AI/ML enabled smart battery management systems in electric vehicles

1 Introduction

The global transition toward sustainable energy systems and electrified transportation has significantly increased the demand for efficient and reliable energy storage technologies. Lithium-ion batteries (LIBs) have become the dominant energy storage solution due to their high energy density, long cycle life, and relatively low self-discharge rate (Wei Q. et al., 2023; Schulz-Mönninghoff et al., 2021). These batteries are widely used in electric vehicles (EVs), renewable energy storage systems (ESS), and portable electronics devices. However, the performance, safety, and longevity of LIBs depends heavily on effective monitoring and control mechanisms provided by a battery management system (BMS) (Kim et al., 2020). A BMS is responsible for monitoring critical battery parameters such as voltage, current, temperature, state of charge (SoC), and state of health (SoH), while also performing essential functions including cell balancing, thermal management, and fault protection (Abdolrasol et al., 2024). Accurate battery monitoring is particularly important for large battery packs used in EVs and grid-connected ESS, where hundreds of cells operate simultaneously and require coordinated control strategies (Zhou L. et al., 2023).

Tradional BMS rely on model-based estimation techniques such as coulomb counting, equivalent circuit models (ECM), and kalman filtering algorithms to estimate battery state (Li et al., 2024; Li et al., 2023; Krstić et al., 2023). While these approaches have been widely used in industrial applications, they often struggle to accurately represent the non-linear electrochemical behaviour of LIBs under varying operating conditions (Wang J. et al., 2025; Schwab et al., 2024). Battery characteristics are influenced by temperature variation, aging mechanisms, and dynamic load profiles., which makes accurate estimation of internal battery states challenging. As a result, conventional operating environments (Chen et al., 2023). Moreover, the increasing complexity of battery packs in modern EVs requires advanced monitoring approaches capable of processing large amounts of real-time data and identifying complex degradation patterns (Camboim et al., 2024).

In recent years, artificial intelligence (AI) and ML techniques have emerged as promising solutions for enhancing battery management capabilities (Bhushan et al., 2022). ML algorithms can analyze large datasets obtained from battery sensors and experimental measurements to identify non-linear relationships between battery parameters. These data-driven approaches allow for improved prediction accuracy compared with traditional model-based methods. artificial neural networks (ANNs) have been widely used for estimating the SoC of LIBs due to their ability to capture complex non-linear battery dynamics (Li et al., 2019). Experimental studies have shown that ANN-based models can achieve significantly improved accuracy in SoC estimation compared with conventional estimation techniques, particularly under dynamic operating conditions (Hannan et al., 2020). Furthermore, ML algorithms such as support vector machine (SVM) and ensemble learning methods have been applied to detect battery faults and abnormal operating condition at early stages, thereby improving system safety and reliability (Obuli Pranav et al., 2024).

Deep learning (DL) approaches have further enhanced the capabilities of intelligent BMS (Yang Z. et al., 2021; Ben Youssef et al., 2021). Deep neural network (DNN), convolutional neural network (CNN), and LSTM models have been widely explored for predicting battery degradation and RUL (Li et al., 2017; Yang et al., 2023; Liu H. et al., 2023; Ghosh et al., 2025). These models are capable of extracting complex patterns from high dimensional sensor data and capturing temporal dependencies in battery degradation processes. LSTM networks are particularly effective in analysing time-series battery data and predicting ling-term performance degradation, making suitable for estimating RUL and SoH parameters (Ren and Du, 2023). Recent studies published in leading energy journals have demonstrated that DL based models can achieve high prediction accuracy and improved robustness when compared with traditional analytical models (Kurucan et al., 2024).

Another emerging direction in intelligent BMS involves the integration of internet of things (IoT) technologies with ML algorithms (Pushpavanam et al., 2024). IoT-enabled BMS framework employ distributed sensors and communication networks to collect real-time battery data from individual cell and module (Waseem et al., 2023). These data are transmitted to centralized processing platform where ML models analyze system behaviour and predicts potential failures. IoT-integrated BMS systems have been shown to significantly improve operational reliability and reduce maintenance cost in large-scale ESS applications (Kha et al., 2023). Reinforcement learning (RL) techniques is also explored to optimize battery charging and energy management strategies (Amir et al., 2024). This approach enables adaptive charging strategies that reduce battery degradation and improve energy efficiency. Such intelligent charging control methods are particularly beneficial for EVs and renewable ESS where battery performance is influenced by dynamic environmental condition and load demands (Yun et al., 2025).

