Abstract
Electrochemical energy storage technologies, represented by lithium-ion batteries, play a pivotal role in supporting new energy vehicles and novel power systems. However, frequent safety accidents in lithium-ion battery power and energy storage systems pose significant challenges, primarily due to thermal runaway triggered by safety failures during operation. This can lead to disastrous consequences such as fires or explosions. Therefore, timely and accurate active safety warnings are crucial before battery failure to prevent thermal runaway and ensure safe operation of energy storage battery systems. This paper systematically reviews four types of early active safety warning methods for lithium-ion batteries based on signal characteristics, model prediction, data-driven, and hybrid strategies. The theoretical basis and implementation paths of each method are introduced, along with an analysis of their advantages and disadvantages. Finally, the key issues remaining in the field of lithium-ion battery active safety warnings are summarized, and potential solutions are proposed, providing valuable references for future research.
1. Introduction
In the face of declining fossil fuel resources and the urgent need to mitigate global climate change, renewable energy sources have become increasingly important. China, as a major energy consumer, has embarked on a path to achieve carbon peaking and carbon neutrality by 2060, emphasizing the development of new energy systems . Renewable energy sources like solar and wind power are intermittent, posing challenges to grid stability when integrated into the power grid. Energy storage systems, particularly lithium-ion batteries, offer a solution by balancing power supply and demand, enhancing grid flexibility .
Despite their advantages, lithium-ion batteries are prone to thermal runaway, leading to safety incidents. Thermal runaway, often triggered by internal short circuits caused by overheating, overcharging, or mechanical abuse, can escalate into catastrophic failures . To address this, active safety warning technologies that can predict and prevent thermal runaway before it occurs are crucial. This paper reviews current research in this field, focusing on four key methods and their applications.
2. Early Warning Methods Based on Signal Characteristics
Early warning methods based on signal characteristics rely on sensors to capture physical signals such as voltage, current, temperature, and gas emissions. These signals are then processed using algorithms to identify anomalies indicative of battery failure.
2.1 Voltage and Current Signals
Voltage and current are commonly monitored parameters in battery management systems (BMS). However, they often show significant deviations only in the final stages of battery failure, limiting their usefulness for early warning . Zhang et al. analyzed voltage, gas emissions, and temperature during overcharging and proposed a three-stage warning strategy based on voltage inflection points and temperature thresholds.
2.2 Temperature Signals
Temperature sensors placed on battery surfaces can detect overheating, but internal temperatures can differ significantly due to battery size and design. Shi et al. compared different gas detectors for overcharge safety warnings and found hydrogen detectors to be the most effective, detecting issues earlier than thermal or smoke detectors.
2.3 Gas Emissions
Gas emissions, particularly hydrogen, have been shown to be early indicators of thermal runaway. Jin et al. used hydrogen sensors to detect overcharge-induced lithium dendrite formation, achieving early warnings 639 seconds before smoke detection and 769 seconds before fire.
2.4 Strain and Force Signals
Battery expansion due to internal reactions can also serve as an early warning signal. KOCH et al. compared various sensor responses and found strain signals to be the most sensitive, indicating potential failures well before thermal runaway.
Table 1: Summary of Signal-Based Warning Methods
| Method | Parameter Monitored | Advantages | Disadvantages |
|---|---|---|---|
| Voltage/Current | Voltage, Current | Widely available in BMS, easy to implement | Late indication of failure, high sensitivity to threshold settings |
| Temperature | Surface Temperature | Direct indicator of overheating | Difficulty in detecting internal temperatures, delayed response |
| Gas Emissions | Hydrogen, CO, VOCs, etc. | Early indication of internal reactions | Requirement for specialized gas detectors, potential for false alarms |
| Strain/Force | Expansion, Pressure | Sensitive to early failure signs | Complex installation, high cost, limited application scenarios |
3. Early Warning Methods Based on Model Prediction
Model-based prediction methods leverage theoretical models of battery behavior to forecast potential failures. These models include equivalent circuit models (ECMs), electrochemical models, and mathematical models.
3.1 Equivalent Circuit Models (ECMs)
ECMs simplify battery behavior using electrical components like resistors and capacitors. Common ECMs include the Rint model, Thevenin model, and PNGV model. While computationally efficient, they lack detailed chemical insights.
3.2 Electrochemical Models
Electrochemical models describe battery behavior at the molecular level, capturing phenomena like lithium-ion insertion/deinsertion and solid electrolyte interphase (SEI) formation. These models are more accurate but computationally intensive.
