Beyond standard battery testing: a physics-informed framework for validating next-generation battery packs

1 Introduction

A battery or battery pack may successfully pass all standard factory tests before shipment, but there is still a certain gap between that and real-world operating conditions (Schreiber et al., 2026). This apparent paradox reflects a fundamental challenge facing modern battery research: while battery technologies have advanced rapidly, the conceptual basis of battery validation has remained largely unchanged. (Schreiber et al., 2026; Fan Y. et al., 2025) Standardized testing protocols continue to provide essential benchmarks for reproducibility, quality assurance, and regulatory compliance, yet they are often interpreted as evidence of meaningful validation (Weltsch et al., 2025; ELJarray et al., 2026). In fact, passing a testing procedure does not necessarily prove that the battery has been evaluated under conditions close to its actual operating conditions.

This gap is becoming increasingly evident as batteries evolve from individual laboratory cells to highly integrated battery packs deployed in electric vehicles, grid-scale energy storage, aviation, and other demanding applications. These systems are subjected to mixed real-world impacts such as climatic and environmental conditions, thermal, electrochemical, and mechanical loads; these mixed interactions occur across multiple spatial and temporal scales, creating complex degradation pathways that conventional testing strategies rarely capture. Although numerous studies have introduced multi-stress testing and advanced diagnostic techniques, (Fan Y. et al., 2025; Rücker et al., 2024; Liu et al., 2025; Zhang C. et al., 2025) the field still lacks a unified conceptual framework that defines what battery validation should represent and how validation protocols should be designed to reflect real operating physics.

In this perspective, the battery validation is suggested to move beyond protocol compliance toward physics-informed validation, a mechanism-oriented philosophy in which validation strategies are derived from the dominant degradation processes expected during service. Rather than treating temperature, current, pressure, vibration, or humidity as isolated testing variables, physics-informed validation considers their interactions through the underlying mechanisms governing battery degradation. Building on this concept, a Validation Maturity Framework is proposed that organizes validation strategies from conventional electrochemical testing to mechanism-informed, physics-informed, and digital twin-assisted validation. This framework aims to provide a common conceptual language for designing more physically meaningful validation strategies and facilitate the reliable translation of next-generation battery technologies from laboratory demonstrations to practical deployment.

2 The growing gap between standardized testing and real-world operation

Standardized testing protocols have played a fundamental role in developing batteries by providing reproducible procedures for benchmarking electrochemical performance, comparing materials, and ensuring quality control (Dubar et al., 2020; Miao et al., 2019). However, these protocols are primarily designed to isolate individual variables under controlled laboratory conditions rather than reproduce the coupled physical environments. Thus, standardized testing and real validation should not be regarded as equivalent objectives (Figure 1).

In real-world applications, battery packs operate under multiple interacting physical loads (Rufino Júnior et al., 2024; Zhao et al., 2024; Li H. et al., 2025). High current operation generates heat; thermal gradients alter local reaction kinetics; mechanical constraints influence cell contact pressure; environmental conditions affect aging behavior; and cell-to-cell heterogeneity further amplifies these interactions (Schreiber et al., 2026; Edge et al., 2021; Zhao et al., 2025). Rather than acting independently, these factors evolve simultaneously through interconnected degradation mechanisms that ultimately determine battery safety, durability, and reliability. The physical environment of a battery is therefore inherently coupled rather than isolated. By contrast, conventional validation strategies frequently investigate temperature, current, pressure, vibration, or humidity as independent testing variables (Dubar et al., 2020; Shi et al., 2025). While such approaches are invaluable for understanding the influence of individual factors, they often overlook the mechanistic interactions that dominate battery degradation under practical operating conditions. As battery systems become increasingly integrated and application-specific, this disconnect between isolated laboratory testing and coupled operating physics continues to widen (Zhao et al., 2025; Shi et al., 2025). Therefore, the challenge is not simply to introduce additional testing conditions but to redesign validation strategies around the physical mechanisms governing battery degradation. This shift from testing isolated variables to validating coupled operating physics, which forms the conceptual foundation of the physics-informed framework proposed in this perspective. The existing standards already encompass a broad range of battery testing objectives, spanning cell-level performance and safety to pack-level reliability, abuse, vibration, and system-level assessment (Table 1). However, these requirements remain largely organized around specific test objectives and compliance needs, rather than a unified framework linking operating conditions, coupled physical loads, degradation mechanisms, and predictive capability. The need for more application-specific qualification becomes particularly evident in second-life battery deployment, where heterogeneous degradation histories, residual state of health, and diverse operating profiles complicate conventional qualification procedures. Recent work on second-life batteries has highlighted the importance of systematic characterization, continuous state monitoring, performance evaluation, and adaptive management strategies for reliable integration into electric-vehicle infrastructure (Zabihi et al., 2026).