In addition to software-based intelligence, latest research is focused on advanced hardware implementation to support intelligent BMS. Field programmable gate array (FPGA) framework and embedded processors is used to implement real-time monitoring and control algorithms for BMS (Waseem et al., 2024). These hardware platforms provide high computational capability and low-latency processing required for implementing advanced AI approaches in practical BMS. Furthermore, predictive analytics frameworks is proposed to enable adaptive control strategies for grid-connected battery storage systems., allowing battery packs to dynamically respond to energy demand fluctuations and renewable energy generation applications (Haraz et al., 2024). Recent advancement in BMS have increasingly integrated AI, ML and advance computational architectures to improve battery monitoring, performance optimization and predictive maintenance. Different AI and ML based algorithms employed in EVs for estimating battery performance are provided in Table 1 and discussed below.

TABLE 1. Different AI/ML based techniques, their functions and key findings.

In 2020, a distributed wireless cell-level BMS architecture was developed to enhance the scalability and efficiency of monitoring systems in EVs. The study focused on a balancing algorithm based on SoC and SoH parameters to achieve efficient cell balancing. The methodology involved wireless communication between battery cells and a centralized monitoring unit. The findings demonstrated improved fault tolerance and scalable architecture suitable for large battery packs. However, the system did not incorporate ML algorithms, limiting its capability for predictive analytics and intelligent decision-making. Also in 2020, M-based ANN were applied for accurate SoC estimation in LIBs. The ANN model was trained using battery voltage, current and temperature datasets to capture non-linear battery behavior. Results showed that ANN-based estimation significantly improved accuracy compared with traditional columb-counting methods. The main advantage of this approach lies in its ability to model complex battery dynamics. Meanwhile, the model requires large training datasets and computational resources, which may limit real-time embedded implementation.

In 2021, SVM technique was explored for battery fault diagnosis and anomaly detection. The methodology employed by SVM classifiers trained with battery operational data to identify abnormal voltage and temperature patterns. Experimental results showed that SVM can detect early sign of battery degradation and potential failure issues. The key advantage of this approach is its high classification, accuracy, and robustness with limited datasets. However, SVM models may struggle with highly non-linear battery behaviors and require careful feature selection. Another key approach introduced hybrid methods combining ML with traditional estimation algorithms such as the extended kalman filter (EKF). This hybrid technique integrates ML prediction capabilities with the filtering and noise-reduction feature of KF. The findings indicate that improved SoC and SoH estimation accuracy under dynamic operating conditions in EVs. While this hybrid increases estimation reliability, its implementation complexity and parameter tuning remain challenges.

In 2022, several studies investigated deep learning methods for battery health prediction. LSTM and NN were widely adopted for RUL prediction due to their ability to capture temporal dependencies in battery degradation data. The LSTM model analyzed time-series battery operational data to predict future battery health conditions. The results demonstrated superior prediction accuracy compared with conventional statistical models. However, LSTM models typically require large training datasets and may involve longer training times. Furthermore, comprehensive review studies in 2022 summarized state-of-the-art BMS technologies, including intelligent algorithms for state estimation, charge balancing, thermal management, and fault detection. These studies highlighted the transition from conventional rule-based systems toward AI-driven BMS architectures. Despite providing valuable insights into research trends, review studies generally lack experimental validation of proposed techniques.

In 2023, intelligent BMS architectures integrating ML with IoT technologies were proposed for real-time battery monitoring and predictive maintenance. The system utilized distributed IoT sensors to collect battery operational parameters such as voltage, temperature, and current, which were processed by machine learning algorithms for health monitoring and adaptive control. Experimental results demonstrated improved predictive maintenance capabilities and enhanced system reliability. However, IoT-based BMS architectures may face challenges related to communication latency, cybersecurity, and data management. Deep learning approaches, including DNN, were also explored in 2023 for battery degradation modeling. These models analyze large datasets generated from battery cycling experiments to identify degradation patterns and predict SoH. The primary advantage of deep learning lies in its ability to automatically extract complex features from high-dimensional data. However, deep learning models typically require significant computational resources and large labeled datasets.

