Digital twin technologies for battery systems: advancements, applications, and future directions

1 Introduction

1.1 Lithium-ion batteries and future trends in energy storage

Basic intercalation-type materials have evolved into sophisticated cathode, anode, and electrolyte technologies as a result of significant advancements since the creation of lithium-ion batteries (LIBs) in the 1970s and 1980s. Since the 1970s, advanced technology has made it possible for LIBs to be widely commercialized, beginning in 1991. Although nickel-rich NCM cathodes are high-capacity materials, their increased energy densities raise safety concerns such as gas evolution and thermal runaway, necessitating further study (PbSO et al., 2008). As lithium-sulfur (LSB) and lithium-oxygen (LOB) batteries show promise for developing high-energy storage solutions, LIBs continue to hold their leading position in energy storage. Although alternative batteries have higher energy densities, they operate at developing levels and face difficulties with cycle life and over-potential (Carbone et al., 2017; Shabeer et al., 2025b; Shabeer et al., 2025a).

Because of their high energy density, long lifespan, and suitability for electric vehicle (EV) and hybrid electric vehicle (HEV) applications, LIBs are gaining a lot of attention. By using nickel-manganese-cobalt (NMC) cathodes rather than lithium cobalt oxide (LCO), current research has advanced electrode technology and improved structural stability and storage capacity. Lithium-rich and nickel-rich layered oxide research has increased LIB energy capacity; however, electrolyte degradation and capacity fading remain problems for researchers. Emerging technologies like lithium-sulfur and lithium-air and lithium-CO₂ batteries are the focus of current research because they allow conversion-based energy storage to surpass existing capacity limitations. However, a number of technological obstacles stand in the way of the successful commercialization of developing technology (Nazri et al., 2008; Choi and Wang, 2018; Zhao et al., 2024).

As the world deals with increased energy demands and the challenges posed by climate change, the significance of integrating renewable energy sources continues to grow. Because renewable energy is sporadic, portable electronics like EVs, laptops, and cellphones require battery systems to store it. Accurate modeling of battery charging patterns is necessary for optimal performance, and as battery usage increases, so will this requirement. This study discusses the advantages and disadvantages of the various battery modeling techniques that incorporate electrochemical, mathematical, and circuit-based techniques as well as hybrid models. Through the use of LIB modeling and gray-box/black-box parameter extraction methodologies, the study examines how machine learning and metaheuristic algorithms affect battery management systems (BMS) (Meng et al., 2018).

Power battery modeling systems and state-of-charge (SoC) estimation are necessary for energy management in EVs. A study investigates power battery modeling using data-driven models with strength and weakness analyses, electrochemical mechanism models, and similar circuit models. To evaluate the operational boundaries of SoC estimate techniques in real-world applications, the study further explores their performance. The ability of advanced deep learning techniques, such as gated recurrent units (GRU) and long short-term memory (LSTM) networks, to improve model accuracy and estimation precision is examined (Vaideeswaran et al., 2019).

Developers must concentrate on developing sophisticated BMSs in conjunction with improved battery modeling skills to maximize LIB performance, safety, and durability. The development of complete BMS platforms that integrate features for energy distribution, management, and failure identification with battery condition analysis has lately become the focus of research. Along with many other applications, the use of multiphysics models that combine electrical and thermal activities has become commonplace for EVs. These models provide better simulation accuracy by considering both mechanical stress factors and aging mechanisms. To optimize BMS performance, advanced computational techniques based on digital twins are gaining popularity because they provide precise insights on battery behavior that were previously difficult to get (Xiong et al., 2011; Tapaskar et al., 2024).

Due to its high potential for enhancing battery system performance, digital twin technology is being researched for LIB applications in electric vehicles and renewable energy systems. Digital twins are physical system duplications that assess battery system issues and make forecasts in real time. To improve battery design capabilities and performance, digital twins enable crucial insights on electrochemical behaviors, such as electrochemical reactions and mass transfer. Through a review of current developments involving digital twin-driven all-solid-state batteries and their implications for further research and development, the paper investigates digital twin applications for battery systems (Soleimani et al., 2022; Park et al., 2020; Merkle et al., 2019).

1.2 Overview of digital twins in batteries and applications

Digital twins (DT) technology has become a widely adopted solution in BMS that allows real-time observation and predictive analysis. Operating reliability demands evaluation of LIB degradation because it affects system performance with time. The assessment of LIB performance degradation remains difficult because of changing charging and discharging patterns. The LSTM algorithm enables development of a digital twin together with a health indicator (HI) for implementing a performance degradation model. A developed estimation model uses virtual discharge simulations to determine battery capacity levels through a method that accurately evaluates degradation under dynamic operating environments. The experimental results demonstrate how the model uses measurable parameters during operation to observe performance decline and extend its use to simulate full battery packs. The proposed method allows users to optimize when maintenance should occur as well as schedule missions and extend battery life duration. The development of additional studies should investigate both battery pack simulation parallelization and digital twin model establishment of power systems (Qu et al., 2020; Madani et al., 2025a).

A comprehensive review highlights the role of machine learning, deep learning, and digital twins in predicting and managing the state of health (SoH) of batteries in EVs. Traditional SoH estimation methods are limited, whereas data-driven techniques can model complex relationships between battery parameters and SoH. These techniques enhance traditional approaches and improve result interpretability. The review discusses the challenges and future directions in SoH management using digital twin technologies (Renold and Kathayat, 2024).

Optimization techniques and uncertainty quantification were addressed in digital twins. After discussing virtual-physical world modeling techniques, the study shifts its focus to optimization techniques aimed at improving the precision of digital twins and uncertainty quantification. These methods demonstrate Industry 4.0 readiness for asset management systems and provide better battery lifecycle management through cell retirement optimization in a battery digital twin application (Thelen et al., 2022). Figures 1a,b captures the technological and structural foundations underpinning modern battery intelligence frameworks. Figure 1a presents the evolution of battery modeling methodologies, transitioning from equivalent-circuit abstractions to physics-based electrochemical formulations and, ultimately, to AI-driven data-centric models capable of learning degradation trajectories, nonlinear behaviors, and operational dependencies directly from usage data. Building upon this modeling progression, Figure 1b extends the concept into a multiscale digital-twin hierarchy spanning the entire battery ecosystem, from micron-scale component and material twins at the cell-manufacturing stage, to device-level twins embedded in individual cells and battery modules during vehicle integration, and finally to large-scale system twins governing grid storage assets and fleet-level energy infrastructures. This hierarchical construct highlights how digital twins evolve across manufacturing, deployment, and operational domains, enabling parameter inheritance across scales, lifetime performance tracing, and predictive optimization of cell, module, and system behavior. Together, Figures 1a,b illustrates how advances in modeling paradigms and digital-twin architectures are converging to form an intelligent, multilevel representation of batteries throughout their lifecycle.

1.3 Importance of digital twins in battery systems

High-power LIBs are critical for applications such as EVs, power grid stabilization, and high-pulsed power loads. Traditional experiment-based battery design, however, is costly, inefficient, and provides limited insights into the complex internal processes of batteries. To address this, a digital twin model incorporating electrochemical-thermal coupling has been developed to optimize ultrahigh-power LiFePO₄/graphite LIBs. This model accurately predicts performance and identifies optimal design parameters, enabling a high energy density of 92.38 Wh/kg at 50 °C discharge and stability under 150 °C pulse discharges. By improving Li+ transport and enhancing discharge capacity at extreme rates, this model-driven approach contributes to the advancement of ultrahigh-power LIBs (Zhang et al., 2023). Table 1 provides an overview of applications of digital twin technology in battery management and optimization.

TABLE 1. Applications of digital twin technology in battery management and optimization.

The lifetime of LIBs significantly impacts the operation of EVs and grid energy storage systems. It depends on design and material quality, as well as operational conditions that require improved control strategies. The development of machine learning (ML) and battery degradation modeling diagnostics creates new opportunities for the formation of battery digital twins, which improves control and extends battery life. From employing empirical models, battery control system development advanced to physics-based systems. Despite these restrictions, a few issues and performance deterioration make their practical application difficult. By predicting battery performance and linking charge/discharge trends with capacity loss measurements, machine learning models address this issue. Rapid performance forecasts from hybrid prediction systems made of neural networks and multiphysics allow charging process optimization to proceed more quickly. Artificial intelligence (AI)-driven solutions will address present challenges such as data integration, model accuracy, and real-world application challenges (Wu et al., 2020).

A study utilizing wavelet analysis filters voltage and current signals to estimate the internal complex impedance of a battery based on SoC and SoH. This processed data is then used to create a battery digital twin that outputs realistic voltage signals based on SoC and SoH inputs. Additionally, a novel approach for extracting open-circuit voltage (OCV) from pulsed discharge voltage signals is introduced. The battery digital twin replicates laboratory-tested battery data, making it suitable for monitoring, diagnostics, and testing cell balancing algorithms (Di Fonso et al., 2024).

Because digital twins offer accurate energy system forecasts for diagnosis and optimization capabilities, they hold great promise for advancing EV battery technology. In addition to enabling vehicle to grid (V2G) connectivity, these solutions improve battery engineering, extend battery life, and ensure vehicle safety. Artificial intelligence and machine learning techniques increase the forecast of battery SoC and SoH, optimize performance, and enable sustainable EV infrastructure using digital twins. To improve system operational performance, future technological advancements in digital twins should focus on building safe cyber platforms and different physics simulation methodologies, as well as improved data processing capabilities (Madani et al., 2025a; Rajesh et al., 2024).

By analyzing historical battery data, a digital twin model with AI capabilities is now available to optimize BMS state estimation. The battery status prediction model is an integrated system that makes use of back-propagation neural networks (BPNN) improved by the whale optimization algorithm (WOA). In charging and discharging activities, this model has experimental prediction accuracy levels of 89% and 96%, respectively. Future studies will focus on enhancing accuracy using LSTM deep learning technologies, in addition to the integration of load profile data and equivalent series resistance (ESR) analysis (Li et al., 2024a).