3.3 Mathematical Models
Mathematical models employ statistical and machine learning techniques to establish relationships between battery parameters. They can be data-driven or based on physical principles.
3.4 Internal Short Circuit Detection
Internal short circuits are a major cause of thermal runaway. Model-based methods detect these by analyzing electrical and thermal signatures. However, most models focus on a single failure mode and may not generalize well.
Table 2: Summary of Model-Based Warning Methods
| Method | Advantages | Disadvantages |
|---|---|---|
| ECMs | Simple, computationally efficient | Limited by lack of chemical insights |
| Electrochemical Models | Accurate, detailed chemical insights | Computationally intensive, complex implementation |
| Mathematical Models | Flexible, can incorporate multiple data sources | May lack physical interpretability, sensitive to data quality |
| Internal Short Circuit Detection | Can detect early signs of failure | Limited to specific failure modes, model parameter convergence issues |
4. Data-Driven Early Warning Methods
Data-driven methods leverage large datasets and statistical/machine learning techniques to identify patterns indicative of battery failure.
4.1 Statistical Methods
Statistical methods use historical data to establish normal operating ranges and detect deviations. WANG et al. proposed a method using Shannon entropy to detect voltage faults.
4.2 Machine Learning Methods
Machine learning algorithms excel at feature extraction and pattern recognition. JIANG et al. used an isolated forest algorithm to detect internal short circuits, while WAN et al. employed LSTM networks for voltage-based fault detection.
Table 3: Summary of Data-Driven Warning Methods
| Method | Advantages | Disadvantages |
|---|---|---|
| Statistical Methods | Simple, easy to implement | Limited by assumptions of data distribution, may miss complex patterns |
| Machine Learning | Powerful feature extraction, good generalizability | Require large, high-quality datasets, may lack interpretability |
5. Hybrid Early Warning Strategies
Hybrid strategies combine multiple methods to enhance performance and reliability. They can improve model accuracy, expand application scenarios, or reduce false alarms.
5.1 Model Enhancement
Combining models can improve accuracy. ZHENG et al. used a combination of RVR and UKF to predict battery capacity, enhancing prediction accuracy.
5.2 Scenario Adaptation
Hybrid methods can adapt to different operating conditions. ZHANG et al. trained separate BPNNs for different driving conditions to predict battery voltage accurately.
5.3 Multi-level Warning Systems
Multi-level warning systems reduce false alarms. By setting multiple warning levels and corresponding actions, these systems balance sensitivity and specificity .
Table 4: Summary of Hybrid Warning Strategies
| Strategy | Advantages | Disadvantages |
|---|---|---|
| Model Enhancement | Improves accuracy, robustness | Higher computational complexity |
| Scenario Adaptation | Expands applicability, improves performance | Requires extensive training data |
| Multi-level Warnings | Reduces false alarms, increases reliability | May lead to complexity in deployment |
6. Comparison and Application Scenarios
Table 5: Comparison of Early Warning Methods
| Method | Signal-Based | Model-Based | Data-Driven | Hybrid |
|---|---|---|---|---|
| Timeliness | Moderate | Good | Good | Best |
| Accuracy | Moderate | High | High | Highest |
| Generalizability | Limited | Moderate | Good | Best |
| Data Dependency | Low | Moderate | High | High |
| Computational Cost | Low | Moderate | High | Highest |
6.1 Signal-Based Methods
Suitable for early R&D phases and small-scale systems due to cost and sensor reliability concerns. Ideal for environments with controlled temperatures.
6.2 Model-Based Methods
Ideal for cell-level research and small-scale internal short circuit monitoring. Limited by parameter convergence issues and computational costs.
6.3 Data-Driven Methods
Well-suited for large-scale systems with extensive data collection capabilities. Require high-quality datasets and may struggle with uncommon operating conditions.
6.4 Hybrid Methods
Optimal for high-reliability, mission-critical applications. Address limitations of individual methods but require substantial computational resources and careful deployment planning.
7. Conclusion and Future Directions
This paper systematically reviewed early active safety warning technologies for lithium-ion batteries in energy storage systems. Key findings and future directions include:
- Physical Layer: Develop intelligent batteries with built-in multi-sensor fusion for accurate internal state monitoring.
- Data Layer: Research data transmission and storage solutions, enhance AI-based prediction models, and improve dataset quality and updating strategies.
- System Layer: Analyze warning methods across diverse application scenarios, providing customized design guidelines for the power grid, home energy storage, and electric vehicles.
By addressing these areas, future research can significantly enhance the safety and reliability of lithium-ion battery energy storage systems, supporting the transition to sustainable energy systems.