The advent of li ion battery technology, with its superior energy density and lightweight characteristics, has been a cornerstone in the rapid evolution of modern society. It powers the revolution in electric vehicles, enables large-scale energy storage for renewable integration, and is indispensable in portable electronics, aerospace, and robotics. The operational lifespan of a li ion battery is finite and degrades over time due to complex electrochemical processes. Predicting its Remaining Useful Life (RUL) is therefore paramount for ensuring safety, reliability, and economic efficiency across all applications. RUL is typically defined by the number of charge-discharge cycles before the capacity fades to 80-70% of its initial value or the internal resistance increases to 160%. Accurate RUL prediction forms the basis for proactive maintenance, timely replacement, and effective planning for secondary use or recycling. To systematically delineate the intellectual landscape and evolutionary trajectory of this critical research field, this article employs a bibliometric analysis using CiteSpace software. By visualizing and analyzing literature from the CNKI and Web of Science (WOS) databases spanning 2000-2024, this study aims to uncover foundational knowledge structures, collaborative networks, research fronts, and emerging trends in li ion battery lifespan prognostics.
The methodology centers on CiteSpace, a Java-based application designed for visualizing and analyzing trends and patterns in scientific literature. It transforms bibliographic data into knowledge maps, revealing networks of authors, institutions, keywords, and their co-citations. The analysis combines these visual metrics with quantitative assessments. The data was sourced from two major repositories: 758 relevant articles were retrieved from the CNKI database (Chinese core journals) and 1,836 from the WOS core collection (English journals), covering publications from January 1, 2000, to February 1, 2024. After deduplication and format conversion, these 2,594 articles constituted the dataset for constructing knowledge maps on author/institution collaboration, keyword co-occurrence, and citation bursts, providing a dual-perspective (Chinese and international) on the field’s development.
Collaborative Landscape: Authors and Institutions
The analysis of author collaboration networks reveals distinct patterns. In the WOS dataset, the network comprised 198 authors connected by 235 collaboration links, indicating a relatively dense and interconnected research community. A significant proportion (approximately 75.8%) of authors published at least two papers. The CNKI network was larger in node count (335 authors) but with a slightly lower link density (319 links), suggesting a broader but potentially less tightly knit community. Applying Price’s Law to identify core authors, where the minimum publication count (M) is defined as $$M = 0.749 \times \sqrt{N_{max}}$$ (with $N_{max}$ being the highest publication count by a single author), yields interesting insights. For the WOS corpus ($N_{max}=22$), $M \approx 3.5$, setting a core author threshold of 4 publications. The subset of authors meeting this criterion accounted for only about 34% of total publications, below the 50% threshold that signifies a mature, stable core author group. The CNKI corpus ($N_{max}=6$, $M \approx 1.8$, threshold of 2 publications) showed a similar pattern, with core authors contributing only about 30% of the publications. This indicates that the field of li ion battery life prediction, while active, is still in a growth phase where influential, highly prolific author teams have not yet fully consolidated.
Examining the institutional affiliation of prolific authors highlights the profoundly interdisciplinary nature of this research. The field is driven by convergence between several classical engineering and science disciplines, as summarized in the table below:
| Academic Discipline | Primary Research Focus in Li-ion Battery Life Prediction |
|---|---|
| Chemical Engineering & Materials Science | Investigation of degradation mechanisms at the electrode-electrolyte interface, development of novel anode/cathode materials, and synthesis of stable electrolytes to intrinsically extend cycle life. |
| Electrical Engineering | Design of Battery Management Systems (BMS), development of algorithms for State of Charge (SOC) and State of Health (SOH) estimation, and system-level energy management for packs and grids. |
| Mechanical & Automotive Engineering | Integration of li ion battery packs into vehicles and drones, thermal management system design, structural analysis for vibration and impact resistance, and RUL prediction for maintenance scheduling. |
| Computer Science & Data Science | Development of data-driven prognostic models, application of machine learning (ML) and deep learning (DL) algorithms for feature extraction and RUL prediction, and big data analytics from battery cycling. |
This cross-disciplinary fusion is essential for a holistic approach to li ion battery lifespan management, addressing challenges from the atomic scale of material degradation to the system-scale of operational control.
Mapping the Intellectual Core: Research Hotspots and Evolution
Keyword Co-occurrence and Cluster Analysis
The co-occurrence network of keywords crystallizes the central themes of the field. High-frequency keywords such as “state of health (SOH)”, “remaining useful life (RUL) prediction”, “machine learning”, “deep learning”, “capacity fade”, “aging”, and “Battery Management System (BMS)” form the dense core of the network in both databases. Cluster analysis, performed using the Log-Likelihood Ratio (LLR) algorithm, groups these keywords into coherent research themes. The high modularity Q values (WOS: Q=0.7984; CNKI: Q=0.6855, both >0.3) and high mean silhouette S values (WOS: S=0.9502; CNKI: S=0.9271, both >0.7) confirm the significant structure and reliability of the clustering. The clusters can be broadly synthesized into two overarching methodological paradigms:
- Model-Based Prognostics: This paradigm relies on constructing physical or empirical models of the li ion battery.