Smart BMS operations benefit from AI-based predictive models when digital twins are integrated since their models deliver higher accuracy for SoC and SoH calculations. This research evaluates deep neural networks (DNNs) and LSTMs as components of digital twin-based BMSs with Shapley additive explanations (SHAP) and local interpretable model-agnostic explanations (LIME) and surrogate models for model explanation using explainable artificial intelligence (XAI) techniques. The research findings confirm LSTM superiority over DNNs because LSTMs show better performance when processing temporal data from evolving BMS datasets. The study needs examination of predictive models including support vector regression (SVR) and GRUs to boost robustness improvements (Njoku et al., 2025a).

A study built a technological framework by uniting digital twin technology with deep deterministic policy gradient (DDPG) method for battery degradation process analysis alongside SoH evaluation and fuel consumption measurement for intelligent transportation systems (ITS). The digital twin system tracks battery deterioration by analyzing connections between SoC and health indicators and their relation to cell voltage values. A cloud-based DT solution delivers greater performance than static models during all predictive maintenance operations and energy efficiency optimization tasks. The primary areas of investigation target predictive accuracy enhancement alongside development of a model which provides broader utility across EV platform domains (Saba et al., 2024).

Using digital twin technology, the study presents a real-time predictive monitoring system that focuses on new energy commercial vehicle battery mounts. To assess structural behavior in response to variations in temperature and loading conditions, the digital twin structural-reliability assessment framework (DTMAR) model employs a hybrid machine learning model in conjunction with a finite element model and neural network surrogate (response-surface) model (NN-RSR). In addition to improving design processes, DTMAR is a technique for accurately assessing structural reliability and tracking the condition of equipment while it is in use (He et al., 2023).

Cyberattacks and equipment malfunctions have the potential to interfere with BESS system functionality. A digital twin model should be utilized for frequency management since it requires dependable performance. Along with digital battery supervision and security threat identification, the system produces comprehensive battery predictions that improve asset monitoring and attack prevention. According to the study, when compared to conventional model-based forecasting methods, artificial neural networks (ANN) outperform them in terms of SoC forecasting outcomes. Because it exhibits the best accuracy, the GRU model with six hidden layers and 20 iterations can be applied to cybersecurity platforms. Future studies will evaluate real-world SoC data in addition to looking at other machine learning approaches (Kharlamova et al., 2022).

Battery digital twins play a key role in enhancing battery operations and design during maintenance, cutting costs, and promoting environmentally friendly projects. By reducing emissions while preserving regulatory compliance and raising overall battery costs over the course of their lives, the technology serves two functions (Madani et al., 2025b). Since BMS and electrical components have gotten the greatest attention to date, most of the research on battery digital twins that is currently available comes from the EV industry. Even while scholarly research on battery DTs has been performed, more study is necessary for practical implementations. By examining DTs’ applicability throughout all battery industry processes, the paper explores the possible applications of DTs. Before requiring a more thorough analysis of battery DT use in various applications, the review integrates electrical and software viewpoints through an assessment process of various DT platforms and industry insights that uncovers potential and difficulties.

2 Digital twin applications and technologies

Real-time sensor information is necessary for the precise prediction of LIB’s SoC and SoH. Continuous battery health tracking is made possible by a DT that incorporates this data, and it motivates both performance improvement and preventive maintenance initiatives. The battery system operates more reliably and efficiently due to the real-time monitoring system (Bhatti et al., 2021). To achieve this level of precision, digital twin models employ a variety of simulation techniques, including electrochemical, mechanical, and thermal modeling, which replicate the complex behavior of batteries. By using these approaches, the predictive accuracy is significantly improved, ultimately contributing to an extended battery lifespan and more effective operation (Njoku et al., 2025b). In this context, BMS play a crucial role by continuously monitoring, estimating, and controlling battery states to ensure safe and efficient functioning. The integration of digital twins into BMS further refines the estimation of SoC and SoH, leading to enhanced situational awareness and optimized battery performance (Jafari and Byun, 2022). However, as battery systems evolve, so do the challenges. Reconfigurable battery systems (RBS), designed to adjust their configuration in response to fluctuating load demands, must address issues such as thermal management, short-circuit risks, and cost-effectiveness. A DT-based intelligent system framework offers a solution by optimizing the design and operation of RBS, supported by cloud-based big data storage for comprehensive lifecycle management (Garg et al., 2022). This holistic approach helps mitigate potential risks while maximizing the efficiency of RBS, ensuring their reliable operation over time. Table 2 describe key findings and conclusions from studies on digital twin technology in battery management and optimization.

TABLE 2. Key findings and conclusions from studies on digital twin technology in battery management and optimization.

2.1 Digital twin in battery life cycle and end-of-life management

A digital twin-based framework for lithium-ion battery packs in EVs enables real-time monitoring, intelligent management, and autonomous control throughout the entire battery lifecycle. This framework encompasses key stages such as design, manufacturing, operation, and second-life applications, addressing critical challenges like safety, durability, and efficiency. A case study highlighted the positive impact of incorporating digital twins, demonstrating improved battery safety and an extended service life. Future research will focus on developing robust virtual battery models, enhancing IoT infrastructure, securing cloud computing, and integrating data-driven models with physics-based approaches to further optimize battery management (Anandavel et al., 2021).

Beyond the general lifecycle, digital twins offer significant potential in end-of-life (EOL) management, particularly for battery recycling and disassembly. The integration of the human-robot collaboration disassembly (HRCD) approach with digital twin technology addresses challenges related to varying battery models and conditions. This system uses a genetic algorithm (GA) for sequencing tasks and transfer learning to adapt strategies across different battery types, enhancing disassembly flexibility and efficiency. The framework effectively balances tasks between human and robotic operators, reducing disassembly time and adapting to various EOL conditions (Madani et al., 2025b; Qu et al., 2023).

In addition to lifecycle and EOL applications, digital twins have proven valuable in more specialized domains, such as fission battery systems. When coupled with machine learning and physics-based models, digital twins can predict and regulate performance, improving safety and efficiency. A recent study on fission batteries showed that a digital twin could autonomously adjust system parameters, predict temperature fluctuations, and mitigate risks, demonstrating its capacity for anticipatory control and risk reduction, especially in complex systems like nuclear energy (Wilsdon et al., 2023).

For real-time degradation prediction in lithium-ion batteries, a digital twin framework integrates machine learning models to forecast battery behavior with high accuracy. This framework uses a BPNN for voltage curve completion, achieving 99.6% accuracy, and a convolutional neural network (CNN)-LSTM-Attention model for capacity prediction with over 99% accuracy. By leveraging cloud-based storage for extensive historical data, this system enhances battery management and lifespan forecasting (Li et al., 2024b).

Further, a digital twin for lithium-ion batteries has been developed to capture real-time data and model complex interactions, including the solid electrolyte interphase (SEI) layer, anode crack propagation, and lithium plating. This model provides predictions on battery aging behavior at both macroscopic and microscopic levels, accurately estimating capacity fade. It also identifies aging mechanisms like SEI growth and anode cracking, showing that multi-step charging protocols (MCCCV) are less damaging compared to continuous charging methods (Guo et al., 2023).

Lastly, a study integrated a fully physics-based combustion engine model into real-time co-simulation with an electrical power plant model to optimize renewable energy systems, including battery storage. The study demonstrated the feasibility of real-time simulation with high accuracy, though further simplifications are needed to fully integrate the combustion engine model into a power plant digital twin (Söderäng et al., 2022). This integration underscores the growing potential of digital twins in optimizing energy systems and advancing battery technologies across different sectors.

2.2 Applications of digital twins in battery systems

Digital twins enable continuous monitoring of battery performance, optimizing charging and discharging cycles while facilitating predictive maintenance. A Microgrid Digital Twin (MGDT) model simulates the dynamics of a microgrid (MG) to estimate the required cycle count and stress levels of a battery energy storage system (BESS) without needing specific battery type details. This model assists designers in selecting the appropriate BESS during the planning and expansion of MGs. By leveraging forecast data on irradiance and demand, the MGDT predicts battery stress and cycle count, supporting cost-effective BESS selection and maintenance scheduling. It also helps predict the health of BESS in existing MGs, enabling predictive maintenance and enhancing operational efficiency despite not having an exact BESS model (Padmawansa et al., 2023).

Additionally, a digital twin framework for real-time SoH estimation of LIBs addresses the limitations of traditional end-of-cycle estimation methods. This framework enables real-time SoH estimation based on partial discharge data, integrating energy discrepancy-aware cycling synchronization, a time-attention SoH estimation model, and similarity analysis-based data reconstruction. Experimental results demonstrate that this method achieves real-time SoH estimation with errors below 1% for most sampling times, providing substantial improvements over conventional methods (Qin et al., 2023).

A cloud-based BMS integrates cloud computing and the Internet of Things (IoT) to enhance battery system safety and performance. By utilizing a digital twin of the battery, the system employs diagnostic algorithms for SoC and SoH estimation. Adaptive extended H-infinity filters are used for SoC estimation, while particle swarm optimization (PSO) is applied for SoH monitoring. Experimental validation in both stationary and mobile applications marks a significant step in transitioning the digital twin concept from theoretical research to practical, real-world industrial applications (Li et al., 2020).

Digital twins also play a pivotal role in continuously monitoring and optimizing battery performance. One study presents a multi-purpose BESS integrated into a large wind farm to improve system stability. The BESS mitigates fluctuations in active and reactive power, participates in frequency regulation, and dampens low-frequency oscillations. Digital twin technology is used to optimize the BESS’s capacity, with no adverse effects observed from integrating the controls. This integration significantly enhances system design and performance, highlighting the key role digital twins play in improving the stability and economic efficiency of BESSs (Li et al., 2024c).

