The widespread integration of renewable energy sources like wind and solar power into the grid is fundamentally constrained by their inherent intermittency and volatility. To ensure grid stability and maximize the utilization of this clean energy, large-scale cell energy storage system technology has become indispensable. Among various electrochemical storage solutions, Lithium Iron Phosphate (LFP) batteries are currently the dominant choice for stationary cell energy storage system deployments in many regions, prized for their superior safety profile, long cycle life, and cost-effectiveness compared to other lithium-ion chemistries. However, the safe operation of these massive cell energy storage system installations remains a paramount concern. Thermal runaway (TR), a vicious cycle of self-heating leading to fire or explosion, poses a severe threat. This risk can be triggered by factors such as cell inconsistency, long-term performance degradation, and protection system failures, often culminating in abusive conditions like overcharging.
Current state monitoring in LFP-based cell energy storage system primarily relies on external measurements at the module level (voltage, current, temperature) and at the cluster or container level (gas, smoke). While these methods provide valuable data, they detect manifestations of internal faults—such as gas venting or surface deformation—only after significant internal damage has occurred. This inherent lag limits the effectiveness of early warning systems. Consequently, there is a critical need for an in-situ, non-destructive technology capable of probing the internal state of a battery cell in real-time, providing a much earlier indication of impending failure.
Ultrasonic testing (UT) emerges as a highly promising solution for this challenge. As a mature non-destructive evaluation technique, UT offers high sensitivity, rapid response, and the unique ability to detect micro-structural changes within a material. Recent research has successfully applied UT to characterize the state of charge (SOC) and detect faults like gas formation in small-format, pouch-type lithium-ion cells. However, the practical application of UT to large-format, prismatic aluminum-cased LFP batteries—the workhorse of modern grid-scale cell energy storage system—presents distinct challenges. Their greater thickness (often >20 mm) and the highly reflective aluminum casing significantly attenuate and reflect ultrasonic signals, making signal acquisition and interpretation more difficult than with thin pouch cells.
In this study, we address this gap by developing and validating an in-situ ultrasonic transmission method specifically designed for monitoring the internal state of prismatic LFP batteries. We integrate this core technology with a comprehensive suite of external parameter measurements—including temperature, voltage, current, surface strain, and gas composition—to construct a multi-parameter, internal-external testing system for battery thermal runaway. Our work systematically investigates the correlation between ultrasonic signal features and the cell’s internal evolution under various states: static, normal cycling, and the critical overcharge condition leading to thermal runaway. We elucidate the underlying physical mechanisms linking ultrasonic wave propagation to changes in electrode properties, electrolyte state, and gas generation. Crucially, we demonstrate that ultrasonic signal attenuation serves as a highly sensitive early-warning indicator for overcharge-induced thermal runaway, far preceding traditional warning signs like venting. Building on this finding, we propose and validate an enhanced early-warning algorithm based on the Improved Mahalanobis-Taguchi System (IMTS), which successfully predicts safety vent opening with a significant lead time. This research provides a vital technological foundation for advancing the safety and reliability of large-scale cell energy storage system through proactive, internal-state-based diagnostics.
1. Methodology and Experimental Setup
1.1 Battery Samples and Preparation
The subject of this study is a commercial 20 Ah prismatic aluminum-cased Lithium Iron Phosphate (LFP) battery, representative of the cells used in many stationary cell energy storage system. To ensure experimental consistency and replicate a stable initial condition, all test cells underwent a standardized pre-treatment protocol. This involved three complete charge-discharge cycles at a 0.5C rate (10 A), using constant current-constant voltage (CC-CV) charging and constant current (CC) discharging, followed by a resting period of at least 12 hours before any experimental tests.
1.2 Integrated Multi-Parameter Testing System
We designed and constructed a comprehensive experimental platform to simultaneously capture the internal (ultrasonic) and external evolution of a battery during abuse testing. The system centers on an in-situ ultrasonic through-transmission measurement setup, augmented by a previously developed multi-parameter characterization platform for battery thermal runaway.
Core Ultrasonic Transmission System: The system utilizes a pair of ultrasonic transducers (one transmitter, one receiver) placed on opposite, wide faces of the battery. A signal generator produces a Gaussian-windowed, 5-cycle sinusoidal voltage pulse, which is amplified to drive the transmitting transducer. The ultrasonic wave propagates through the battery’s internal layers, and the received signal is captured by an oscilloscope. The choice of transducer frequency (optimized in preliminary tests) balances penetration depth and resolution for the specific battery thickness.
External Parameter Monitoring Suite: This includes:
- Electrical Parameters: A battery cycler records voltage and current with high precision.
- Thermal Monitoring: Multiple K-type thermocouples are attached to strategic locations: the cell surface near the ultrasonic transducers (center), beneath the terminal posts, and at the bottom corners to map temperature gradients.