In 2024, RL algorithms were introduced to optimize battery charging strategies. RL-based approaches learn optimal charging policies through interaction with the battery system and environmental conditions. The results indicated improved energy efficiency and reduced battery aging compared with traditional charging methods. Nevertheless, RL algorithms require extensive training and careful reward function design to achieve stable performance.

In 2025, several studies focused on advanced system architectures for intelligent BMS. One proposed computational system architecture for EV battery packs that integrates advanced monitoring and control features. The methodology involved hardware-software integration for real-time monitoring of battery parameters such as voltage, current, and temperature. The system demonstrated improved operational efficiency and safety. However, the study mainly focused on system design rather than advanced AI algorithms. Another IoT-enabled machine learning BMS prototype was proposed in 2025 for real-time monitoring and predictive degradation detection. The system integrated ML algorithms with IoT sensors to provide adaptive charging control and early fault detection. Experimental prototypes demonstrated improved reliability and predictive maintenance capabilities, although large-scale deployment may require advanced data management infrastructure.

More recently, in 2026, hybrid ML frameworks combining support vector regression (SVR), gradient boosting models such as XGBoost, and EKF algorithms have been proposed for accurate battery SoH estimation. These hybrid models leverage both data-driven learning and physical system modeling to improve prediction accuracy. The results showed enhanced performance in estimating battery impedance and degradation parameters in real-time BMS applications. However, hybrid frameworks may increase computational complexity and require careful model integration. Additionally, ensemble machine learning models have been investigated for simultaneous prediction of SoC, SoH, and RUL. These approaches combine multiple algorithms to improve prediction robustness and reduce model uncertainty. While ensemble models achieve high predictive accuracy, their computational overhead may limit embedded system implementation.

Overall, the literature demonstrates a clear transition from conventional BMS architectures toward intelligent AI-driven BMS. Modern research increasingly emphasizes hybrid AI models, predictive analytics, IoT-based monitoring, and deep learning techniques to enhance battery performance, safety, and lifecycle management. However, challenges such as computational complexity, data requirements, real-time implementation, and system integration remain key research directions for future intelligent BMS development.

Despite the growing body of literature on AI-assisted battery management systems, most existing review studies focus on specific aspects such as SoC/SoH estimation, battery thermal management, fault diagnosis, or individual machine learning algorithms. In contrast, the present review provides a comprehensive system-level assessment of AI/ML-enabled smart BMS architectures by integrating state estimation, cell balancing, thermal management, fault detection, predictive maintenance, intelligent charging, FPGA-based edge implementation, and IoT-cloud connectivity within a unified framework. Furthermore, this study presents a comparative analysis of contemporary AI/ML techniques, highlights practical deployment challenges, and identifies future research directions involving physics-informed machine learning, digital twins, edge intelligence, and cloud-connected battery ecosystems. Therefore, this review offers a broader and more application-oriented perspective for the development of next-generation intelligent battery management systems for electric vehicles and energy storage applications.

1.1 Research gap and organization

Based on the identified research gaps, this study proposes an intelligent BMS framework that integrates advanced monitoring architecture with predictive analytics to improve battery performance and reliability. Unlike conventional BMS approaches that rely primarily on rule-based control or simulation models, the proposed system combines real-time battery parameter monitoring with adaptive algorithms for enhanced state estimation and fault detection. The system is designed to support scalable battery pack architectures while maintaining efficient computational performance suitable for embedded implementation. Furthermore, the proposed framework aims to improve battery lifecycle management and enable better integration with smart energy systems, addressing limitations observed in existing AI-based BMS approaches.

Organization of the current study is structural as follows: Section 2 provides the framework for BMS and its different functional role including SoC, SoH, cell balancing and practical control. Section 3 presents thermal management and safety associated with lithium-ion battery technology. Section 4 represents the role of AI and ML in BMS to make it more advanced. Comparative study of different AI/ML enabled BMS system is presented in Section 4.7. concluding remark and future scope of work are presented in Section 5.

1.2 Contribution of the present review

Unlike previous reviews that primarily focus on individual BMS functions or isolated AI algorithms, this study critically evaluates the integration of AI/ML techniques with battery state estimation, thermal management, fault diagnosis, intelligent charging, edge computing, and cloud-based battery ecosystems. The review further identifies key research gaps and future opportunities for developing scalable and intelligent battery management solutions.