TABLE 1. Representative standards and recommended practices for battery testing, safety, and qualification.

*Abbreviations: EV, electric vehicle; HEV, hybrid electric vehicle; PHEV, plug-in hybrid electric vehicle; RESS, rechargeable energy storage system; BESS, battery energy storage system; EES: electrical energy storage.

3 A new conceptual framework based on physics-informed validation

The widening gap between standardized tests and real-world operating conditions indicates that battery validation processes require more than just increasingly complex testing protocols. Rather than asking how many operating conditions should be included in the validation process, a more fundamental question is how to select these conditions. From this perspective, it is suggested that the answer should be based on the physical principles governing battery degradation.

In this perspective, physics-informed validation is proposed as a conceptual validation philosophy in which testing strategies are designed according to the dominant thermal, electrochemical, mechanical, and environmental degradation mechanisms expected during battery operation. This definition shifts the objective of validation from reproducing standardized protocols to reproducing the physical conditions responsible for battery aging, failure, and performance evolution. More important, physics-informed validation should not be confused with simply combining multiple stress factors within a single experiment. Conventional multi-stress testing typically investigates the effects of thermal, electrical, or mechanical abuse (Le et al., 2023). While highly valuable, these approaches can make it difficult to clearly link operating conditions and measurable performance changes to their underlying degradation mechanisms (Qu et al., 2022; Luo et al., 2023). Physics-informed validation instead begins by identifying the dominant degradation mechanisms associated with a target application, then designing validation strategies that reproduce the coupled physical environments responsible for those mechanisms (O’Kane et al., 2022). Based on this philosophy, four guiding principles for physics-informed validation are suggested in Figure 2: (i) define validation objectives according to the intended application rather than standardized protocols; (ii) identify the dominant degradation mechanisms governing battery performance; (iii) design testing strategies that reproduce coupled physical loads instead of isolated variables; and (iv) select validation metrics capable of reflecting the targeted degradation mechanisms rather than relying solely on conventional performance indicators. Together, these principles provide a conceptual framework for designing validation strategies that are both physically meaningful and application relevant. This framework changes the role of testing from reproducing predefined conditions to interrogating the physical causes of battery degradation. The key distinction is not coming from the number of stress factors included in a test, but whether the selected conditions can be causally connected to the degradation mechanisms relevant to the intended application. This shift also changes how validation metrics should be interpreted: capacity retention, resistance growth, thermal response, or mechanical deformation become more informative when they are linked to specific physical processes rather than treated solely as empirical performance indicators. Thus, physics-informed validation represents a battery surviving a prescribed test meaningfully reproducing the conditions and mechanisms that govern its behavior in operation.

Recent analysis of standardized and commercially deployed charging protocols has shown that charging profiles can be interpreted in relation to state-of-charge-dependent (SOC) material transformations, ionic transport limitations, reaction heterogeneity, and heat generation. For example, differences in current profiles across practical fast-charging strategies can produce distinct polarization and thermal responses, suggesting that charging protocols can be evaluated not only by charging time but also by the physical mechanisms they activate or suppress (Li et al., 2026). This mechanistic interpretation provides a direct pathway from observed operating conditions to the selection of more physically relevant validation conditions.

4 Validation framework

The transition toward physics-informed validation requires more than the incorporation of testing conditions. It requires a systematic way to evaluate how validation strategies evolve with increasing understanding of the mechanisms governing battery degradation. As summarized in Table 2, existing studies have progressively incorporated coupled stresses, degradation mechanisms, physical models, and predictive tools into battery evaluation. However, these developments remain distributed across different testing strategies and levels of physical representation. Building on these developments, the concept of validation maturity is proposed in Figure 3, defined as the extent to which a validation strategy reflects the physical mechanisms governing battery behavior under practical operating conditions. Thus, validation maturity represents a progression in physical understanding and predictive capability rather than merely an increase in experimental or computational complexity. Figure 3 presents this progression as a conceptual roadmap from protocol-oriented testing toward predictive validation.

TABLE 2. Evolution of battery validation strategies from standardized testing to predictive qualification.