- Mechanical Deformation: High-temperature resistant strain gauges are bonded to the cell surface (center and upper corner) to measure the evolution of surface strain, an indicator of internal pressure build-up.
- Gas Analysis: A small fan continuously extracts gas from the vicinity of the safety vent into a sealed collection box equipped with electrochemical sensors for CO, CO2, and H2.
All sensors are connected to data loggers synchronized with the ultrasonic oscilloscope, enabling time-correlated analysis of all parameters.
1.3 Ultrasonic Signal Processing and Feature Extraction
Two key features are extracted from the received ultrasonic signal to characterize the battery’s internal state:
- Ultrasonic Signal Intensity (Urms): The root-mean-square voltage of the received signal, representing the transmitted energy. It is calculated over the entire signal window containing the main wave packet:
$$ U_{rms} = \sqrt{\frac{1}{N}\sum_{i=1}^{N} U_i^2} $$
where $U_i$ is the voltage of the i-th data point and $N$ is the total number of points. - Time of Flight (TOF): The time interval between the trigger pulse sent to the transmitter and the arrival of a specific feature (e.g., the first peak or zero-crossing) of the received signal. It is related to the effective ultrasonic wave speed and propagation path length within the battery.
To facilitate comparison between different cells or different test conditions, and to mitigate the effects of initial coupling variations, we normalize these features to their initial values at the start of a test, defining relative parameters:
- Relative Urms (RU) = Urms(t) / Urms(t=0)
- Relative TOF (RT) = TOF(t) / TOF(t=0)
2. Results and Discussion: Ultrasonic Response to Battery States
2.1 Static State Characterization: Influence of Temperature and SOC
We first established baseline correlations between ultrasonic features and fundamental battery states under non-operating conditions.
2.1.1 Temperature Dependence: Tests were conducted on batteries at different initial SOC levels while controlling the ambient temperature. As shown in the data summary below, $R_U$ exhibits a strong, monotonic decrease with rising temperature, while $R_T$ increases.
| Condition | Trend of $R_U$ | Trend of $R_T$ | Key Observation |
|---|---|---|---|
| Temperature Increase (20°C to 45°C) | Decreases sharply (~80% drop) | Increases gradually (~10% increase) | $R_U$ is far more sensitive to temperature than $R_T$. A distinct change in slope is observed near 35°C, likely due to phase change in the electrolyte solvent. |
The underlying mechanism involves changes in material properties: increased temperature reduces electrolyte viscosity and density, lowering its acoustic impedance and increasing the impedance mismatch at electrode-electrolyte interfaces, thereby reducing ultrasonic transmission ($R_U$). Simultaneously, thermal expansion increases the propagation path length, and a reduction in the elastic moduli of components decreases wave speed, both contributing to an increased $R_T$.
2.1.2 State of Charge (SOC) Dependence: Ultrasound scans were performed on batteries at rest at various SOC levels.
| Condition | Trend of $R_U$ | Trend of $R_T$ | Key Observation |
|---|---|---|---|
| SOC Increase (0% to 100%) | Increases to a maximum, then slightly decreases (~11.5% max change) | Decreases linearly (~0.7% total change) | $R_U$ is more sensitive to SOC variation than $R_T$. The non-monotonic trend in $R_U$ is linked to changing acoustic impedance matching between electrodes as their elastic properties evolve with lithium concentration. |
The linear decrease in $R_T$ with increasing SOC is primarily governed by the graphite anode, whose elastic modulus increases upon lithiation, leading to a higher effective wave speed that outweighs the effect of slight electrode thickening.
2.2 Dynamic State: Normal Charge-Discharge Cycling
Under normal operation (0.25C, 0.5C, 0.75C cycling), the ultrasonic features show characteristic, repeatable patterns.
- $R_U$ vs. SOC: The relationship observed during static tests largely holds during dynamic cycling. However, abrupt changes in $R_U$ occur at the instant current is applied or removed, attributable to concentration polarization at the electrodes. After current interruption, $R_U$ relaxes to a stable value as the lithium concentration gradient equilibrates.
- $R_T$ during Cycling: The evolution of $R_T$ during a cycle closely mirrors the cell’s surface temperature curve. This indicates that during normal operation, the temperature-induced effects on wave speed and expansion dominate the $R_T$ signal, overshadowing the more subtle SOC-dependent mechanical changes observed in static tests.
Conclusion for Normal Operation: $R_U$ proves to be a more robust indicator of the internal electrochemical state (SOC) during both static and dynamic conditions, as it is less dominated by concurrent temperature changes compared to $R_T$.
2.3 Overcharge Abuse Leading to Thermal Runaway
This is the critical test for safety warning. Cells at 100% SOC were subjected to continued constant current charging (0.25C, 0.5C, 0.75C) until thermal runaway occurred. The process before venting can be divided into three stages based on ultrasonic signal evolution.