Furthermore, a study proposes a four-layer networked architecture for cloud-side-end collaboration aimed at improving the efficiency and performance of BMS. The digital twin model enables precise real-time management of LIBs, supporting state estimation and cloud-assisted equalization. The architecture enables the implementation of high-performance algorithms, with results confirming accurate state estimation and successful battery equalization, paving the way for more sustainable and efficient e-mobility solutions (Wang et al., 2022).

Finally, ongoing research explores the integration of digital twin technology into BMS to address challenges such as battery aging, safety, and performance optimization in new energy vehicles. The study emphasizes the potential of digital twins to provide solutions for battery modeling, real-time state estimation, dynamic charging control, and thermal management. While digital twins offer considerable promise for improving BMS functions, challenges remain, such as aging mechanisms, data privacy concerns, and the integration of AI with big data (Wang et al., 2021).

2.3 Advancements in digital twin models for battery systems

A study presents a physicochemical model-based digital twin for Li-S batteries, integrating continuum ion transport equations that account for species migration, dissolution, precipitation, and electrochemical reactions. The model incorporates the cathode microstructure, including pore size distribution, hierarchy, and tortuosity, and accurately predicts experimental data across various case studies. Notably, it reveals that a cathode with 70 wt% sulfur and porous hollow particles traps polysulphides, optimizing sulfur utilization. The study also identifies limiting factors for sulfur utilization, such as undissolved sulfur in micropores and sulfur leaching near the separator. This computationally efficient model supports routine evaluation of cathode microstructures and compositions, with potential updates to incorporate solid-phase reactions and precipitate formats in the future (Grabe et al., 2023).

Building on this, another study introduces a calendering discrete element method (DEM) model to examine the impact of calendering on the electrochemical performance of LiNi₀.₃₃Mn₀.₃₃Co₀.₃₃O₂-based cathodes. The model is coupled with a coarse-grained molecular dynamics model for slurry equilibration and drying, as well as a 4D-resolved finite element method (FEM) model for performance prediction. The findings highlight how calendering affects the electrode mesostructure, influencing key factors such as pore size distribution, tortuosity, and particle arrangement. This comprehensive modeling approach offers valuable insights for optimizing LIB manufacturing, with implications for both improved performance and cost reduction, contributing to the development of a DT for LIB manufacturing (Ngandjong et al., 2021). The framework includes four layers: physical system data collection, data processing with noise reduction using improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN), SoC estimation models, and visualization.

By combining various SoC models such as k-nearest neighbors (KNN), SVR, light gradient boosting machine (LightGBM), gradient boosting decision tree (GBDT), GRU, and extra trees (ET) with an ensemble weighting network (EWN), the framework significantly enhances performance through time and accuracy-based weight scaling strategies. Experimental results show that the EWN model outperforms other models, achieving root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) values lower than 1.10%, 0.96%, and 1.92%, respectively. Demonstrating robustness across different temperatures and battery types, the framework paves the way for integrating advanced SoC estimation models to further enhance battery management, particularly for electric vehicles and renewable energy storage (Wu et al., 2024).

2.4 Digital twin for battery failure data generation

A DT framework significantly enhances battery lifecycle management by optimizing performance throughout the R&D, manufacturing, operation, and post-operation phases. In the R&D phase, DT integrates data from various technical fields to simulate and refine battery performance, ensuring efficient design. During manufacturing, digital production lines and real-time data feedback continuously improve processes, while in operation, DT simulations detect faults and adjust battery management strategies to extend battery life. In the post-operation phase, recycling frameworks support the recovery of materials through remanufacturing, addressing potential material shortages. However, challenges remain, including the need for multidimensional data, consistent model verification, and real-time model evolution. Emerging technologies such as IoT, 5G, AI, and cloud integration hold the potential to further optimize DT applications in battery manufacturing (Garg and Panigrahi, 2021).

Figure 2 illustrates the multi-layered architecture of digital twins within the battery value chain, emphasizing their scalability, interoperability, and lifecycle integration. Part (a) presents a hierarchical mapping of digital twin deployment from micron-level representations of raw materials and cell components to centimeter-scale battery modules, device-level twins, and ultimately meter-scale system and grid-level twins managed by asset operators. This progression highlights how detailed material-level information provided by cell manufacturers propagates upward to inform system-level lifetime predictions, performance assessments, and operational strategies. Part (b) extends this hierarchy into a functional ecosystem enabled by cloud and edge computing. Here, data originating from mining, material synthesis, cell fabrication, pack construction, and controlled testing are continuously transferred into cloud-based platforms where OEMs and operators leverage diagnostics tools, thermal management models, safety analytics, and prognostic algorithms. These insights drive downstream decision-making for battery operation, fault detection, load management, second-life utilization, and end-of-life logistics such as disassembly and recycling. Together, Figure 2 demonstrates how digital twins not only span multiple physical scales but also unify disparate lifecycle stages, creating a cyber-physical continuum in which cause-effect relationships can be tracked, analyzed, and optimized across decades of battery use.

Building on this, another study demonstrates the application of DT technology in generating data for early warning algorithms, particularly for internal short circuits. A thermal-electrical coupled model simulates battery behavior during faults, incorporating variables like load current, short circuit position, and ambient temperature, which influence failure characteristics. A key finding is the temperature mismatch between sensor locations and fault points, which can delay fault detection. Figure 3 presents a unified perspective on how digital twin frameworks support fault analysis throughout the lifecycle of battery systems. The workflow in part (a) depicts the integration of real-time online fault diagnosis with offline model training, where data originating from operating environments, abnormal system behavior, and multi-source diagnostic interfaces are parsed, cleaned, and matched with predefined fault categories. These data streams feed cloud-based machine learning services that enable iterative parameter optimization, joint prediction, and continuous model refinement, ultimately producing fault risk assessments and early-warning indicators that enhance operational safety. Complementing this diagnostic pipeline, part (b) illustrates a comprehensive map of fault mechanisms, capturing the diverse origins of battery failures, including thermal imbalance, mechanical stress, calibration inaccuracies, manufacturing defects, electrical anomalies, and external environmental triggers, and tracing their propagation into severe events such as capacity fade, electrolyte leakage, internal short circuits, and thermal runaway. In (Yuan et al., 2023) identifies the thermal runaway threshold and illustrates how DT accelerates the development of fault diagnosis algorithms. This approach significantly reduces testing time and costs, generating 78 fault cases in 1 week instead of several months, thus enabling faster development of detection algorithms.

Figure 3 presents an integrated digital-twin framework for battery fault management, linking real-time diagnosis with data-driven prognostics and multiscale failure analysis. Part (a) illustrates the operational pipeline in which heterogeneous fault signatures, arising from electrical, thermal, mechanical, and aging conditions, are captured within the operating environment and processed through an onboard diagnostic interface. Sensor data are parsed, cleaned, and matched against predefined fault categories, enabling multi-fault coupling and risk assessment. If deviations exceed adaptive thresholds, early-warning signals and safety actions are triggered. Simultaneously, the diagnostic data streams are uploaded to cloud platforms, where machine-learning models perform batch-wise preprocessing, parameter optimization, and joint training to refine predictive accuracy. These continuously improving models are then redeployed to the edge layer for online state prediction, forming a closed feedback loop in which offline learning enhances real-time diagnostics, reliability, and system stability.

Part (b) complements this workflow by mapping the generative pathways of battery faults from their root causes to catastrophic failure modes. The diagram identifies primary triggers, including manufacturing defects, environmental stressors, operational misuse, thermal imbalance, and sensor inaccuracies, and traces their escalation through intermediate degradation phenomena such as capacity fade, impedance rise, electrolyte decomposition, gas evolution, lithium dendrite growth, and separator damage. These interactions converge toward hazardous end states such as internal short circuits, thermal runaway, smoke emission, fire, or explosion. By visualizing both the origin and propagation chains, Figure 3 highlights the value of digital twins in correlating causal factors with fault evolution, enabling predictive maintenance strategies that intervene before irreversible damage occurs.

2.5 Optimizing fast charging protocols with battery digital twin models

A new approach to optimizing fast charging protocols using a battery digital twin model (BDTM) and Bayesian optimization (BO) is proposed to balance charging time and battery health. This method reduces the computational complexity of electrochemical models by utilizing BO to find near-optimal solutions more efficiently. The study reveals that four parallel optimization channels reduce the number of evaluations required, shortening the optimization time. Compared to traditional charging protocols, the optimized strategy decreases lithium inventory loss by up to 4.76% within 90 cycles. The research also demonstrates the impact of aging parameters and ambient temperature on charging efficiency. The integration of data-driven algorithms with BDTMs enhances real-time charging optimization, providing practical implications for future electric vehicle implementations (Luo et al., 2024).

2.6 Thermal management in battery systems using digital twin technology

A microchannel liquid cooling battery thermal management system (BTMS) integrated with a DT framework is proposed to mitigate the impact of high temperatures and high discharge rates on battery modules. The system uses a computational fluid dynamics (CFD) model combined with Gaussian process regression to optimize cooling capacity. Results show that microchannel plates improve coolant uniformity and enhance direct cooling performance, reducing maximum temperature and temperature difference by 4.02 °C and 5.05 °C, respectively. Structural parameters influence heat dissipation, while increasing coolant flow alone provides limited improvements. The DT framework effectively analyzes factor interactions and enables global optimization, facilitating more efficient and intelligent BTMS designs (Xu et al., 2022; Mevawalla et al., 2022).