In this context, battery validation strategies can be broadly categorized into six maturity levels. Level 1 focuses on protocol-oriented validation, where batteries are evaluated according to standardized testing procedures and predefined performance metrics (Dubar et al., 2020). Level 2 introduces parameter-oriented validation by investigating the influence of individual operating variables such as temperature or current (Qu et al., 2022). Level 3 advances to stress-oriented validation through the simultaneous application of multiple external stresses (Yan et al., 2018). Although these approaches provide increasingly comprehensive experimental data, they remain largely descriptive because validation is still organized around testing conditions rather than degradation mechanisms. A conceptual transition occurs at Level 4, where validation becomes mechanism-oriented. Instead of asking which testing conditions should be applied, validation begins by identifying the dominant degradation mechanisms expected under a given application and subsequently designing experiments to reproduce these mechanisms (Li R. et al., 2025). Level 5 extends this philosophy to physics-informed validation, where coupled thermal, electrochemical, mechanical, and environmental interactions are explicitly considered during validation design (Guo et al., 2023). Finally, Level 6 represents predictive validation, integrating digital twins, mechanistic models, advanced sensing, and data-driven approaches to continuously evaluate battery behavior throughout its operational lifetime (Sun et al., 2025).

An important distinction is made between mechanism-oriented validation (Level 4) and physics-informed validation (Level 5). At Level 4, validation focuses on identifying and reproducing the dominant degradation mechanisms associated with a specific failure or performance-loss pathway. Testing may therefore be designed to isolate or emphasize individual mechanisms, such as lithium plating, loss of active material, interfacial degradation, or mechanical damage. Level 5 extends this mechanistic understanding to the coupled physical environment in which multiple degradation pathways can interact during practical operation. Thermal, electrochemical, mechanical, and environmental conditions are therefore considered not only as individual stress factors but also through their interactions and their combined influence on degradation. The transition from Level 4 to Level 5 consequently represents a shift from understanding individual degradation pathways to reproducing the coupled operating physics that governs their evolution.

Figure 3 presents a conceptual Validation Maturity Framework to describe the progressive integration of physical understanding into battery validation, based on insights from existing standards and literature. At the lower levels, validation is primarily protocol- or parameter-oriented, focusing on predefined tests or the response of batteries to controlled changes in individual operating variables. Systematic aging studies have demonstrated how parameters such as temperature, state of charge, and depth of discharge influence degradation and performance evolution (Luo et al., 2023). Multi-stress approaches subsequently introduce coupled thermal, electrical, or mechanical conditions, with experimental studies demonstrating that such interactions can produce degradation behavior that cannot be inferred from isolated stresses alone (Yan et al., 2018; Wang F. et al., 2024). Mechanism-oriented validation moves one step further by explicitly linking observed performance changes to underlying degradation pathways (Li R. et al., 2025; Dia et al., 2022). At the highest levels proposed here, physics-informed validation uses such mechanistic understanding to design application-relevant testing, while predictive validation further integrates physical models, data-driven inference, and digital representations of battery behavior to anticipate degradation under untested or dynamic operating conditions (Guo et al., 2023; Qu et al., 2020; Jafari and Byun, 2022). Importantly, these maturity levels should not be interpreted as a universal ranking of validation quality. Here, they provide a conceptual framework for selecting validation strategies appropriate to different research objectives and technology readiness levels. Early-stage material discovery may require only protocol-oriented validation, whereas battery pack development for safety-critical applications increasingly demands physics-informed and predictive validation. The proposed framework therefore serves as a roadmap for aligning validation strategies with both the scientific understanding of degradation mechanisms and the intended application of battery technologies. An important implication of this framework is that validation maturity should not be equated with the number of variables, tests, or computational tools incorporated into an evaluation protocol. A highly complex multi-stress experiment may still provide limited mechanistic insight if the relationships between operating conditions, physical loads, and degradation pathways remain unresolved. In contrast, a more targeted validation strategy can provide greater physical relevance when the selected conditions are directly linked to the dominant failure mechanisms of the intended application. The proposed framework therefore shifts the emphasis from testing complexity toward physical observability, mechanistic attribution, and predictive capability.

5 From validation to predictive battery qualification

The proposed validation maturity framework also points toward a broader transformation in how battery technologies may be qualified for practical deployment. As batteries become increasingly integrated with sensors, battery management systems, and computational models, validation need not remain a discrete laboratory activity performed before deployment (Li et al., 2024; Zeng and Berecibar, 2025; Urishov et al., 2026). Here, it can evolve toward a continuous and predictive process in which physical measurements, mechanistic understanding, and operational data are combined throughout the battery lifetime.