• This review presents a critical analysis of 120 high-quality research articles and review papers published in leading journals and conferences indexed in IEEE Xplore, ScienceDirect (Elsevier), MDPI, Springer, Wiley, Taylor & Francis, Frontiers, and other reputable scientific databases.
• Existing literature/reviews mainly focus on SoC/SoH estimation methods. Limited reviews integrate AI/ML, thermal management, cell balancing, and predictive maintenance in a single framework.
• Emerging technologies such as FPGA-based BMS, IoT-enabled monitoring, and cloud analytics have not been comprehensively assessed.
• Practical implementation challenges including explainability, cybersecurity, and real-time deployment remain insufficiently discussed.
• This review provides a unified comparative framework and future roadmap for intelligent BMS development.

2 Key functions of BMS

Modern BMS plays a critical function in ensuring the safety, reliability, and optimal performance of LIB battery packs used in EVs, grid-connected ESS, and renewable energy applications (Ranjith Kumar et al., 2023). Among the fundamental role of a BMS, accurate state estimation and effective cell balancing are essential for maintaining operational stability and extending battery lifetime. Recent advancements in intelligent BMS architecture integrate advanced estimation algorithms and adaptive balancing strategies to address battery aging, cell heterogeneity, and dynamic operating conditions (Habib et al., 2023). BVMS and various associates with it is depicted in Figure 1.

2.1 State estimation role

State estimation is a core function of the BMS because it provides essential information for control, safety monitoring, and lifecycle optimization (Xing et al., 2011). The two most important internal battery states estimated by modern BMS systems are the SoC and SoH. SoC represents the remaining charge in the battery relative to its nominal capacity, while SoH indicates the degradation level and remaining useful capacity of the battery over its lifetime (Balasingam et al., 2020).

Accurate estimation of SoC and SoH is challenging because these parameters cannot be measured directly and must instead be inferred using computational algorithms and real-time sensor data (Kurkin et al., 2025). Typically, voltage, current, and temperature measurements are collected through embedded sensing units and processed by estimation algorithms to determine internal battery states. Conventional techniques such as coulomb counting and open-circuit voltage methods are commonly used but often suffer from accumulated errors and environmental sensitivity (Ng et al., 2009). Therefore, modern BMS architectures employ model-based estimation approaches, including equivalent circuit models integrated with filtering techniques such as the EKF and unscented kalman filter (UKF) to improve estimation accuracy under dynamic operating conditions (Peng et al., 2022).

In addition to model-based methods, recent research has increasingly explored data-driven and ML-based state estimation algorithms, which can capture nonlinear battery behavior more effectively (Rojek et al., 2023). These approaches combine real-time measurement feedback with adaptive models to enhance prediction accuracy and battery lifecycle performance. As a result, SoC and SoH estimation outputs serve as primary inputs for critical BMS control functions, including power limitation, charging control, and fault detection (Bandla et al., 2024).

2.2 Cell balancing goal

Battery packs consist of multiple cells connected in series and parallel configurations. Due to manufacturing tolerances, temperature gradients, and aging effects, individual cells often exhibit variations in capacity, internal resistance, and degradation rate (Nyamathulla and Dhanamjayulu, 2024). These variations lead to cell imbalance, where certain cells reach their voltage limits earlier than others during charging or discharging cycles.

Cell balancing mechanisms are implemented in the BMS to reduce the spread in SoC and capacity among cells, thereby preventing overcharging or overdischarging of weaker cells (Cao et al., 2024). Without effective balancing strategies, the overall capacity of the battery pack becomes limited by the weakest cell, resulting in reduced usable energy and accelerated degradation. Therefore, the main objective of balancing algorithms is to maintain uniform distribution across cells while maximizing the operational lifetime of the battery pack (Cao and Emadi, 2009).

2.3 SoH-aware balancing strategies

Recent developments in intelligent BMS systems incorporate SoH-aware balancing strategies, where both SoC and SoH information are used to determine balancing actions (Lelie et al., 2018). This approach is particularly beneficial for battery packs operating under long-term cycling conditions where cells age unevenly. By integrating SoH information into balancing algorithms, the BMS can distribute charge more effectively and protect weaker cells from excessive stress (Sudhapriya and Jaisiva, 2025).