2.7 Psuedo digital twin technology for optimizing battery performance in energy storage systems

A study investigates the implementation and functionalities of DT technology in the battery industry, specifically for optimizing battery performance across its lifecycle, including degradation evaluation, usage optimization, manufacturing inconsistencies, and second-life applications. The study uses the Doyle-Fuller-Newman (DFN) pseudo-two-dimensional (P2D) model to assess the DT’s effectiveness in enhancing battery representation, performance estimation, and behavioral predictions based on real-world data. The implementation approach follows five structured steps, distinguishing DTs from traditional battery models. Challenges include integrating operational battery data via cloud-based solutions and updating model parameters over time. Despite these challenges, the use of DTs improves performance estimation, optimizes strategies, and integrates lifecycle attributes, extending BMS capabilities for battery degradation assessment. However, challenges related to cost uncertainty, infrastructure demands, and return-on-investment (ROI) remain. Key performance indicators (KPIs) outlined in this study will help quantify DT attributes, with future research focused on refining DT feasibility for battery applications (Singh et al., 2021). Recent developments have further coupled the physics-based P2D model with data-driven architectures, such as LSTM networks, enabling adaptive parameter estimation and improved predictive accuracy under dynamic operating conditions. This hybrid approach bridges electrochemical interpretability with real-time learning capability, reinforcing the digital twin’s role in lifecycle performance optimization.

Figure 4 illustrates a hybrid digital twin architecture that fuses a physics-based pseudo-two-dimensional (P2D) electrochemical model with a data-driven long short-term memory (LSTM) network to achieve adaptive and predictive battery behavior modeling. The P2D model provides interpretable physical states, such as voltage, temperature, and state of charge (SOC), along with core electrochemical parameters, while the LSTM module learns temporal patterns from sequential input data consisting of current, voltage, temperature, and SOC histories. By iteratively refining model parameters based on the discrepancies between predicted and observed outputs, the system enables continuous updating of the P2D model and enhances prediction accuracy across varying operating conditions.

Figure 4 illustrates a hybrid digital-twin architecture in which a physics-based pseudo-two-dimensional (P2D) electrochemical model is coupled with a data-driven long short-term memory (LSTM) network to enable adaptive, high-fidelity battery state prediction. In this framework, the P2D model provides mechanistic outputs, including voltage response, temperature evolution, and state of charge (SOC), derived from embedded electrochemical parameters. These outputs are continuously fed into the LSTM network, which learns temporal dependencies from historical input sequences comprising current, voltage, temperature, and SOC measurements. The LSTM module identifies latent system dynamics and estimates updated parameter values that reflect real operating conditions, degradation progression, and usage variability. These updated parameters are then fed back into the P2D model, closing the loop for iterative model refinement. By combining the interpretability and physical rigor of the P2D formulation with the adaptability and pattern-recognition capabilities of neural networks, Figure 4 demonstrates how hybrid digital twins overcome limitations of standalone modeling techniques, yielding a self-evolving virtual replica capable of accurate prediction, anomaly detection, and condition-aware battery management.

2.8 Integration with battery management systems: real-time state-of-charge and health estimation

Digital twins are integral to estimating the SoC and SoH of lithium-ion batteries, providing valuable insights into battery performance, degradation, and operational safety. One study develops a digital twin for tracking SoC and SoH using data from an e-Golf’s diagnostics interface. The system employs an OBD logger to record cell-level data, which is transmitted to a cloud-based platform. Battery models applied to this data estimate parameters such as internal resistance, highlighting significant variations across cells. The study finds that accurate state estimation relies on segmenting data at different levels (pack, module, cell) and considering cell characteristics like OCV. Despite limitations in data logging, the study emphasizes the importance of long-term data collection for statistical significance (Merkle et al., 2021; Madani et al., 2025c; Madani et al., 2025).

Figure 5 illustrates how initial parameter inconsistencies originating at the cell level, such as variations in capacity, internal resistance, state of charge (SoC), and manufacturing tolerances, create non-uniform electrical and thermal behavior within a battery pack. These discrepancies accumulate as cells are interconnected, leading to uneven current distribution, SOC imbalance, and differing aging rates across cells. As the figure shows, these local inconsistencies evolve into pack-level performance deviations, complicating health estimation algorithms, accelerating degradation, and reducing the accuracy and reliability of SoC and state-of-health (SoH) predictions.

Figure 5 demonstrates how localized parameter inconsistencies at the cell level evolve into large-scale performance disparities across an entire battery pack. The diagram traces the progression from fundamental sources of variation, such as material impurities, manufacturing tolerances, and unequal contact resistances, to initial cell-level divergences in capacity, resistance, and state of charge. These intrinsic discrepancies disrupt current distribution and temperature uniformity during operation, generating imbalance in SOC trajectories, depth-of-discharge behavior, and thermal responses across cells. As these deviations accumulate, the pack exhibits non-homogeneous performance characteristics, including accelerated aging of certain cells, unequal utilization rates, and increased stress on control algorithms. Ultimately, these effects manifest as measurable pack-level performance degradation, reduced reliability, and compromised safety margins. By clarifying this causal chain, Figure 5 highlights the critical need for digital-twin-enabled monitoring, parameter harmonization, and predictive control strategies to mitigate the cascading influence of microscopic inconsistencies on macroscopic battery behavior.

Expanding on this, another study introduces a digital twin framework for SoC estimation using an industrial internet-of-things (IIoT)-based system and Microsoft Azure services. This system integrates real-time measurements with machine learning techniques, achieving high accuracy in SoC estimation, demonstrated by a normalized root mean square error (NRMSE) of 0.00047. The framework also incorporates historical driving data from APIs to refine the model, with future research focusing on integrating additional data sources, such as charging infrastructure and grid information. These developments aim to enhance the reliability, adaptability, and predictive accuracy of the digital twin in real-world applications (Issa et al., 2023).

Figure 6 delineates the information flow that transforms raw pack-level measurements into actionable control signals within a digital twin-enabled battery management architecture. The process begins with distributed sensing across the battery pack, capturing cell voltages, currents, and temperatures, which are aggregated and processed as system-level variables. These measurements feed algorithmic layers responsible for state-of-charge estimation, thermal management, and capability assessment, each generating diagnostic and prognostic indicators used for functions such as charge/discharge current limitation, cycle-life prediction, and remaining useful life (RUL) forecasting. Complementary paths support cell balancing, fault diagnosis, leakage detection, and operational safeguards. The resulting decisions and status signals are integrated and transmitted through a communication interface, ultimately directing a controlled transceiver that executes real-time adjustments to maintain safe, efficient, and balanced pack operation. By articulating this end-to-end signal pathway, from sensor acquisition to digital twin-driven actuation, Figure 6 underscores how cyber-physical integration enables predictive, intelligent, and self-adaptive battery control strategies.

In a related study, a DT-supported method for battery state estimation is proposed, utilizing a temporal convolutional network (TCN) and LSTM for accurate estimates of SoC, SoH, and remaining useful life (RUL). The 4-layer hierarchical DT model tackles computational and storage challenges by dynamically updating neural network parameters using transfer learning. This approach significantly outperforms conventional methods, achieving low RMSE values (SoC: 1.1%, SoH: 0.8%, RUL: 0.9%). Future research could focus on refining the framework, incorporating dynamic models, and enhancing real-time adaptability using advanced computing techniques (Zhao et al., 2025).

2.9 Optimization and control of battery management with digital twin technology

A study presents a digital twin-driven framework for BMS aimed at overcoming challenges such as limited computing power, data storage constraints, and inaccurate SoC estimation. The proposed joint HIF-PF online algorithm, tested under the Beijing bus dynamic stress test conditions, significantly outperforms conventional methods, achieving an average SoC error of just 0.14%. The system provides real-time monitoring and visualization of battery parameters, with future work focusing on implementing a closed-loop feedback system for automatic adjustments to further enhance adaptability and accuracy in battery management (Tang et al., 2022).

Figure 7 illustrates a multilevel battery management architecture in which physical batteries and their digital counterparts are tightly coupled through cyber-physical data exchange and advanced computational services. Part (a) shows how sensor-derived measurements of voltage, current, temperature, and mechanical stress from real batteries are continuously fed into a digital twin, where artificial intelligence, cloud computing, big-data analytics, blockchain security, and IoT connectivity constitute the core technological enablers. These foundations support a spectrum of high-level BMS functions, including remaining useful life (RUL) prediction, state-of-health (SOH) estimation, thermal analysis, optimization of charging strategies, and product lifecycle decision-making, while enabling visualization and feedback loops that refine control actions in real time. Part (b) details the hierarchical interaction between local and cloud BMS modules. Local BMS units perform essential on-board tasks such as cell balancing, embedded processing, threshold detection, and preliminary data acquisition. The collected metadata are transmitted via an IoT gateway to cloud-based BMS services, where machine learning, parallel computation, and analytic tools generate enhanced diagnostic insights, predictive models, and visualization dashboards. These refined outputs are then sent back to the local layer, enabling adaptive, intelligent, and feedback-driven control. Together, Figure 7 demonstrates how digital twin technologies elevate traditional BMS frameworks into scalable, interconnected, and self-optimizing ecosystems that support reliable, efficient, and data-centric battery operation across the entire lifecycle.

Another paper discusses the application of digital twins in battery cell production, proposing a framework to optimize sustainable manufacturing processes. The study emphasizes the importance of standardized definitions for digital twins across different contexts (machine, building, product) to ensure effective integration. By demonstrating prototypical implementations, it shows how digital twins can collaborate to conserve resources, improve efficiency, and optimize production steps. The research also addresses the challenges of transitioning from product to machine twins, laying the groundwork for further exploration of digital twin applications in manufacturing (Kies et al., 2022).

A review of advancements in BMS from 2006 to 2020 explores key BMS functions such as cell balancing, thermal management, and SoC/SoH estimation. A study identified gaps in current methodologies and presents a framework for improving BMS efficiency, particularly for electric vehicles. It highlights the potential of emerging technologies, such as digital twins, cyber-physical systems (CPS), and battery swapping, while emphasizing the need for expertise in battery technology to address challenges like range anxiety, slow charging, and battery cost (Panwar et al., 2021).