One more key direction is the integration of physics-informed models and digital twins into battery validation (Borah et al., 2024; Telmasre et al., 2025). In spite of using models solely to reproduce experimental data, these approaches could continuously evaluate whether observed battery behavior remains consistent with the degradation mechanisms expected under actual operating conditions (Amiri et al., 2024; Plankovskyy et al., 2025; Keçeci and Oktal, 2026). This could enable validation to move from retrospective assessment toward early detection of abnormal degradation and prediction of future performance. Similarly, distributed temperature, pressure, strain, and electrochemical sensing could provide spatially resolved information that is currently lost when battery packs are evaluated primarily through aggregate metrics (Sulaima et al., 2026). However, the new systems should not rely too heavily on machine learning and complex computational protocols. The key challenge is to provide reliable links between measurable signals, physical mechanisms, and application-level failure modes (Thelen et al., 2022). Physics-informed AI may assist this process, but only when constrained by physically meaningful relationships and validated against independent operating conditions (Navidi et al., 2024; Chen et al., 2025; Eltamaly and Almutairi, 2026). In this context, future validation systems should combine mechanistic models, real-time sensing, data-driven inference, and application-specific operating profiles rather than treating any single approach as sufficient.

In the short term, the most practical opportunity is to extend existing qualification protocols instead of replace them. Application-specific operating profiles, coupled thermal–electrical testing, improved spatial sensing, and mechanism-oriented diagnostic metrics could be incorporated into established testing procedures without fundamentally changing current certification structures. In the mid-term, validation is likely to become increasingly model-assisted, with mechanistic models, physics-informed machine learning, digital twins, and real-time sensing used to connect measured signals with internal degradation states. For the long term, battery qualification could evolve from a discrete pre-deployment activity toward continuous and predictive assessment throughout the battery lifetime. Connected battery systems could combine real-time sensing, mechanistic models, fleet-level operational data, and digital representations of battery behavior to continuously evaluate whether individual cells, modules, or packs remain within their expected degradation trajectories. Qualification could consequently become adaptive, with validation requirements dynamically informed by battery history, operating environment, and emerging degradation signatures.

Machine learning (ML) can further extend this validation framework by converting multidimensional sensing and operational data into quantitative indicators of battery state and degradation (Shi et al., 2022). However, the usefulness of ML for validation depends not simply on prediction accuracy but on whether the selected metrics reflect physically meaningful degradation and remain reliable under operating conditions beyond the training dataset (Navidi et al., 2024; Shi et al., 2022). Metrics such as mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²) can quantify predictive performance, while errors in state-of-health (SOH), remaining useful life (RUL), or degradation-rate estimation provide more application-relevant measures of validation quality. Importantly, these metrics should be evaluated under independent operating conditions, dynamic load profiles, and different temperatures or aging states to assess model generalizability (Che et al., 2024). In this context, ML should function as a component of the validation framework rather than as a replacement for mechanistic understanding. Recent studies provide quantitative evidence that machine-learning-assisted approaches can substantially improve battery state estimation and degradation prediction, although their reported performance varies considerably with the dataset, operating conditions, prediction target, and validation strategy. Selected examples are summarized in Table 3 to illustrate not only reported prediction accuracy but also address dynamic or previously unseen operating conditions.

TABLE 3. Selected quantitative evidence from advanced battery characterization and machine-learning-assisted validation.

*Physics-informed long short-term memory network: LSTM; Model-Agnostic Meta-Learning: MAML; Gated Convolutional Neural Networks: GCNNs; Principal Component Analysis (PCA); Root-mean-squared error: RMSE; Mean absolute error: MAE; Principal Component Analysis: PCA; Fractional-order equivalent circuit model: FOECM; Random Forests: RF; Physics-informed hybrid neural network: PIHNN.

The reliability of predictive validation, however, depends on more than predictive accuracy alone. Model verification, uncertainty quantification, sensor accuracy, and the quality of the underlying operational data are essential for determining whether predictions remain physically meaningful. Predictive models should also be evaluated under operating conditions beyond those used for model development to assess their robustness and generalizability. These considerations are particularly important for battery systems, where changes in temperature, load profile, degradation state, and environmental conditions can move the system beyond the domain represented by the available training data. Thus, predictive validation should include not only model performance metrics but also verification of physical consistency, uncertainty, sensing reliability, and performance under previously untested operating conditions.