Advanced wireless cell-level BMS architectures have demonstrated the feasibility of implementing distributed balancing strategies. In such systems, individual cells or modules communicate wirelessly with the central control unit, allowing scalable monitoring and decentralized balancing control. Experimental prototype studies have shown that these architectures can balance large battery packs containing cells with different aging levels and capacity deviations while maintaining system reliability and flexibility (Gozuoglu, 2025).

2.4 Practical control and implementation

In practical implementations, balancing operations can be performed at either the cell level or module level, depending on the battery pack architecture and control strategy. Two primary balancing approaches are widely used in modern BMS systems: passive balancing and active balancing (Kertész and Szabó, 2024). Passive balancing dissipates excess energy from higher-charged cells using resistive elements, while active balancing redistributes energy among cells through power electronic converters, improving overall energy efficiency (Waag et al., 2013).

The BMS control unit must schedule balancing operations carefully to minimize energy loss and avoid unnecessary thermal stress. Typically, balancing actions are triggered when the SoC difference between cells exceeds a predefined threshold or when voltage deviations approach safety limits. During operation, the BMS continuously monitors cell voltage, current, and temperature to ensure that balancing processes occur within safe operating ranges (Baumann et al., 2018).

By integrating accurate state estimation algorithms with adaptive balancing control strategies, modern BMS architectures can maintain cell voltage uniformity, prevent premature degradation of weaker cells, and maximize the operational efficiency of battery packs (; Waseem et al., 2025). These integrated control mechanisms are particularly important in electric vehicles and renewable energy storage systems where reliability, safety, and long-term performance are critical design requirements (Yang S. et al., 2021).

3 Thermal management and safety controls

Thermal management and safety controls are critical functions of modern BMS, particularly for LIB packs used in EVs and grid-connected energy systems (Challoob et al., 2024). Due to the strong coupling between electro-chemical behavior and temperature, improper thermal regulation can lead to accelerated degradation, reduced efficiency, and severe safety hazards such as thermal runaway. Therefore, intelligent BMS architectures incorporate advanced thermal monitoring, active thermal control, and fault detection mechanisms to ensure safe and reliable battery operation (Zhang et al., 2025).

3.1 Thermal monitoring

Smart BMS designs incorporate per-cell or per-module temperature sensing to continuously monitor the thermal state of the battery pack. Temperature sensors, typically thermistors or integrated sensing devices, are strategically placed within the battery modules to capture spatial temperature variations (He et al., 2023). These measurements are critical inputs not only for thermal regulation but also for improving the accuracy of SoC and SoH estimation, as battery parameters such as internal resistance, capacity, and reaction kinetics are highly temperature-dependent (Hannan et al., 2017). Temperature gradients within a battery pack can lead to non-uniform aging and imbalance among cells. Therefore, real-time thermal monitoring enables the BMS to detect localized hotspots and abnormal temperature rises, which are early indicators of potential failure modes. Accurate thermal data thus plays a fundamental role in both performance optimization and safety assurance in modern BMS implementations (Peng et al., 2022). Thermal management system block architecture for battery pack is depicted in Figure 2.

The thermal management system highlights the critical role of temperature regulation in ensuring the safe, efficient, and reliable operation of battery systems. It illustrates how real-time temperature sensing at the cell or module level feeds into the BMS, enabling continuous monitoring of thermal conditions. Based on this data, the control unit dynamically activates cooling or heating mechanisms and adjusts operating limits such as charge/discharge rates to maintain the battery within a safe temperature range. This coordinated control helps prevent thermal runaway, reduces uneven temperature distribution across cells, and minimizes degradation caused by thermal stress. Additionally, the integration of thermal data with state estimation and control strategies enables adaptive system responses, improving battery lifespan, performance consistency, and overall system safety.

3.2 Active thermal control

In addition to monitoring, the BMS implements active thermal control strategies to regulate battery temperature within safe operating limits (Zhou W. et al., 2023). These strategies involve coordination with cooling and heating subsystems such as air cooling, liquid cooling, phase-change materials, or resistive heating elements. The BMS dynamically adjusts these systems based on real-time temperature measurements and operating conditions (Singh et al., 2023).