Figure 8 delineates the interconnected pathways through which operational stressors, material degradation, and electrochemical side reactions collectively accelerate battery aging and precipitate thermal runaway. The left panel identifies the primary influence factors, temperature, pressure, current/voltage load, and time, that govern the aging trajectory of electric vehicle battery systems. These conditions activate fundamental aging mechanisms, including lithium precipitation, SEI layer formation and growth, stress-induced material deformation, and transition-metal dissolution. As these mechanisms evolve, they induce macroscopic performance decay characterized by loss of active material, reduced lithium inventory, impedance rise, declining capacity, and diminished power capability. The right panel links these degradation effects to high-risk thermal behaviors. Electrolyte decomposition, lithium plating, particle cracking, and SEI/CEI breakdown generate gas accumulation, heat generation, and reactive species that undermine structural integrity, lowering thermal stability thresholds. This cascade ultimately manifests as sudden voltage collapse, uncontrolled temperature escalation, and catastrophic thermal runaway. Figure 8 therefore highlights how digital-twin-driven analysis enables the tracing of mechanistic aging origins to hazardous end states, providing a predictive basis for mitigating battery safety risks before irreversible failure occurs.

3 Technologies involved

Digital twins and machine learning are cognitive techniques that improve battery management systems by evaluating predicted data, detecting systems instantly, and increasing capacity. As DTs develop performance simulations for dynamic battery modifications, ML models improve the precision of state-of-charge, state-of-health, and remaining usable life estimates. Different studies attempt to improve battery health analysis efficiency and diagnosis quality by combining deep learning and reinforcement learning methods. Table 3 summarizes the key findings for applications and advancements in machine learning and digital twin models for battery management.

TABLE 3. Applications and advancements in machine learning and digital twin models for battery management.

3.1 Machine learning and AI in digital twin models

Machine learning models, including AdaBoost, LSTM, and Extreme Gradient Boost (XGBoost), are integrated into DT frameworks to enhance the accuracy of SoC and SoH predictions. These models are optimized using metaheuristic algorithms like Improved Gray Wolf Optimization (IGWO) to improve predictive performance, offering valuable insights into battery behavior and lifespan (Nair et al., 2024).

AI and ML techniques are increasingly incorporated into digital twin models to optimize battery performance, enhance predictions, and detect anomalies. Techniques such as neural networks, support vector machines, fuzzy logic, and deep learning are applied to improve SoC and SoH estimation for Li-ion batteries in EVs. These methods contribute to the efficiency and safety of EV batteries by refining state estimation, while the integration of cloud-based systems facilitates real-time data processing and inter-vehicle data sharing, improving BMS (Bin Kaleem et al., 2024).

However, challenges persist when using ML-driven state estimation, particularly the need for large datasets, effective data management, and robust communication protocols. Issues such as data privacy, security, and handling large data volumes must be addressed to fully leverage ML and digital twin approaches. Despite these obstacles, these technologies hold great promise for improving EV battery performance and state estimation in real-world conditions (Rivera-Barrera et al., 2017).

Innovative applications of digital twin technology also contribute to advanced monitoring of EV batteries. For example, a digital twin model combining LSTM algorithms for accurate SoC predictions and a cloud-based IoT network allows for real-time monitoring without requiring sensor calibration or additional hardware. This model also integrates a time-series generative adversarial network (TS-GAN) to generate synthetic data, addressing challenges related to real-time data availability (Pooyan et al., 2023).

Another key challenge in the electric vehicle industry is scaling battery production. Digital twin technology is applied to optimize robotic work cells for efficient assembly of battery modules, overcoming production bottlenecks. A three-stage digital twin methodology is used to enhance assembly lines, meeting the growing demand for zero-emission vehicles. This solution emphasizes the importance of cross-sector collaboration to develop cost-effective, fast, and sustainable battery production processes (Sharma and Kumar Tiwari, 2023).

Digital twin technology also plays a significant role in forecasting SoC for BESSs used in frequency regulation. Machine learning methods such as recurrent neural networks (RNNs), LSTM, GRU, random forest (RF), and AdaBoost have been tested for accurate forecasting. The results highlight the potential of these methods to model digital twins of utility-scale BESSs, offering significant insights for frequency regulation applications (Kharlamova and Hashemi, 2023).

Furthermore, digital twin simulations have been used to model the relationship between microstructural properties and the performance of lithium-ion batteries. Multiphysics approaches, including AI-driven multiscale modeling and dynamic simulations, provide valuable insights for optimizing battery design and improving performance and safety. These advancements underscore the growing potential of digital twin technology to enhance lithium-ion battery design for real-world applications (Kim et al., 2024).

Digital twin models have also been used to monitor and predict battery temperature and degradation patterns. By incorporating LSTM neural networks, the framework achieves high accuracy in real-time temperature predictions, optimizing BTMS and improving battery longevity. This model can also monitor battery health, enhancing the overall performance of battery systems (Yi et al., 2023).

The concept of Dual Digital Twin (DDT) is introduced for real-time monitoring and control of battery systems in electric vehicles. The DDT framework leverages cloud-edge collaboration and AI models to improve battery state predictions and reduce control penalties. Simulation results using a Li-ion battery digital twin model demonstrate improved prediction accuracy, suggesting further research in federated learning frameworks to enhance monitoring while preserving data privacy (Xie et al., 2024).

Finally, digital twin technology is applied to battery cell production, where a virtual digital twin of a flexible cell stack formation machine optimizes production parameters and allows for low-risk testing. This approach improves computational speed, enabling near-real-time performance and ensuring optimal production quality. This application highlights the role of digital twins in enhancing manufacturing processes, reducing waste, and improving production efficiency (Husseini et al., 2022).

3.2 Advanced battery materials and technologies

Advances in battery materials, such as solid-state batteries, are set to enhance the capabilities of digital twins, enabling them to model more complex behaviors and improve performance predictions. A study presents a 3D microstructure-based digital twin model for a LIB module designed for unmanned railway vehicles. Unlike conventional lumped thermo-electrochemical models, this digital twin accurately represents the internal cell structures, allowing for precise calculations of heat generation within each cell component. The study showed higher accuracy in voltage and thermal behavior predictions, with the digital twin model showing good agreement with experimental data across various discharge rates (0.1C–5C). The model also predicts local thermal runaway risks by simulating inhomogeneous temperature gradients, estimating higher local maximum temperatures (137.2 °C vs. 123.9 °C at 10 °C discharge) compared to lumped models. This underscores the potential thermal runaway risks and demonstrates the model’s relevance in real-world applications, such as railway driving patterns. The study highlights the importance of detailed internal structure modeling for accurate thermal analysis and emphasizes the role of digital twin-based battery modeling in optimizing thermal management and safety assessments for advanced energy storage systems (Park et al., 2024).

4 Advancements and future trends

Artificial intelligence is essential to the development of digital twin technology for lithium-ion battery management because it enhances monitoring performance, makes forecasts about remaining useful life, and manages uncertainties more skillfully. Digital twins that incorporate machine learning and deep learning systems increase prediction accuracy levels and allow for real-time system optimization. Future trends are focused on implementing adaptive algorithms with uncertainty quantification capabilities, which will lead to better battery systems. Table 4 outlines the contributions from various studies that apply advancements and challenges in digital twin applications for battery energy storage and management systems.

TABLE 4. Advancements and challenges in digital twin applications for battery energy storage and management systems.

4.1 Advancements in digital twin technology for battery energy storage systems

A study reviewed the advancements in DT technology for BESS, focusing on its functions and architectures. DT integrates with BESS to improve fault detection, prognosis, real-time monitoring, optimization, and temperature control. Key applications of DT are observed in EVs and aircraft. The study identified several key functions of DTs, such as parameter estimation, fault diagnosis, real-time monitoring, and temperature control. However, most studies focus on the physical layer of the DT architecture, with fewer addressing the computing and network layers, suggesting a lack of standard procedures. Trends in research include the use of supervised machine learning for parameter prediction and unsupervised learning for optimization and temperature control, while gaps include a focus on Li-ion batteries and limited integration of BMS in DT architectures (Semeraro et al., 2023a).

4.2 Future trends and opportunities in digital twins for battery systems

Emerging technologies such as AI, edge computing, and deep learning are set to significantly enhance the capabilities of digital twins in battery systems, particularly for real-time optimization and predictive analytics. A study presented a DT for reliability, focusing on life prediction and reliability evaluation of lithium-ion batteries while addressing the stochastic and dynamic nature of battery degradation. The framework integrates various models, including capacity degradation, stochastic degradation, life prediction, and reliability evaluation, using a Bayesian-based adaptive evolution method. This approach achieves a prediction error within 5%, leading to a reduction in predictive maintenance costs by 62.0% and 52.5% for two test batteries. The study underscores the importance of real-time, data-driven battery health monitoring and optimized predictive maintenance to reduce operational costs (Yang et al., 2022). Additionally, a structured approach for designing a digital product twin (DPT) for battery cells was presented. The DPT integrates geometry, material properties, and manufacturing history, enabling optimization of quality, performance, and cost. Key findings highlight the data-driven structuring of the digital twin, comprehensive parameter analysis for simulations and modeling, and a standardized information model for continuous process updates. This facilitates real-time monitoring and optimization of battery production and provides a solid foundation for implementing digital twins in pilot production and quality assessment (Kampker et al., 2023).

As battery digital twins increasingly incorporate data-driven and hybrid artificial intelligence models for state estimation, degradation prediction, and control, model interpretability becomes a critical requirement for safety assurance, validation, and regulatory acceptance. While black-box models often achieve high predictive accuracy, their limited transparency constrains trust and hampers fault diagnosis in safety-critical battery management applications. Interpretable and explainable AI techniques, such as SHAP and LIME, provide systematic frameworks for quantifying feature importance and explaining model behavior at both global and local levels. When embedded within battery digital twin frameworks, these methods enable insight into the influence of operating conditions, thermal states, and degradation indicators on digital twin outputs, thereby supporting verification, certification, and informed decision-making in safety-critical environments (Qu et al., 2025; Chen et al., 2025; Njoku et al., 2025c), (Etem, 2025).

5 Design and implementation

By enabling real-time monitoring and predictive maintenance, DTs offer solutions to current BMS challenges. Their integration significantly improves battery performance, safety, and longevity, benefiting not only electric vehicles but also broader energy storage applications (Gilbert et al., 2023).

