge-Based Implementations: Deployment of lightweight ML models on embedded devices for real-
time fault detection.
COMPARATIVE ANALYSIS
•
Focus Areas: Battery health monitoring and anomaly detection dominate research efforts.
•
Trends: Research peaked in 2023, with sustained developments in 2024 and 2025.
•
Strengths: Accuracy improvements through ML/DL, reinforcement learning in hybrid systems, and
integration of digital twins.
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING, MANAGEMENT &
APPLIED SCIENCE (IJLTEMAS)
ISSN No. 2278-2540 | DOI: 10.51584/IJLTEMAS | Volume XV Issue I January 2026
Page 6
www.rsisinternational.org
•
Weaknesses: Limited real-world deployment, lack of standardized datasets, and resource constraints
on edge hardware.
Distribution of Research Focus in PHM of EV Powertrains (2020-2025)
Research Trends in Predictive Health Monitoring of Powertrains (2020-2025)
HARDWARE IMPLEMENTATIONS
Several studies have demonstrated the feasibility of deploying PHM systems on embedded hardware:
•
Hardware Platforms: Raspberry Pi (5), MCP2515 CAN interface modules.
•
Data Acquisition: CAN signals accessed via the OBD-II port for real-time monitoring.
8
7
6
5
4
3
2
1
0
2020
2023
20224
2025
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING, MANAGEMENT &
APPLIED SCIENCE (IJLTEMAS)
ISSN No. 2278-2540 | DOI: 10.51584/IJLTEMAS | Volume XV Issue I January 2026
Page 7
www.rsisinternational.org
•
Software Tools: Python with TensorFlow Lite, Scikit-learn, and python-can.
•
Algorithms: One-Class SVM, Isolation Forest, and lightweight LSTM variants.
•
Output Interfaces: Dashboards, LEDs, or mobile apps for driver alerts.
FUTURE DIRECTIONS
1.
Standardized Datasets: Development of open CAN bus datasets for benchmarking predictive
models.
2.
Optimized Edge Models: Model compression and pruning techniques for resource-constrained
hardware.
3.
Digital Twins: Virtual replicas for safe and accurate fault simulation.
4.
Explainable AI: Enhancing interpretability and trust in fault predictions.
5.
Cybersecurity Integration: Strengthening PHM systems against in-vehicle cyber threats.
6.
Hybrid Edge–Cloud Frameworks: Combining edge responsiveness with cloud scalability.
CONCLUSION
This paper surveyed predictive health monitoring methods for EV powertrains using Edge AI and CAN bus
data. The review highlighted progress in battery diagnostics, anomaly detection, security, and real-time
implementations. Edge computing offers significant advantages over cloud-based systems, including reduced
latency, privacy, and cost-effectiveness. However, gaps remain in dataset availability, scalability, and model
optimization. Addressing these challenges will be critical to establishing PHM as a standard feature in future
EV ecosystems.
REFERENCES
1. S. Patil, A. Mehta, and R. Kulkarni, “AI-driven predictive maintenance in electric vehicles using LSTM
and federated learning,” in Proc. Int. Conf. Smart Mobility Syst., 2025, pp. 1–6.
2. H. Wang, Y. Li, and P. Zhang, “OBD-II based machine learning applications for vehicle monitoring,”
IEEE Sensors Journal, vol. 25, no. 4, pp. 1123–1132, 2025.
3. M. Ahmed and D. Patel, “Quantum-enhanced artificial intelligence for fault diagnosis in electric vehicles,”
IEEE Transactions on Vehicular Technology, vol. 74, no. 2, pp. 456–467, 2025.
4. A. Roy, K. Sharma, and P. Nair, “Privacy-preserving predictive maintenance using homomorphic
encryption and edge AI,” in Proc. IEEE Int. Conf. Edge Computing, 2025, pp. 201–208.
5. Y. Zhang, T. Chen, and S. Huang, “A survey of anomaly detection in in-vehicle networks,” IEEE Access,
vol. 12, pp. 54123–54139, 2024.
6. J. Lee, R. Park, and H. Kim, “AI for battery health monitoring using deep belief networks and recurrent
neural networks,” Energies, vol. 17, no. 6, pp. 3112–3128, 2024.
7. R. Singh, P. Verma, and S. Gupta, “Reinforcement learning for energy management in hybrid vehicles,”
Chinese Journal of Mechanical Engineering, vol. 37, no. 2, pp. 215–225, 2024.
8. M. Fernandez, L. Torres, and G. Romero, “Digital twin framework for EV powertrains,” World Electric
Vehicle Journal, vol. 15, no. 5, pp. 234–245, 2024.
9. P. Gupta and A. Sharma, “Predictive maintenance in the automotive sector using machine learning and
statistical models,” Journal of Industrial Engineering Research, vol. 62, no. 3, pp. 154–166, 2023.
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING, MANAGEMENT &
APPLIED SCIENCE (IJLTEMAS)
ISSN No. 2278-2540 | DOI: 10.51584/IJLTEMAS | Volume XV Issue I January 2026
Page 8
www.rsisinternational.org
10. T. Hossain, N. Alam, and M. Rahman, “AI applications in industrial equipment maintenance using CNN
and SVM,” IEEE Transactions on Industrial Informatics, vol. 19, no. 7, pp. 8452–8464, 2023.
11. V. Kumar and S. Singh, “Cybersecurity in electric vehicles: A review of CAN bus protocols and intrusion
detection systems,” IEEE Transactions on Intelligent Transportation Systems, vol. 24, no. 1, pp. 122–
135, 2023.
12. H. Zhang, J. Liu, and F. Zhao, “AI integration in electric vehicle technology: A comprehensive review,”
IEEE Access, vol. 11, pp. 78232–78249, 2023.
13. S. Das, A. Mondal, and K. Banerjee, “Remaining useful life estimation using DBN and LSTM models,”
in Proc. Int. Conf. Prognostics and Health Management (PHM), 2023, pp. 211–219.
14. “Integrated vehicle health management: A conceptual framework,” Wikipedia, 2023. [Online].
Available: https://en.wikipedia.org/wiki/Integrated_vehicle_hea lth_management. [Accessed: Oct. 18,
2025].
15. “Intelligent maintenance system (IMS),” Wikipedia, 2023. [Online]. Available:
https://en.wikipedia.org/wiki/Intelligent_maintenanc e_system. [Accessed: Oct. 18, 2025].
16. “Industrial Internet of Things (IIoT) for predictive maintenance,” Wikipedia, 2023. [Online]. Available:
https://en.wikipedia.org/wiki/Industrial_Internet_of_ Things. [Accessed: Oct. 18, 2025].
17. L. Zhao, Q. Wang, and J. Sun, “Vehicular edge computing and networking framework for AI
deployment,” IEEE Internet of Things Journal, vol. 9, no. 8, pp. 7120–7134, 2022.
18. P. Mishra and K. Rao, “Distributed vehicular computing for predictive maintenance using fog AI,” in
Proc. IEEE Int. Conf. Distributed Systems, 2022, pp. 134–140.
19. Y. Chen, L. Wu, and D. Xu, “CAN bus cyberattacks and countermeasures using intrusion detection
systems,” in Proc. IEEE Vehicular Networking Conf., 2022, pp. 88–95.
20. R. Yadav, M. Sharma, and K. Singh, “Machine learning-based intrusion detection techniques for real-
time CAN bus security,” Journal of Embedded Systems and Vehicular Security, vol. 14, no. 2, pp. 99–
110, 2022.