Furthermore, the BMS modifies charge and discharge limits in response to thermal conditions to prevent excessive heat generation. During high-temperature operation, the BMS may reduce charging current or limit discharge power to avoid thermal stress (Wiriyasart et al., 2020). Conversely, at low temperatures, preheating strategies may be activated to improve battery performance and prevent lithium plating. Such adaptive thermal control is essential for mitigating the risk of thermal runaway, a critical failure mechanism in lithium-ion batteries characterized by uncontrolled heat generation and potential fire or explosion (Li and Wang, 2025). By integrating thermal regulation with operational control, modern BMS designs enhance battery lifespan and ensure safe operation under varying environmental and load conditions (Hasani et al., 2025; Pesaran, 2001).

3.3 Fault detection and safety mechanisms

A key responsibility of the BMS is the rapid detection of abnormal operating conditions, including overvoltage, undervoltage, overcurrent, and abnormal temperature rise (Yu et al., 2023). Advanced BMS systems employ rule-based logic, model-based diagnostics, and increasingly machine learning-based anomaly detection algorithms to identify faults at an early stage (Kumar et al., 2018). Once a fault condition is detected, the BMS initiates appropriate protective actions, such as disconnecting the battery from the load (cutoff), isolating faulty cells or modules, or derating the system by limiting current flow. These actions are essential to prevent catastrophic failures and ensure the safety of both the battery system and its surrounding environment (Zhao et al., 2024). Thermal faults, in particular, require rapid response due to their potential to escalate into thermal runaway. Therefore, BMS safety mechanisms are designed with strict thresholds and fast response times to ensure effective protection under critical conditions (Peng et al., 2019).

3.4 Integration consequence

The integration of thermal monitoring with SoC and SoH estimation enables adaptive and intelligent control strategies in advanced BMS systems (Yang et al., 2019). Since battery degradation is strongly influenced by temperature, combining thermal data with state estimation allows the BMS to implement adaptive derating and optimized charging strategies. This integrated approach improves overall system reliability by reducing degradation risks, enhancing energy utilization, and extending battery lifespan (Liu J. et al., 2023). It also enables more accurate prediction of battery performance under real-world operating conditions, which is essential for both electric vehicle and grid-scale energy storage applications (Singh et al., 2023; Pesaran, 2001).

4 Role of AI and machine learning in BMS

The integration of AI and ML into BMS has significantly enhanced the capability of modern EV and energy storage systems (Wang X. et al., 2022). Unlike conventional BMS approaches that rely on deterministic and model-based techniques, AI-driven frameworks enable data-driven decision-making, adaptive control, and predictive analytics (Swarnkar et al., 2023; Njoku et al., 2025). These capabilities are essential for improving battery performance, extending lifecycle, and ensuring safety under dynamic operating conditions.

4.1 Estimation and prediction

ML models such as ANN, LSTM, and gaussian process regression are widely used to enhance SoC and SoH estimation accuracy (Lehtola, 2025). These models capture nonlinear battery characteristics and temporal dependencies in degradation data, enabling more precise estimation compared to traditional model-based approaches. Additionally, ML-based predictive analytics can forecast degradation trends and RUL, allowing the BMS to implement lifecycle-aware control strategies (Elnady and Ozana, 2025; Wang C. et al., 2025).

4.2 Predictive maintenance and fault detection

AI-driven BMS systems utilize data-driven techniques to identify early signs of battery degradation and potential failures (Mahek et al., 2025). ML models such as SVM, random forests, and DNN analyze operational data to detect anomalies in voltage, current, and temperature profiles. This enables predictive maintenance, reducing unexpected downtime and improving system reliability. Deep learning techniques, including CNN and autoencoders, are particularly effective in detecting hidden fault patterns and thermal anomalies (Zhang et al., 2020; Andre et al., 2013).

4.3 Adaptive charging and optimization

AI-based optimization algorithms, including RL, are increasingly used to design adaptive charging strategies. These algorithms dynamically adjust charging rates based on real-time inputs such as SoC, SoH, temperature, and grid conditions (Miraftabzadeh et al., 2024). The objective is to balance charging speed with battery longevity while preventing degradation mechanisms such as lithium plating and thermal stress. Such intelligent charging strategies are essential for next-generation EVs and smart grid-integrated energy storage systems (de la Iglesia et al., 2025).

4.4 Edge implementation and hardware acceleration

To enable real-time decision-making in EV environments, AI algorithms are increasingly deployed on embedded hardware platforms (Sai Ramesh et al., 2023). FPGA-based intelligent battery management units (i-BMU) and edge AI processors allow low-latency execution of ML models for state estimation, fault detection, and power distribution control. These hardware-accelerated implementations improve system responsiveness and enable real-time optimization in safety-critical applications, although they introduce additional design complexity and cost considerations (Lee et al., 2025).