5.1 Guidelines for designing a digital twin for Li-ion battery energy storage systems

A study presented guidelines for developing a digital twin for Li-ion BESS, aimed at optimizing system performance and enabling predictive fault detection. The proposed framework consists of three layers: the physical layer (BESS and sensor data), the digital layer (behavioral, geometric, and data-driven models), and the communication layer (enabling data exchange). The digital twin helps with fault detection, system optimization, and performance improvement by providing actionable insights. While the guidelines effectively focus on fault detection and optimization, they are limited in broader applications. This structured approach enhances BESS performance by ensuring the reliability and accuracy of the digital twin (Semeraro et al., 2023b).

To improve practical adoption, the five structured steps of battery digital twin design can be operationalized through a generalized implementation blueprint (Figure 9):

• System definition and sensing strategy, including battery configuration, sensor selection, and data acquisition frequency;
• Model development, integrating physics-based, data-driven, or hybrid representations appropriate to the target application;
• Computing and deployment strategy, specifying on-board, edge, or cloud execution based on latency and hardware constraints;
• Synchronization and adaptation, enabling continuous updating through online learning, parameter estimation, or state correction;
• Verification, validation, and performance assessment, ensuring accuracy, robustness, and safety across operating conditions.

5.2 Real-world implementation of digital twins in battery production

A study proposed a cyber-physical production system (CPPS) using a DT to enhance resilience in rechargeable battery production, addressing performance degradation from complex processes, dynamic disturbances, and uncertainties. A novel model for resilience activities and a DT-based CPPS architecture are introduced, incorporating service composition procedures and asset descriptions for efficient interoperability. The system follows asset administration shell (AAS) principles and ensures both static and dynamic functionalities through service composition phases. The proposed system was validated in an industrial case study, achieving 95.24% accuracy in providing technical functions, minimizing data, analysis, and decision latencies. It supports event diagnosis, dynamic response, and resilient operation, and integrates virtual commissioning, prognostic simulation, and reactive simulation. Future work could involve parallel and distributed DT simulations, enhanced 3D visualization, mixed reality for decision-making, and federated DT for end-to-end supply chain control using reinforcement learning-based production control (Park et al., 2023).

5.3 Extending digital twin frameworks to emerging battery chemistries

While lithium-ion batteries dominate current digital twin research due to their widespread deployment, extending digital twin frameworks to emerging battery chemistries represents an important future design research direction. Chemistries such as lithium-sulfur, lithium-oxygen, and solid-state batteries exhibit degradation mechanisms that differ fundamentally from those of lithium-ion systems, including polysulfide shuttling, interfacial instability, dendrite formation, and solid electrolyte degradation (Sun et al., 2025c).

Adapting digital twins to these systems will require chemistry-aware modeling strategies, tailored sensing and diagnostic approaches, and revised validation protocols capable of capturing these unique failure modes. Hybrid physics-data-driven digital twins, combined with adaptive learning and uncertainty-aware state estimation, are expected to play a critical role in enabling reliable monitoring and control of next-generation battery technologies (Bartolucci et al., 2025; Adam et al., 2025).

For lithium-sulfur systems, digital twins must represent soluble polysulfide speciation, dissolution and migration, and precipitation-induced porosity and tortuosity evolution, since the shuttle effect and morphology changes strongly modulate sulphur utilisation and coulombic efficiency. Physicochemical digital-twin formulations that incorporate continuum transport with microstructure-aware cathode descriptions illustrate how electrode architecture directly constrains performance and degradation, making microstructure-informed parameterisation and online recalibration central requirements for practical Li-S twins (Grabe et al., 2023).

For solid-state batteries, the governing “twin state” must include interfacial contact quality, interphase growth, stress fields, and stack-pressure dependent resistance, because performance and safety are tightly linked to chemo-mechanical coupling and dendrite-mediated shorting risks. Recent solid-state digital twin efforts highlight the importance of resolving microstructure and interface variability, while broader solid-state literature shows that pressure regulation and mechanics-driven evolution are central determinants of stability, suggesting that solid-state battery twins should prioritize interface-aware observables and mechanically consistent reduced-order models suitable for online updating (Wang et al., 2025; Sun et al., 2025a).

For lithium-oxygen systems, extending digital twins require inclusion of oxygen transport, discharge product formation and removal, and parasitic chemistry that drives electrolyte degradation and impedance rise. Mechanistic work on Li-O₂ degradation indicates that failure attribution can be non-intuitive, reinforcing the need for digital twins that track chemical uncertainty, update hypotheses as new evidence arrives, and separate competing degradation channels under realistic operating conditions (Zor et al., 2019; Sun et al., 2025b).

6 Challenges and limitations of digital twins in battery management

A study introduced a self-adaptive DT-based battery monitoring and management system (DT-BMMS) to overcome the limitations of existing digital twins in BESS. The proposed system dynamically adjusts to changes in battery aging and operating conditions, improving decision-making. By integrating a self-adaptive algorithm, the DT-BMMS enhances SoC estimation accuracy using an Extended Kalman Filter (EKF) and Model Predictive Control (MPC) for SoC balancing. The system outperforms traditional methods, achieving precise SoC estimates and equalization in under 2 minutes, even under high or low dynamic conditions. Future work could focus on calibrating the system with real battery data to further validate its performance (Fu et al., 2024).

Beyond technical considerations, the large-scale adoption of battery digital twins is constrained by economic and infrastructural challenges. High-fidelity digital twin implementations often require substantial investment in sensing hardware, data acquisition systems, edge or cloud computing infrastructure, and integration with existing battery management architectures. These requirements increase system complexity and capital expenditure, particularly for industrial-scale and fleet-level deployments. Moreover, the cost-benefit and ROI outcomes of digital twin adoption remain highly application-specific and are not yet well quantified, limiting their near-term deployment beyond research prototypes and high-value use cases. Recent quantitative assessments further highlight the economic uncertainty associated with digital twin deployment at scale. A comprehensive analysis by the National Institute of Standards and Technology (Thomas, 2024) indicates that the cost effectiveness of digital twins depends strongly on system complexity, model fidelity, and the cost consequences of non-optimal operation. While high-fidelity digital twins can deliver substantial economic benefits in complex, high-risk systems, their costs increase nonlinearly with model accuracy, precision, and flexibility, often resulting in diminishing marginal returns for moderate-complexity applications. As a result, digital twins are not universally cost effective, and simplified or hybrid modelling approaches may be more economically viable for many battery management applications, particularly during early-stage deployment or retrofitting of legacy systems. Evidence from large-scale energy and smart-grid digital twin deployments further suggests that infrastructure and integration costs remain major barriers to widespread adoption. Recent study (Al-Shetwi et al., 2025) report that digital twin-enabled energy systems often require extensive sensor networks, high-bandwidth communication infrastructure, cloud or edge computing resources, and skilled personnel for system integration and maintenance. Although such investments have been shown to reduce operational costs by approximately 15%–20% in mature grid-scale applications, these benefits are typically realized only after significant upfront capital expenditure and long deployment timelines. Consequently, the scalability of battery digital twins remains constrained by infrastructure readiness, organizational capability, and the availability of high-quality operational data, particularly outside of large industrial or utility-scale settings.

Another study developed a digital twin for a large-scale 1 MWh grid battery system to assess the impact of cell-level monitoring and control over a 10-year simulation. The study emphasizes the importance of managing cell-to-cell variations in energy storage systems, as these variations significantly impact system performance and lifespan. The results highlight the need for advanced techniques, such as digital twins, to optimize the operation and performance of large-scale energy storage systems, addressing these variations and improving efficiency (Bai et al., 2023).

A major technical limitation of battery digital twin deployment arises from the computational constraints of commercial battery management system hardware. High-fidelity physics-based models and data-driven digital twins often require substantial processing power, memory, and real-time numerical solvers, which exceed the capabilities of typical embedded BMS microcontrollers. As a result, fully integrated on-board execution of digital twin models remains impractical for most commercial battery systems, particularly when strict latency, reliability, and power-consumption requirements must be met (Shen et al., 2025). To address these constraints, most practical implementations adopt hierarchical or hybrid architectures in which only lightweight estimation or control-oriented models are executed on-board, while computationally intensive digital twin functions are offloaded to edge or cloud platforms. Reduced-order models, surrogate machine-learning predictors, and event-triggered updates are commonly employed to balance model fidelity with real-time feasibility. While these approaches enable near-term deployment, they also introduce additional system complexity related to model synchronization, communication latency, and validation across distributed computing layers (Barreto et al., 2025).

A novel data pre-processing method for SoH estimation in lithium-ion batteries is proposed, eliminating manual feature extraction by converting 1D voltage data into 2D data using a sliding window. This method automatically extracts features during training and predicts SoH by forecasting battery voltage in the next cycle. The results show a substantial reduction in RMSE, with a CNN-LSTM model achieving RMSE values below 1.5%, and further optimization techniques lowering it to 1%. This approach offers a promising solution for SoH estimation in battery digital twins, with future work focusing on hybrid models to improve RUL predictions and integrate real-time updates through online learning techniques (Safavi et al., 2024).

Despite growing interest in battery digital twins, standardized design and integration procedures across physical, computing, and network layers remain limited. Most existing implementations are developed for application-specific contexts, resulting in fragmented architectures that hinder interoperability, scalability, and cross-platform deployment. While emerging frameworks in cyber-physical systems, asset administration shells, and cloud-based BMS architectures provide partial solutions, a unified and standardized digital twin reference architecture for battery systems has yet to emerge (Yu et al., 2025).