Recent literature increasingly addresses individual elements of this validation philosophy, including application-relevant aging profiles, coupled electrochemical–thermal–mechanical behavior, degradation-mode analysis, real-world battery data, and physics-informed predictive modeling. For example, realistic aging experiments on automotive-grade cells have demonstrated that accelerated laboratory cycling can reproduce warranty-equivalent mileage substantially faster while simultaneously altering the dominant degradation pathways, highlighting the limitations of accelerated tests when their physical equivalence to field operation is not established (Schreiber et al., 2025). Real-world fleet studies have further incorporated temperature, charging rate, and irregular-use effects into degradation prediction and reliability assessment (Li F. et al., 2025). Recent mechanistically guided learning approaches have also demonstrated improved state monitoring under dynamic and previously unseen operating conditions (Che et al., 2026). These developments indicate that important components of physics-informed validation are already emerging across different research directions, although they are often investigated separately. Table 4 compares representative studies according to the key dimensions considered in this perspective.

TABLE 4. Comparison of representative literature across key dimensions of physics-informed battery validation.

*✓ = Explicitly demonstrated.

△ = Partially addressed/indirect evidence.

✗ = Not demonstrated as a primary part of the study or not applicable.

In summary, the transition from standardized testing to physics-informed validation should not be viewed as a replacement of existing standards. Standardized protocols will remain essential for benchmarking, certification, and quality control. The opportunity lies instead in placing these protocols within a broader validation hierarchy in which the level of physical representation increases as a technology approaches practical deployment (Figure 4). Such a transition could transform battery qualification from a collection of predefined tests into a continuous, mechanism-aware, and predictive assessment of battery reliability under real-world conditions.

Translating the proposed physics-informed validation framework requires a careful transition from enhanced experimental testing toward model-assisted and ultimately predictive validation. In the short-term, existing qualification practices can be strengthened by incorporating application-relevant duty cycles, coupled physical stresses, distributed sensing, and mechanism-relevant diagnostics (Wang Z. et al., 2024; Liu et al., 2024; Yang et al., 2025). In the mid term, multiphysics models, physics-informed learning, uncertainty quantification, and self-adaptive digital twins can provide a bridge between experimental qualification and continuously updated representations of battery state and degradation (Fu et al., 2024; Nkechinyere Njoku et al., 2024; Zhang H. et al., 2025). For the long-term, continuous integration of operational data and adaptive models could enable battery validation to evolve from periodic qualification toward lifecycle-aware and predictive assessment (Fan Z. et al., 2025; Rasheed et al., 2026). A staged roadmap for this transition is summarized in Table 5.

TABLE 5. Future development pathway for physics-informed battery validation.

6 Conclusion

Battery validation is becoming increasingly challenging as battery systems evolve toward greater heterogeneity, dynamic operation, and application-specific requirements. Standardized testing will remain essential for reproducibility, benchmarking, and certification; however, compliance with predefined protocols alone may not fully capture the physical conditions governing battery degradation and failure in practical operation. This perspective therefore proposes a conceptual physics-informed validation framework that connects application conditions, coupled physical loads, degradation mechanisms, validation strategies, and mechanism-relevant metrics. The proposed six-level Validation Maturity Framework further organizes the progression from protocol-oriented testing toward stress-oriented, mechanism-oriented, physics-informed, and predictive validation, with increasing emphasis on physical understanding and predictive capability rather than experimental complexity alone. Recent studies have demonstrated promising predictive performance for machine-learning-assisted battery validation under real-world and physics-informed conditions, while real-world battery-pack studies have achieved strong predictive performance. These findings highlight the potential of data-driven approaches while also showing that predictive reliability depends strongly on operating conditions, available data, and physical information.

The framework proposed in this report is conceptual and requires systematic evaluation across battery chemistries, cell-to-pack architectures, operating environments, and failure modes before it can support standardized qualification practices. Future research should therefore focus on cross-condition validation, independently evaluated physics-informed models, distributed sensing, and integration of real-time operational data with digital representations of battery behavior. In the longer term, these developments could enable battery qualification to evolve from discrete laboratory testing toward continuous, mechanism-aware, and predictive assessment throughout the battery lifetime. Such a transition could improve the reliability and practical qualification of batteries deployed in electric transportation and stationary energy-storage systems, helping reduce the gap between laboratory performance and real-world energy-system reliability.

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.

Footnotes

Edited by: Mohammad Waseem, Jamia Millia Islamia, India

Reviewed by: Wenlong Li, The University of Texas at Austin, United States

Alireza Zabihi, University of Coimbra, Portugal