With the rapid growth of renewable energy sources like wind and solar power, the intermittent and fluctuating nature of their output necessitates the use of energy storage technologies. Among various energy storage methods, electrochemical energy storage, particularly lithium-ion batteries, has gained significant attention due to high conversion efficiency, long cycle life, and flexibility. Energy storage cells based on lithium iron phosphate (LFP) chemistry are widely adopted in large-scale energy storage systems because of their stability, safety, and cost-effectiveness. However, issues such as performance degradation and protection system failures can lead to abusive conditions like overcharging, potentially causing thermal runaway—a critical safety challenge. Traditional monitoring methods for energy storage cells, such as voltage, current, and temperature measurements at the module level, or gas and smoke detection at the cluster level, often exhibit delayed responses to internal abnormalities. To address this, we propose an in-situ ultrasonic transmission method for internal state monitoring during thermal runaway in energy storage cells, combined with multi-parameter detection including temperature, voltage, current, and surface deformation. This integrated approach enables precise monitoring of internal states under various operating conditions and facilitates early warning for overcharge-induced thermal runaway.
Ultrasonic testing (UT) is a non-destructive technique that offers high sensitivity and real-time capabilities. Unlike external parameters, ultrasonic signals can directly reflect internal changes in energy storage cells, such as electrolyte saturation, gas generation, and electrode structural evolution. In this study, we focus on large-capacity prismatic aluminum-shell LFP energy storage cells, which are commonly used in energy storage stations but pose challenges for ultrasonic penetration due to their thickness and reflective casing. By selecting appropriate ultrasonic frequencies, we establish a relationship between ultrasonic transmission signals and the spatiotemporal evolution of thermal runaway. Key ultrasonic parameters include the root mean square signal intensity (Urms) and time of flight (TOF), which are normalized as relative Urms (RU) and relative TOF (RT) for comparative analysis. The experimental system comprises an ultrasonic transmission setup with transducers, signal generator, amplifier, and oscilloscope, integrated with a multi-parameter testing platform for overcharge thermal runaway. Sensors are arranged on the cell surface to measure voltage, current, temperature at multiple points, surface strain, and gas emissions (e.g., CO, CO2, H2).
The ultrasonic signal intensity Urms is calculated as follows:
$$ U_{rms} = \sqrt{\frac{1}{N} \sum_{i=1}^{N} U_i^2 } $$
where Ui represents the voltage value of the i-th data point in the received ultrasonic signal, and N is the total number of data points. The relative parameters RU and RT are defined as:
$$ R_U = \frac{U_{rms}(t=n)}{U_{rms}(t=0)} $$
$$ R_T = \frac{TOF(t=n)}{TOF(t=0)} $$
where t=n denotes the time instance, and t=0 is the initial reference time. This normalization allows for consistent comparison across different energy storage cells and conditions.
Under static conditions, the effects of temperature and state of charge (SOC) on ultrasonic signals are investigated. As temperature increases, RU decreases due to reduced electrolyte viscosity and density, which amplifies the impedance mismatch between electrodes and electrolyte, leading to lower ultrasonic transmission. Conversely, RT increases with temperature because of thermal expansion of electrodes and changes in material properties that reduce ultrasonic wave velocity. The sensitivity of RU to temperature is more pronounced than that of RT, with RU decreasing by approximately 80% when temperature rises from 20°C to 45°C, while RT increases by about 10%. A notable inflection point occurs around 35°C, attributed to the phase transition of ethylene carbonate in the electrolyte, which abruptly changes elastic modulus and attenuation. The following table summarizes the ultrasonic response to temperature variations:
| Temperature (°C) | RU (Normalized) | RT (Normalized) |
|---|---|---|
| 20 | 1.00 | 1.00 |
| 25 | 0.85 | 1.02 |
| 30 | 0.70 | 1.04 |
| 35 | 0.50 | 1.07 |
| 40 | 0.30 | 1.09 |
| 45 | 0.20 | 1.10 |
Regarding SOC effects, RU increases by up to 11.56% as SOC rises from 0% to 100%, primarily due to reduced impedance mismatch and attenuation from changes in electrode elastic moduli and lithium-ion concentration. In contrast, RT decreases linearly by up to 0.67% with increasing SOC, as the wave velocity increase from graphite anode lithiation outweighs the thickness expansion. The table below illustrates ultrasonic parameters at different SOC levels:
| SOC (%) | RU (Normalized) | RT (Normalized) |
|---|---|---|
| 0 | 1.00 | 1.00 |
| 20 | 1.05 | 0.995 |
| 40 | 1.09 | 0.990 |
| 60 | 1.11 | 0.985 |
| 80 | 1.10 | 0.980 |
| 100 | 1.08 | 0.993 |
During normal charge-discharge cycles, ultrasonic signals exhibit dynamic behaviors influenced by current rates. RU follows a similar trend to static conditions but with abrupt changes at the start and end of cycling due to concentration polarization. RT variations are dominated by temperature changes, showing a strong correlation with surface temperature. For instance, at a 0.5 C rate, RT increases by 5% during charging, aligning with a temperature rise of 10°C. The consistency of RU across different cycling rates makes it a reliable indicator for internal state monitoring in energy storage cells during operation.