Data availability and quality remain fundamental constraints for AI-enabled battery digital twins. Robust training of data-driven and hybrid digital twin models requires large volumes of high-resolution, long-term operational data spanning diverse operating conditions, aging states, and failure modes. In practice, such datasets are difficult to obtain due to limited sensor availability, proprietary restrictions, safety considerations, and the long timescales associated with battery degradation. These limitations directly affect model generalizability and reliability when deployed outside controlled laboratory environments (Liu et al., 2025). To mitigate data scarcity, several studies employ synthetic or augmented datasets, including generative models such as TS-GAN. While such approaches can improve data diversity and support algorithm development, synthetic data often fails to fully capture real-world variability, sensor noise, rare fault events, and complex degradation pathways observed in practical battery systems. Consequently, digital twins trained predominantly on synthetic data may exhibit reduced robustness when exposed to unseen operational conditions, underscoring the need for continuous validation and recalibration using real battery measurements (Chatterjee et al., 2025).

Although many battery digital twin frameworks demonstrate promising performance in simulation or laboratory environments, their transition to real-world deployment remains limited. Practical implementation requires systematic calibration using field-acquired battery data and simplification of high-fidelity models to satisfy real-time execution, robustness, and hardware constraints. Bridging this gap is essential for translating digital twin concepts into deployable battery management solutions. The most common challenges associated with digital twin development are summarized in Table 5.

TABLE 5. Key challenges in battery digital twin development and corresponding solution directions.

7 Cloud, edge, cyber-physical systems for battery management

The integration of cloud computing and edge technologies with digital twins enables decentralized control and real-time data processing of battery systems, enhancing both efficiency and safety. From a system-level perspective, battery digital twins extend beyond physics-based modeling and can be formally decomposed into three interdependent layers: the physical layer, the computing layer, and the network layer. The computing layer governs where and how digital twin models are executed, spanning on-board BMS hardware for latency-sensitive state estimation and safety functions, edge computing nodes for near-real-time analytics, and cloud platforms for high-fidelity electrochemical modeling, long-term degradation analysis, and fleet-level optimization. This layered computing paradigm enables scalable deployment by distributing computational workloads according to latency, energy, and resource constraints. The network layer provides the communication backbone that synchronizes physical batteries with their digital counterparts. It encompasses in-pack communication protocols such as controller area network (CAN) bus and on-board diagnostics (OBD) interfaces, IoT gateways for secure data transmission, and cloud-edge communication channels that support bidirectional data flow, model updates, and feedback control. Limitations in network reliability, latency, and interoperability directly constrain real-time digital twin evolution and remain a major barrier to large-scale industrial adoption (Mulpuri et al., 2025; Hananto and Veza, 2025).

One of the key advancements in this computing layer area is the development of cloud-based battery management systems (CBMSs), which leverage cloud computing and battery big data. These systems connect a physical BMS with a cloud-based virtual BMS through communication channels like IoT, offering enhanced computational resources compared to traditional BMSs. This enables the use of advanced digital twin models and algorithms, significantly improving BESSs. However, with these advancements come new challenges, particularly concerning cybersecurity vulnerabilities. If exploited, these vulnerabilities could compromise BESS safety, leading to hazardous and costly outcomes. To address these concerns, a scoping review of BMS cybersecurity explores CBMS architecture, cyberattack surfaces, attack scenarios, impacts at both the component and system levels, and potential countermeasures, while also examining applicable cybersecurity standards and outlining future research directions (Naseri et al., 2023).

Figure 10 illustrates how artificial intelligence, cloud computing, and Internet-of-Things infrastructure converge to construct a scalable digital-twin backbone for cyber-physical battery systems. Massive streams of multi-modal sensor data are collected from the physical entity and uploaded via IoT gateways into cloud platforms, where big-data storage, blockchain-enabled traceability, and AI-driven analytics transform raw measurements into actionable digital-twin representations. These digital counterparts continuously learn system behavior, update state and degradation models, and predict future operating trajectories. The insights generated in the cloud are transmitted back to the physical system through feedback-control loops, enabling real-time optimization of operation strategies, fault-abatement interventions, and lifecycle-aware decision-making. This closed-loop interplay between data acquisition, model adaptation, and intelligent control establishes the foundational architecture required for autonomous, self-evolving digital-twin ecosyste

In parallel, the review of onboard BMS and CBMS architectures highlights the critical role of onboard BMS for ensuring EV safety, while positioning CBMS as a valuable complement for enhancing overall system capabilities. Commercial CBMS applications from companies like Bosch, NXP, and Panasonic demonstrate the market potential of these systems. However, cybersecurity remains a major concern, as attacks targeting confidentiality, integrity, and availability can compromise system functionality. Vulnerabilities in communication interfaces such as CAN bus, OBD-II, IoT gateways, and cloud protocols could lead to serious issues, including denial-of-charging, battery depletion, accelerated aging, and thermal runaway. To counteract these threats, the review discusses various countermeasures, such as blockchain, encryption, authentication, and resilient algorithms, while identifying gaps in CBMS cybersecurity standardization. The guidelines set forth by standards such as SAE J3061 and ISO 21434 provide valuable recommendations for future research in this area, ensuring that cybersecurity is addressed as part of the ongoing development of CBMS technologies (Naseri et al., 2023).

Alongside cybersecurity, the broader challenge of optimizing CPS for battery management continues to gain attention in both academia and industry. CPS architectures, which bridge the physical and digital worlds, are being increasingly integrated with edge computing and edge-cloud computing to address the massive volumes of data generated by battery systems. However, integrating edge technologies into CPS introduces several Quality-of-Service (QoS) challenges, such as service latency, energy consumption, security, privacy, and reliability. To navigate these challenges, researchers have been focusing on CPS integration from a QoS optimization perspective. By identifying key challenges and reviewing existing solutions, the study categorizes them based on their similarities and differences and offers valuable insights for improving the performance of CPS using edge computing. As the integration of edge technologies continues to evolve, it holds promise for overcoming many of the current limitations faced by CPS in battery management, ultimately improving efficiency, scalability, and overall system reliability (Cao et al., 2021).

Comparative analysis of digital twin deployment strategies indicates a clear trade-off between model fidelity, latency, and computational feasibility. Cloud-only digital twins enable high-fidelity modeling and fleet-level analytics but suffer from communication latency and increased energy consumption. Edge-only implementations offer low-latency operation suitable for safety-critical functions but are constrained by hardware limitations. Hybrid edge-cloud architectures, as exemplified by DDT and C3-FLOW frameworks, provide a balanced solution by distributing computational tasks according to latency sensitivity and resource availability, enabling scalable and adaptive battery digital twin deployment. Table 6 shows comparative characteristics of cloud-only, edge-only, and hybrid digital twin architectures for battery management.

TABLE 6. Comparative characteristics of cloud-only, edge-only, and hybrid digital twin architectures for battery management.

8 Cyber-physical convergence in wireless communication

The evolving telecommunications landscape has transformed wireless communication into an interdisciplinary field influenced by technological, social, economic, and cognitive factors. This cyber-physical convergence extends beyond the IoT, enabling network devices to learn, think, and make autonomous decisions. AI strategies, big data analytics, and cognition are driving this shift, forming the cognitive internet of people, processes, data, and things. In this context, DTs and network digital twins (NDTs) emerge as key enablers of 6G networks, particularly through edge computing-based implementations and vehicle-to-edge (V2E) applications. A study examined how interdisciplinary research supports cyber-physical convergence, particularly through DT technology for efficient resource management in 6G networks. DTs act as virtual representations of physical assets, addressing digital transformation challenges while offering potential solutions. The study also explores the telecommunications landscape and AI market, justifying the need for NDTs. By integrating DTs within the cognitive internet of people, processes, data, and things, the research envisions autonomous network operations and the realization of digital transformation in 6G networks (Hlophe and Maharaj, 2023).

8.1 Edge-cloud collaboration for IoT performance optimization

A study also aims to alleviate network resource pressure on cloud centers and edge nodes, enhancing service quality and optimizing network performance through an edge-cloud collaboration framework for IoT applications. The framework utilizes Raspberry Pi (RP) devices as working nodes within a three-layer architecture: Edge RP, Monitoring & Scheduling RP, and CC. These layers enable effective collaborative communication between RPs and between RPs and CCs. A novel edge-cloud matching algorithm was introduced to minimize task time delays, particularly in time-sensitive applications. Experimental results from face recognition tasks demonstrate that the edge-cloud collaboration mode achieves the shortest delay (12 s) compared to edge-only and CC-only modes. Furthermore, the framework enhances real-time object detection while reducing energy consumption, lowering system energy usage from 18 W to 16 W. Additionally, resource utilization rates improve as the number of users increases. Despite these improvements, further research is needed to integrate lightweight virtualization technologies, such as Kubernetes and Docker, to further optimize the framework’s efficiency (Feng et al., 2023).

8.2 Digital twin-based approaches for energy cyber-physical systems

As the complexity of DERs and their communication and control requirements increase, a robust platform is needed to handle incoming data and ensure the reliable operation of power systems. DT technology has emerged as a promising solution for modeling energy cyber-physical systems (ECPSs), facilitating real-time monitoring, decision-making, and optimization.

8.3 Digital twin framework for energy cyber-physical systems

A DT-based approach was proposed for modeling ECPSs across various applications, introducing two types of DTs tailored for high-bandwidth and low-bandwidth applications that require centralized decision-making. The DT framework is validated using Amazon Web Services (AWS), integrating physical and data models while receiving real-time measurements from power and control entities. Experimental results demonstrate the feasibility of real-time DT implementation, achieving a normalized mean-square error of 3.7% for low-bandwidth applications and a voltage estimation accuracy of 98.2% for high-bandwidth applications. Furthermore, power system sources and interconnected microgrid clusters are modeled and implemented on AWS, confirming the effectiveness of DTs as a live digital replica for future power systems. Future research aims to integrate deep learning techniques with DTs to enhance the stability, reliability, and resilience of globally interconnected power systems (Saad et al., 2020).

