The safety and efficient operation of the battery energy storage system are paramount for its large-scale application. Among various safety hazards, the external short circuit (ESC) fault stands out due to its high incidence and severe consequences, such as deformation, water ingress, or wiring failures, which can lead to a direct connection between the battery’s positive and negative terminals. This fault triggers an extreme electro-thermal coupling abuse condition characterized by high-rate discharge, rapid internal heat accumulation, and a sharp temperature rise, posing significant risks to the battery energy storage system. My research delves into defining the safety boundaries of batteries under such abuse and developing predictive models to enhance monitoring and prevention strategies for the battery energy storage system.
Experimental Investigation of External Short Circuit Characteristics
To build a comprehensive understanding, I designed and conducted ESC tests on cylindrical lithium-ion batteries, typical in distributed battery energy storage system applications, under various initial conditions. The test platform included a control system, a short-circuit contactor within a vacuum chamber, a high-precision data acquisition unit, and an explosion-proof environmental chamber to ensure safety and control temperature. Key parameters varied were the initial State of Charge (SOC: 1.0, 0.8, 0.6, 0.4, 0.2), ambient temperature (-20°C, 0°C, 20°C, 40°C, 60°C), and external short-circuit resistance (10 mΩ, 20 mΩ, 30 mΩ).
The results reveal a consistent three-stage current response, as illustrated in the experimental data:
- Stage I (Instantaneous High-Rate Discharge): A massive current peak occurs immediately after short initiation (duration <1 s), primarily governed by double-layer capacitor discharge.
- Stage II (Current Plateau): The current rapidly drops to and maintains a relatively steady plateau. Simultaneously, the terminal voltage also stabilizes at a very low level. This stage is often attributed to the high temperature causing separator shutdown, which hinders lithium-ion transport.
- Stage III (Failure/Termination): The discharge process terminates abruptly, with both current and voltage plummeting to zero. This is typically caused by the melting of the positive tab or the activation of an internal current interrupt device (CID), preventing further discharge and potential thermal runaway.
The initial conditions profoundly affect the ESC characteristics. For instance, at a fixed resistance and ambient temperature, a higher initial SOC leads to a higher peak current and a more rapid temperature rise. Conversely, lower ambient temperatures initially suppress the peak current due to increased electrolyte viscosity, but as internal heating progresses, the current may rise. A smaller short-circuit resistance results in a higher peak current, a lower voltage plateau, and a faster temperature rise rate. The time from short-circuit initiation to the abrupt drop of current and voltage to zero is defined as the Failure Boundary (T_f). This boundary varies significantly with initial conditions; for example, it can be as short as ~11 seconds at high SOC and temperature, and as long as ~140 seconds at low SOC and temperature.
Determining the Damage Boundary
While prolonged ESC leads to failure, shorter-duration ESC can cause recoverable yet significant damage. To define the Damage Boundary, I analyzed the coupling between short-circuit duration and battery aging. Tests were performed with controlled short durations (10s, 15s, 20s, 25s) under identical initial conditions (80% SOC, 20°C, 20 mΩ), stopping before the failure boundary (~27s). Post-test capacity fade and Electrochemical Impedance Spectroscopy (EIS) were analyzed.
The capacity fade was more severe for longer short durations, primarily due to the higher peak temperature reached. EIS analysis revealed that the growth of the Solid Electrolyte Interphase (SEI) layer, accelerated by high temperature, was the dominant aging mechanism. The increase in SEI resistance ($R_{SEI}$) was significantly larger than changes in ohmic ($R_{\Omega}$) or charge transfer resistance ($R_{ct}$) for batteries subjected to longer shorts.
To establish a universal damage criterion, I considered temperature from two perspectives: electrolyte stability and battery safety onset. LiPF$_6$ salt decomposition, catalyzed by trace moisture, can begin around 80°C. Furthermore, the exothermic decomposition of the SEI layer, widely considered the initiating reaction for thermal runaway, typically starts between 80°C and 120°C. Therefore, defining the time from ESC initiation until the battery’s internal temperature reaches 80°C as the Damage Boundary (T_d) is a conservative and mechanistically justified threshold for the battery energy storage system.
Internal Temperature Estimation via Lumped-Parameter Thermal Model
Accurately determining the damage boundary requires knowledge of the internal temperature, which is difficult to measure directly. Surface temperature measurements are insufficient due to significant internal gradients (observed to exceed 30°C during ESC). Therefore, I developed a lumped-parameter thermal model coupled with a state observer for real-time internal temperature estimation.