4.5 IoT and cloud integration

Modern smart BMS architectures integrate IoT-enabled telemetry with cloud-based analytics platforms (Shi et al., 2023). Battery data collected from sensors is transmitted to cloud servers, where advanced ML models perform large-scale analysis, fleet-level optimization, and predictive diagnostics. Cloud integration also enables over-the-air (OTA) updates, allowing continuous improvement of ML models and adaptive control strategies (Yalavarthy et al., 2025). This approach supports coordinated energy services and enhances the scalability of battery management in large EV fleets and grid storage systems (Andre et al., 2013; Liu et al., 2021). A focused architecture of an AI and ML-based intelligent BMS up to the core intelligence layer, clearly illustrating the data-driven workflow from sensing to decision-making is depicted in Figure 3.

It begins with the battery pack, where key parameters such as voltage, current, temperature, pressure, and impedance are measured. These raw signals are then processed in the data acquisition and preprocessing stage, which includes sensor integration, IoT-based data transmission, signal conditioning, filtering, and feature extraction. This stage is crucial for ensuring high-quality input data, as the performance of AI/ML models heavily depends on accurate and noise-free measurements. The processed data is fed into the AI/ML core, which represents the central intelligence of the system. This layer performs multiple advanced functions: state estimation (SoC and SoH) using ANN and LSTM models, RUL prediction using time-series learning, fault detection through CNN and autoencoder-based anomaly detection, and degradation modeling using regression techniques such as gaussian process regression or XGBoost. Additionally, a decision-making module based on RL enables adaptive optimization of charging and discharging strategies.

4.6 Practical considerations and challenges

Despite the advantages of AI/ML integration, practical deployment in BMS systems presents several challenges (Satpathy et al., 2025). The literature emphasizes that hybrid approaches combining model-based control with ML-driven prediction represent the most promising direction for future BMS development (Khaleel et al., 2025). However, real-world implementation requires careful consideration of factors such as model validation, computational constraints, communication reliability, and compliance with safety certification standards (Fasano et al., 2025; Badran and Toha, 2024). Ensuring robustness and explainability of ML models remains a key requirement for their adoption in safety-critical battery applications (Saha et al., 2021; Martins and Rodrigues, 2025).

4.7 Comparative analysis of different AI/ML techniques in BMS

Growing role of AI/ML techniques in enhancing the functionality and intelligence of modern BMS is highlighted in Table 2. Techniques such as ANN, LSTM, and gaussian process models are widely used for accurate estimation of SoC and SoH, owing to their ability to capture nonlinear battery dynamics and temporal behavior, although their effectiveness depends heavily on the availability of large training datasets (Gozuoglu, 2025; Nighojkar et al., 2022). Similarly, deep learning and LSTM-based approaches are applied for forecasting battery aging and RUL, enabling lifecycle-aware control and predictive planning; however, these methods often involve high computational complexity (Firke and Kumar Gupta, 2024; Da et al., 2024).

TABLE 2. Role of AI/ML in battery management system.

For system reliability, ML algorithms such as SVM, random forest, and DNN are employed for early fault detection and anomaly identification, significantly reducing downtime, but they may suffer from generalization issues under unseen operating conditions (Li et al., 2021; Avcı and Koca, 2023). In addition, convolutional neural networks (CNN) and autoencoders are particularly effective in detecting thermal anomalies and voltage irregularities due to their high sensitivity to hidden patterns, though they require high-quality labeled datasets for robust performance (Girijaprasanna and Dhanamjayulu, 2022; Poorani et al., 2025).

Reinforcement learning (RL) and other ML-based optimization techniques further enable adaptive charging strategies by dynamically adjusting charging rates based on real-time battery states, thereby balancing fast charging with battery longevity, albeit with challenges related to training stability and complexity (Shibl et al., 2023; Hoffmann et al., 2022). From an implementation perspective, FPGA-based ML and embedded AI solutions facilitate real-time processing and low-latency control suitable for electric vehicle environments, but they introduce additional hardware cost and design complexity (Kim and So, 2023; Tian et al., 2016). Finally, cloud-based machine learning and big data analytics support remote monitoring, fleet-level optimization, and over-the-air updates, making BMS architectures scalable and continuously improving, although concerns related to cybersecurity and communication latency remain critical (Wei Z. et al., 2023; Song et al., 2022; Jiang et al., 2024).