- Electrochemical Models: Use coupled partial differential equations (e.g., Doyle-Fuller-Newman model) to describe ion transport and reaction kinetics. Aging is modeled by incorporating side reactions. The RUL is extrapolated from simulated parameter drift.
$$ \frac{\partial c_s}{\partial t} = \frac{1}{r^2} \frac{\partial}{\partial r} \left( D_s r^2 \frac{\partial c_s}{\partial r} \right) $$ - Equivalent Circuit Models (ECM): Represent the li ion battery with electrical components (resistors, capacitors, voltage sources). Parameters of the ECM (e.g., ohmic resistance $R_0$, polarization resistance $R_p$) are identified online and their evolution is tracked to indicate health.
$$ V_{terminal} = OCV(SOC) – I \cdot R_0 – V_p $$ - Empirical/Semi-Empirical Models: Use mathematical functions fitted to aging data, such as exponential or polynomial models of capacity fade.
$$ Q_{loss} = A \cdot \exp(B \cdot N) + C \cdot N $$
where $Q_{loss}$ is capacity loss, $N$ is cycle count, and $A, B, C$ are fitting parameters.
- Electrochemical Models: Use coupled partial differential equations (e.g., Doyle-Fuller-Newman model) to describe ion transport and reaction kinetics. Aging is modeled by incorporating side reactions. The RUL is extrapolated from simulated parameter drift.
- Data-Driven & AI-Based Prognostics: This paradigm dominates current research, leveraging algorithms to learn degradation patterns directly from operational data.
- Machine Learning: Includes Support Vector Machines (SVM), Relevance Vector Machines (RVM), and Gaussian Process Regression (GPR) for modeling the nonlinear relationship between health indicators and RUL.
- Deep Learning: Employs advanced neural networks like Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), and Convolutional Neural Networks (CNN) to automatically extract temporal features from sequential battery data (voltage, current, temperature) for sequence-to-sequence RUL prediction.
$$ h_t, c_t = \text{LSTM}(x_t, h_{t-1}, c_{t-1}; \theta) $$ - Filtering Algorithms: Particle Filter (PF) and its variants (Unscented PF, Optimized PF) are widely used to combine empirical models with real-time measurement data in a Bayesian framework for probabilistic RUL estimation.
$$ p(x_k | z_{1:k}) \propto p(z_k | x_k) \int p(x_k | x_{k-1}) p(x_{k-1} | z_{1:k-1}) dx_{k-1} $$
A powerful emerging trend is hybrid modeling, which integrates the interpretability of physical/empirical models with the adaptive learning power of AI algorithms, creating more robust and generalizable prognostic frameworks for the li ion battery.
Temporal Evolution and Burst Detection
Keyword burst detection identifies terms that experience a sharp increase in citation frequency over a specific period, marking the emergence of research frontiers. The evolution of the field can be segmented into distinct phases, as illustrated by the burst keywords from the WOS and CNKI analyses:
| Phase | Time Period | Burst Keywords & Concepts | Significance & Direction |
|---|---|---|---|
| Foundational Phase | ~2006-2012 | Cycle life, Capacity fade, Aging mechanism | Focus on fundamental electrochemical degradation studies, establishing baseline understanding of li ion battery failure modes. |
| Growth & System Integration Phase | ~2013-2019 | State of Charge (SOC), Battery Management System (BMS), Electric vehicles, Energy management | Shift towards operational state estimation and system-level control. Research driven by the automotive industry’s needs for reliable BMS algorithms and pack longevity. |
| Data-Driven & AI Dominance Phase | ~2020-Present | Machine learning, Deep learning, Feature extraction, Data-driven, Convolutional Neural Network (CNN), Attention mechanism | Explosive growth in AI/ML applications. Emphasis on automated health feature learning from raw or processed cycling data (e.g., incremental capacity analysis – ICA, differential voltage analysis – DVA) for more accurate and adaptive RUL prediction. |
The current frontier is distinctly characterized by sophisticated AI architectures. Researchers are moving beyond standard LSTMs to employ bidirectional networks, encoder-decoder structures with attention mechanisms, and hybrid models that combine CNNs for spatial feature extraction from voltage/current curves with RNNs for temporal dependency modeling. Furthermore, transfer learning is gaining traction to address the critical challenge of predicting li ion battery RUL with limited early-cycle data. The fusion of model-based and data-driven approaches represents the cutting edge, aiming to create prognostics that are both physically meaningful and highly accurate.
Conclusion and Prospective View
This bibliometric analysis via CiteSpace provides a panoramic view of the dynamic research landscape in li ion battery remaining useful life prediction. The field exhibits vibrant, decentralized collaboration networks without a singular dominant core group, reflecting its broad appeal and distributed innovation. Its strength lies in its inherent interdisciplinary, drawing concerted efforts from materials science, electrochemistry, electrical engineering, mechanical engineering, and computer science.
The intellectual evolution has progressed from foundational studies of degradation mechanisms to the current era dominated by artificial intelligence and big data analytics. The core research hotspots are unequivocally centered on data-driven methods, particularly deep learning algorithms like LSTMs and CNNs, and their integration with physics-based models for hybrid prognostics. Key technical challenges driving future research include:
- Generalization and Transferability: Developing models that can accurately predict RUL for a li ion battery under varying usage conditions, chemistries, and with minimal early-life data.
- Uncertainty Quantification: Moving from point estimates to probabilistic RUL predictions with confidence bounds, which is crucial for risk-informed decision-making.
- Explainable AI (XAI): Interpreting the decisions of complex “black-box” deep learning models to build trust and link predictions back to physical degradation modes.
- Edge Computing for BMS: Deploying lightweight yet accurate prognostic models on embedded BMS hardware for real-time, onboard RUL estimation.
In conclusion, the pursuit of accurate li ion battery lifespan prediction is more critical than ever as our reliance on this technology deepens. The field is poised for continued advancement through the deepening synergy between electrochemical theory, advanced sensor data, and next-generation artificial intelligence, ultimately ensuring the safe, efficient, and sustainable operation of energy storage systems across the globe.