Under overcharge conditions, energy storage cells undergo distinct stages before thermal runaway. In Stage I, voltage increases gradually due to lithium plating and SEI layer decomposition; RU decreases steadily, and RT increases slowly. Stage II involves accelerated gas generation from electrolyte decomposition and lithium dendrite growth, causing rapid attenuation of ultrasonic signals—RU drops to near zero, and RT becomes unmeasurable. Stage III culminates in venting valve activation, characterized by voltage drop, temperature spike, and gas emission. Crucially, ultrasonic signal attenuation occurs significantly earlier than venting; for example, at 0.5 C overcharge, signals are淹没 by noise at 586 s, while venting happens at 2418 s. This early warning capability is consistent across different charging rates, as shown in the table below for overcharge experiments:
| Charging Rate (C) | Ultrasonic Signal Loss Time (s) | Venting Valve Activation Time (s) | Voltage at Venting (V) |
|---|---|---|---|
| 0.25 | 1150 | 5154 | 4.99 |
| 0.5 | 586 | 2418 | 4.98 |
| 0.75 | 257 | 1537 | 5.01 |
To leverage ultrasonic signals for early warning, we develop an algorithm based on the improved Mahalanobis-Taguchi system (IMTS). This method addresses limitations of traditional MTS, such as multicollinearity and feature weighting, by incorporating weighted Mahalanobis distance (WMD) and health index (HI). The process involves: (1) selecting feature variables—voltage, mid-point temperature, ultrasonic Urms, ultrasonic TOF, H2 volume fraction, and mid-point strain; (2) normalizing data using Z-score method; (3) constructing a baseline space with normal samples (e.g., data from charge-discharge cycles); (4) calculating WMD using Gram-Schmidt orthogonalization and Fisher weighting; and (5) setting HI thresholds via 3σ准则 for anomaly detection. The WMD is computed as:
$$ WMD = \sqrt{ ( \mathbf{x} – \mathbf{\mu} )^T \mathbf{W} \mathbf{S}^{-1} \mathbf{W} ( \mathbf{x} – \mathbf{\mu} ) } $$
where $\mathbf{x}$ is the sample vector, $\mathbf{\mu}$ is the mean vector of normal samples, $\mathbf{S}$ is the covariance matrix, and $\mathbf{W}$ is the diagonal weight matrix derived from Fisher scores. The health index HI is then defined as:
$$ HI = \exp(-\varepsilon \cdot WMD) $$
where $\varepsilon$ is a tuning coefficient. A threshold HIt = 0.9222 is established, and if HI falls below this for three consecutive samples, an overcharge fault is flagged.
Validation using test data from a 0.5 C overcharge experiment shows that the algorithm triggers an alert at 7262 s (SOC 104.2%), which is 30 minutes prior to venting valve activation. At this point, ultrasonic Urms has decayed to 0.677 V (from initial 1.0 V), TOF is 20.95 μs, voltage is 4.46 V, temperature is 21.91°C, H2 volume fraction is zero, and strain is 53×10−6. The early detection allows for preventive measures like discharging to avoid thermal runaway in energy storage cells.
In conclusion, ultrasonic detection technology provides a sensitive and early indicator of internal changes in energy storage cells under various conditions. The relative ultrasonic intensity RU is particularly effective for monitoring temperature and SOC dynamics, while the attenuation of ultrasonic signals during overcharge serves as a reliable precursor to thermal runaway. The IMTS-based warning algorithm, incorporating multiple parameters, enables timely fault detection with a 30-minute advance notice. This approach enhances the safety management of energy storage systems by addressing the limitations of external monitoring methods. Future work could focus on optimizing ultrasonic frequencies for different cell formats and integrating machine learning for adaptive threshold setting in real-world energy storage applications.