The simplified thermal model for a cylindrical cell is represented as an electrical analogy:
$$Q = I(U_{ocv} – U_t) – IT\frac{dU_{ocv}}{dT} + Q_{SEI} + Q_{NE} + Q_{PE} + Q_{ELE}$$
where $Q$ is the total heat generation rate, $I$ is current, $U_{ocv}$ is open-circuit voltage, $U_t$ is terminal voltage, $T$ is temperature, and the last four terms represent heat from side reactions (SEI decomposition, negative/positive electrolyte reactions, electrolyte decomposition). The side reaction kinetics follow Arrhenius-type equations. For example, the SEI decomposition rate is:
$$R_S = A_S \exp\left(-\frac{E_{a,S}}{RT}\right) c_S^{m_S}$$
$$Q_{SEI} = H_S W_C R_S$$
Parameters for these reactions, identified from literature, are summarized below:
| Parameter | Symbol | Value |
|---|---|---|
| SEI Decomposition Pre-exponential Factor | $A_S$ | 1.667×10$^{15}$ s$^{-1}$ |
| SEI Decomposition Activation Energy | $E_{a,S}$ | 1.3508×10$^{5}$ J mol$^{-1}$ |
| SEI Reaction Enthalpy | $H_S$ | 257 J g$^{-1}$ |
| Negative Electrolyte Reaction Pre-exp. Factor | $A_N$ | 2.5×10$^{13}$ s$^{-1}$ |
| Negative Electrolyte Reaction Activation Energy | $E_{a,N}$ | 1.3508×10$^{5}$ J mol$^{-1}$ |
| Electrolyte Decomposition Pre-exp. Factor | $A_E$ | 5.14×10$^{25}$ s$^{-1}$ |
| Electrolyte Decomposition Activation Energy | $E_{a,E}$ | 2.74×10$^{5}$ J mol$^{-1}$ |
The thermal dynamics are described by a state-space model discretized from the energy balance equations for the core ($T_c$) and surface ($T_s$):
$$C_c \dot{T_c} = Q – \frac{T_c – T_s}{R_c}$$
$$C_{air} \dot{T_s} = \frac{T_c – T_s}{R_c} – \frac{T_s – T_{air}}{R_{air}}$$
Discretizing yields:
$$\mathbf{x}_{k+1} = \mathbf{A} \mathbf{x}_k + \mathbf{B} \mathbf{u}_k$$
$$\mathbf{y}_k = \mathbf{C} \mathbf{x}_k$$
where $\mathbf{x} = [T_c, T_s]^T$, $\mathbf{y}=T_s$, $\mathbf{u}=[Q, T_{air}]^T$. The model parameters ($R_c$, $C_c$, $R_{air}$, $C_{air}$) were identified using genetic algorithm optimization. A PID observer was then implemented on this model to estimate the internal core temperature ($\hat{T}_c$) from measurable inputs ($I$, $U_t$, $T_s$, $T_{air}$). Validation under various discharge profiles showed a maximum estimation error within ±2°C, confirming its accuracy for application in the battery energy storage system.
Prediction of Damage and Failure Boundaries Using LSTM Neural Network
Fast and accurate prediction of the impending damage and failure boundaries immediately after an ESC fault is crucial for proactive safety management in a battery energy storage system. To achieve this, I developed a data-driven prediction model using a Long Short-Term Memory (LSTM) neural network, renowned for handling sequential data.
The model takes the first 3 seconds of ESC data as input: 1 second of pre-fault data (to capture initial SOC and temperature) and 2 seconds of post-fault data (to capture the dynamic response). The input features are current, voltage, and ambient temperature, sampled at 20 Hz, forming an input matrix of size 3 × 60. The model outputs are the predicted Damage Boundary ($\hat{T}_d$) and Failure Boundary ($\hat{T}_f$).
The LSTM network architecture consisted of an input layer (3 nodes), 6 LSTM hidden layers, a dropout layer for regularization, and an output layer (2 nodes). The model was trained on the experimental dataset (75 samples covering various SOC, temperature, and resistance conditions) using the Adam optimizer.
The trained model demonstrated excellent predictive performance. For example, under test conditions of -20°C ambient temperature and 10 mΩ resistance across different SOCs, the maximum prediction error for the Damage Boundary was less than 3.5%, and for the Failure Boundary, it was less than 2%. This high accuracy indicates that the LSTM model effectively learns the complex mapping between the early-stage electrical-thermal response and the ultimate safety boundaries, providing a powerful tool for real-time risk assessment in a battery energy storage system.
Conclusion
This work systematically investigates the external short circuit fault in lithium-ion batteries, focusing on defining and predicting critical safety boundaries to enhance the reliability of the battery energy storage system.
First, I characterized the ESC behavior under diverse conditions, identifying a clear three-stage response and defining the Failure Boundary as the time to current/voltage collapse. Second, by analyzing the coupling between short-circuit duration and aging, I established the growth of the SEI layer, driven primarily by temperature, as the key damage mechanism. Consequently, I defined the Damage Boundary as the time for the internal temperature to reach 80°C. Third, to access this internal temperature, I developed a lumped-parameter thermal model with a PID observer, achieving accurate internal temperature estimation with errors below ±2°C. Finally, I constructed an LSTM neural network model capable of predicting both the Damage and Failure Boundaries within 3 seconds of fault onset, with errors below 3.5% and 2%, respectively.
The proposed framework—from boundary definition to model-based estimation and data-driven prediction—provides a comprehensive strategy for understanding, monitoring, and mitigating the risks associated with external short circuits. The implementation of such predictive safety models is essential for developing smarter, more resilient battery management systems, thereby ensuring the long-term safety and health of grid-scale and distributed battery energy storage system installations.