8.4 Cloud-edge collaboration in digital twin implementation

The advancement of edge computing has facilitated cloud-edge collaboration, enabling AI models to operate efficiently in cyber-physical environments. A DDT approach was proposed for real-time battery system monitoring and control in electric vehicles. This framework uses an online adaptive model reduction technique to address application time sensitivity and infrastructure limitations. By employing a GRU neural network, the framework estimates battery internal states from IoT sensor data, while incremental learning updates a reduced-order model (ROM) based on streaming data. The integration of the Lyapunov stability theorem enhances synchronization with actual battery behavior. Simulation results show an average ROM prediction error of 1.70%, with a 43.3% improvement in accuracy using a physics-informed adaptive approach. Cloud-edge collaboration allows high-fidelity digital twins to reside in the cloud, while optimized ROMs operate on edge devices to enhance battery control efficiency (Xie et al., 2024).

8.5 Digital twin for low-carbon electrical equipment management

Real-time management of electrical equipment, including renewable energy sources, controllable loads, and storage units, is essential for achieving low-carbon operations in smart industrial parks. DT technology, which integrates cloud-edge-device collaboration with artificial intelligence, has been utilized to optimize these systems. However, challenges such as electromagnetic interference, high communication costs, and inefficient resource allocation limit its effectiveness. To overcome these issues, a cloud-edge-device collaborative framework, C3-FLOW, has been introduced to improve DT reliability and communication efficiency. By optimizing device scheduling, channel allocation, and computational resource distribution, C3-FLOW reduces long-term system losses and communication costs. Simulation results show that it outperforms existing methods, achieving a 49.28% reduction in loss functions and a 14.71% decrease in communication costs. Furthermore, in V2G energy scheduling, C3-FLOW successfully cuts peak-period carbon emissions by 36.78% and boosts renewable energy absorption during off-peak hours by 69.78% (Liao et al., 2023).

8.6 Coordination of distributed energy resources using digital twins

Distributed energy resources are essential for frequency regulation in renewable-integrated power systems. However, uncoordinated DER responses can lead to inefficiencies, such as delayed reactions and significant overshoots. Traditional centralized and distributed control strategies depend heavily on real-time communication, which can cause performance degradation due to latency. To address this, a DT-based approach for DER coordination has been developed, minimizing reliance on continuous communication while ensuring effective system control. The study presents two DT-based control frameworks: one using cloud-hosted DTs for centralized coordination and another utilizing edge-hosted DTs for distributed control. Real-time simulation tests demonstrate that both frameworks improve DER aggregation and enhance grid support during contingency events by accurately estimating DER status and enabling coordinated actions without excessive communication dependency (Han et al., 2023). Table 7 summarizes advancements in digital twin applications for cyber-physical systems and energy management.

TABLE 7. Advancements in digital twin applications for cyber-physical systems and energy management.

8.7 Cybersecurity in battery digital twins

The increasing adoption of digital twin architectures in battery management systems introduces new cybersecurity challenges arising from enhanced connectivity, distributed computation, and reliance on cloud-based analytics. Battery digital twins require continuous bidirectional data exchange between physical assets and virtual models through in-vehicle communication buses, IoT gateways, and cloud–edge infrastructures, thereby expanding the attack surface compared to conventional battery management systems. Potential vulnerabilities include unauthorized access to sensor data, manipulation of state estimation outputs, disruption of communication channels, and malicious model updates, all of which may compromise operational safety in battery systems (Humayed et al., 2017).

From a threat modeling perspective, cyber-attacks targeting battery digital twins can impact data confidentiality, integrity, and availability. In safety-critical applications, such attacks may propagate through the digital twin feedback loop, leading to incorrect control decisions, improper thermal management, or accelerated degradation. Consequently, cybersecurity must be treated as an integral design requirement rather than a post-deployment consideration in DT-enabled battery systems (Achuthan et al., 2025).

Several countermeasures have been proposed to mitigate these risks, including secure edge computing architectures, encrypted communication protocols, authentication mechanisms, and anomaly detection techniques for identifying abnormal data or control behavior. Distributed ledger and blockchain-based approaches have also been explored to enhance data integrity and traceability in collaborative or multi-stakeholder digital twin environments, although their computational overhead and scalability remain active research challenges (Otoom and Almaiah, 2025).

In parallel, compliance with established cybersecurity standards is essential for enabling the safe deployment of battery digital twins in commercial systems. Standards such as ISO 21434 and SAE J3061 (Ward and Wooderson, 2021) provide structured guidelines for managing cybersecurity risks across the vehicle lifecycle and offer a foundation for integrating security considerations into digital twin design, validation, and operation. Aligning battery digital twin architectures with these standards will be critical for achieving regulatory acceptance and ensuring long-term operational resilience.

9 Conclusion and future research directions

The manuscript highlights the integrated operation of digital twin technologies with edge computing, deep learning, and artificial intelligence frameworks to enable advanced lithium-ion battery management systems. DT-based approaches facilitate real-time health assessment, condition monitoring, and improved RUL prediction by continuously linking physical battery behavior with virtual representations. However, existing studies also reveal several persistent challenges, including cybersecurity risks, standardization complexity, and the computational burden associated with on-board implementation.

To address these limitations, continued research into federated learning strategies, adaptive modeling techniques, and mixed-reality applications is essential for scalable and privacy-preserving deployment of DT-enabled BMS architectures. In parallel, responsible artificial intelligence (RAI) methodologies combined with cloud-edge collaboration are increasingly recognized as critical enablers for improving the robustness, transparency, and operational reliability of battery systems used in electric vehicles and distributed energy resources.

While the majority of existing battery digital twin research focuses on lithium-ion systems due to their technological maturity and widespread deployment, the underlying DT principles are broadly applicable to emerging battery chemistries. Future research should extend DT frameworks to chemistries such as lithium-sulfur, lithium-oxygen, and solid-state batteries, accounting for their distinct degradation mechanisms, including polysulfide shuttling, interfacial instability, and solid electrolyte degradation. Adapting digital twins to these chemistries will require chemistry-aware modeling, tailored sensing strategies, and revised validation protocols to ensure accurate and reliable operation.

Despite significant progress, a fundamental research gap remains in the development of fully reliable virtual battery models capable of real-time evolution across diverse operating conditions. Most existing digital twins are tailored to specific datasets, operating regimes, or aging stages, which limits their robustness when deployed in dynamic, real-world environments. Achieving continuously evolving digital twins requires systematic integration of online learning, adaptive model updating, and multi-timescale data assimilation to ensure sustained synchronization between physical batteries and their virtual counterparts throughout the lifecycle.

Equally important is the development of automated verification and validation mechanisms that can quantify digital twin confidence, detect model drift, and ensure safe operation in safety-critical applications. Future research should therefore prioritize standardized validation protocols, uncertainty-aware state estimation, and hybrid physics-data-driven frameworks capable of maintaining consistent performance under varying thermal, electrical, and degradation conditions. Addressing these challenges is essential for transitioning battery digital twins from experimental and decision-support tools into trustworthy, deployable components of next-generation battery management 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.

Author disclaimer

Its content only reflects the author’s view and that the European Commission is not responsible for any use that may be made of the information it contains.

Glossary

AI Artificial Intelligence ANN Artificial Neural Networks AAS Asset Administration Shell AWS Amazon Web Services BPNN Back-Propagation Neural Networks BDTM Battery Digital Twin Model BESS Battery Energy Storage System BMS Battery Management System BTMS Battery Thermal Management System BO Bayesian Optimization CAN Controller Area Network CBMS Cloud-Based Battery Management System CFD Computational Fluid Dynamics CNN Convolutional Neural Network CPPS Cyber-Physical Production System CPS Cyber-Physical System DDPG Deep Deterministic Policy Gradient DNN Deep Neural Network DPT Digital Product Twin DT Digital Twin DT-BMMS Digital Twin-Based Battery Monitoring and Management System DTMAR Digital Twin Structural-Reliability Assessment Framework DEM Discrete Element Method DFN Doyle-Fuller-Newman DDT Dual Digital Twin EoT Edge Of Things EOL End of Life ECPS Energy Cyber-Physical System EWN Ensemble Weighting Network ESR Equivalent Series Resistance XAI Explainable Artificial Intelligence EKF Extended Kalman Filter ET Extra Trees XGBoost Extreme Gradient Boost FEM Finite Element Method GRU Gated Recurrent Unit GA Genetic Algorithm GBDT Gradient Boosting Decision Tree HI Health Indicator HRCD Human-Robot Collaboration Disassembly ICEEMDAN Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise IGWO Improved Gray Wolf Optimization IIot Industrial Internet of Things Iot Internet Of Things KPI Key Performance Indicator KNN K-Nearest Neighbors LightGBM Light Gradient Boosting Machine LCO Lithium Cobalt Oxide LIB Lithium-Ion Battery LOB Lithium-Oxygen Battery LSB Lithium-Sulfur Battery LIME Local Interpretable Model-Agnostic Explanation LSTM Long Short-Term Memory ML Machine Learning MAE Mean Absolute Error MAPE Mean Absolute Percentage Error MGDT Microgrid Digital Twin MPC Model Predictive Control NN-RSR Neural Network Response-Surface Model NDT Network Digital Twin NMC Nickel-Manganese-Cobalt NRMSE Normalized Root Mean Square Error OBD On-Board Diagnostics OCV Open-Circuit Voltage PSO Particle Swarm Optimization P2D Pseudo-Two-Dimensional QoS Quality Of Service RF Random Forest RP Raspberry Pi RBS Reconfigurable Battery System ROM Reduced-Order Model RNN Recurrent Neural Network RUL Remaining Useful Life ROI Return-On-Investment RMSE Root Mean Square Error SHAP Shapley Additive Explanation SEI Solid Electrolyte Interphase Soc State Of Charge Soh State Of Health SVR Support Vector Regression TCN Temporal Convolutional Network TS-GAN Time-Series Generative Adversarial Network V2G Vehicle To Grid V2E Vehicle To Edge WOA Whale Optimization Algorithm

Footnotes

Edited by: Jung Ho Kim, University of Wollongong, Australia

Reviewed by: Xinhua Liu, Beihang University, China

Christopher Columbus C., VIT University Chennai, India