Overall, while AI/ML techniques significantly enhance estimation accuracy, predictive capability, and operational efficiency of BMS, their practical deployment requires careful consideration of computational requirements, data quality, and system integration challenges.

The comparative analysis reveals that no single AI/ML technique is universally optimal for all BMS applications. ANN and LSTM models demonstrate high estimation accuracy for SoC and SoH prediction, whereas CNN-based architectures are particularly effective for fault diagnosis and anomaly detection. SVM and Random Forest methods provide robust performance with lower computational requirements, making them suitable for embedded systems. However, practical implementation requires balance estimation accuracy, computational burden, data availability, explainability, and hardware constraints. Consequently, recent studies increasingly favor hybrid AI architectures that combines model-based estimation with data-driven learning to improve robustness and real-world applicability.

5 Conclusions and future outlook

This review presented a comprehensive assessment of AI/ML-enabled battery management systems (BMS) for electric vehicle applications through a critical analysis of 120 research articles and review papers. The findings demonstrate that artificial intelligence and machine learning techniques have significantly enhanced conventional BMS capabilities by improving the accuracy of state of charge (SoC), state of health (SoH), and remaining useful life (RUL) estimation, while enabling intelligent fault diagnosis, predictive maintenance, adaptive charging, and thermal management. Among the reviewed approaches, deep learning techniques such as ANN, LSTM, and CNN exhibited superior capability in capturing complex battery nonlinearities and degradation patterns, whereas lightweight machine learning algorithms provided greater feasibility for real-time embedded implementation.

The comparative analysis revealed that the suitability of AI/ML techniques depends on application requirements, computational resources, and data availability. While deep learning methods generally provide higher estimation and prediction accuracy, their deployment is often constrained by computational complexity, training data requirements, and limited interpretability. In contrast, lightweight machine learning approaches offer improved feasibility for real-time embedded BMS applications.

Overall, the reviewed literature highlights that integrating AI-driven estimation with thermal management, safety control, and cell balancing is essential for improving battery reliability and lifespan. Emerging technologies such as FPGA-based edge intelligence and IoT-enabled cloud analytics are accelerating the evolution of BMS toward predictive and autonomous operation. Overall, hybrid frameworks combining physics-based battery models with AI-driven analytics provide the most effective balance between accuracy, robustness, and practical implementation.

5.1 Future outlook

The comparative evaluation conducted in this review indicates that the selection of AI/ML techniques should be guided by application requirements, available computational resources, data accessibility, and safety considerations. While deep learning models generally achieve superior predictive accuracy, lightweight machine learning methods and hybrid frameworks offer greater feasibility for real-time embedded BMS implementation. Consequently, future intelligent BMS platforms are expected to rely on hybrid architectures that integrate physics-based battery knowledge with AI-driven predictive analytics. In addition, extensive experimental validation, hardware-in-the-loop testing, and real-world deployment studies are essential to accelerate the practical adoption of AI/ML-enabled BMS in electric vehicles and energy storage systems.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Glossary

AI artificial intelligence ANNs Artificial neural networks BMS battery management system BMU Battery management unit CNN Convolutional neural network DL Deep learning DNN Deep neural network ECM equivalent circuit models EKF Extended Kalman Filter EVs Electric vehicles FPGA Field programmable gate array GPR Ground-penetrating radar GRU Gated recurrent units IEEE The Institute of Electrical and Electronics Engineers IoT Internet of things KF Kalman Filter LCP Life-cycle performance LIBs Lithium-ion batteries LSTM Long-short-term memory MDPI Multidisciplinary Digital Publishing Institute ML Machine learning MPC Model predictive control OTA Over-the-air PID Proportional integral derivative RL Reinforcement learning RUL Reaming useful life SoC state of charge SoE State of energy SoH and state of health SoP State of power SoT State of temperature SVM Support vector machine SVR Support vector regression UKF Unscented kalman filter XGBoost Extreme Gradient Boosting

Footnotes

Edited by: Gaulthier Rydzek, Université de Montpellier, France

Reviewed by: Nayan Kumar, Government Engineering College, Khagaria, India

Deepak Kumar Singh, Government Polytechnic Aurangabad, Bihar, India