Stage I (Early Overcharge): Voltage rises linearly. $R_U$ begins a steady decline, and $R_T$ starts a steady increase. Surface temperature rises slowly. No gas is detected externally, but minute surface strain indicates initial internal pressure build-up.
Stage II (Progressive Degradation): Voltage rise starts to plateau. $R_U$ decay accelerates sharply, and $R_T$ increase accelerates. Temperature rise rate increases. Surface strain grows more noticeably. This stage is marked by severe lithium plating on the anode, accelerated SEI decomposition and reformation, and the onset of gaseous side reactions (e.g., Li + electrolyte → H2, Li2CO3 decomposition → CO2). The generated gas bubbles cause intense scattering and attenuation of ultrasound, leading to the rapid collapse of $R_U$.
Stage III (Imminent Failure): Voltage may peak and begin to drop. $R_U$ decays to near-zero (signal buried in noise). $R_T$ becomes unmeasurable. Temperature rises sharply. Strain reaches a maximum. Finally, the safety vent opens (venting time, tvent), releasing a burst of gas (H2, CO, CO2) and causing a sudden drop in surface strain.
The most significant finding is the temporal relationship between ultrasonic signal failure and venting. The time at which the ultrasonic signal is completely lost (tultra-loss) provides a critical early warning.
| Overcharge Rate | tultra-loss (s) | tvent (s) | Warning Lead Time (s) |
|---|---|---|---|
| 0.25C | ~1150 | ~5154 | ~4004 (~66.7 min) |
| 0.5C | ~586 | ~2418 | ~1832 (~30.5 min) |
| 0.75C | ~257 | ~1537 | ~1280 (~21.3 min) |
This table demonstrates that ultrasonic signal attenuation provides a consistent and substantial early warning window ahead of the safety vent opening, a key last-ditch safety device in a cell energy storage system cell. The lead time decreases with increasing overcharge rate but remains operationally significant.
3. Development of an Integrated Early-Warning Algorithm
While ultrasonic signal loss is a powerful indicator, relying on a single parameter can be risky. We therefore developed a multi-parameter fusion algorithm based on the Improved Mahalanobis-Taguchi System (IMTS) for robust state assessment and early warning.
3.1 Algorithm Framework
The IMTS is a pattern recognition method robust to unbalanced data. It constructs a baseline “healthy space” using data from normal battery operation and quantifies the deviation of real-time data from this space using a weighted Mahalanobis Distance (WMD). A Health Index (HI) is derived, and a threshold is set to trigger an alarm.
Steps of the Proposed Algorithm:
- Feature Selection: Six critical, real-time measurable parameters are chosen as feature variables (X): Cell Voltage, Center Temperature, Ultrasonic Urms, Ultrasonic TOF, H2 Concentration, and Center Surface Strain.
- Data Normalization & Baseline Space Construction: Data from normal charge-discharge cycles (healthy state) are standardized (Z-score) to form the baseline space. The standardized variables are orthogonalized using the Gram-Schmidt Process (GSP).
- Calculation of Weighted Mahalanobis Distance (WMD): The importance of each feature for distinguishing fault conditions is different. We employ Fisher’s discriminant criterion to assign an optimal weight $w_i$ to each i-th orthogonalized feature $Y_i$. The WMD for a sample is calculated as:
$$ WMD = \sqrt{ \sum_{i=1}^{k} \left( \frac{w_i \cdot Y_i}{s_i} \right)^2 } $$
where $s_i$ is the standard deviation of the i-th orthogonalized feature in the healthy baseline, and $k$ is the number of features. - Health Index (HI) and Threshold Setting: The HI is defined to range from 0 (complete failure) to 1 (perfect health), mapped from the WMD. Using the $3\sigma$ principle on the healthy data distribution, an alarm threshold HIt is established. A fault is declared if the HI value falls below HIt for a consecutive number of samples (e.g., 3 samples to avoid false alarms from noise).
The weights calculated for our system highlight the importance of internal sensing:
| Feature Variable | Assigned Weight ($w_i$) | Relative Importance |
|---|---|---|
| Ultrasonic Urms | 0.253 | Highest |
| Center Temperature | 0.221 | High |
| Ultrasonic TOF | 0.198 | High |
| H2 Concentration | 0.148 | Medium |
| Cell Voltage | 0.105 | Medium |
| Center Strain | 0.075 | Lower |
3.2 Algorithm Validation and Performance
The algorithm was trained on data from a healthy cycle and an overcharge-to-failure test at 0.5C. It was then validated on a separate test where a cell was charged from 0% SOC to 110% SOC at 0.5C. The results are compelling:
- The HI remained stable and high (>0.95) during most of the normal charging process.
- At approximately 650 seconds after the start of overcharge (SOC ~104.2%), the HI dropped precipitously below the threshold HIt = 0.922.
- This alarm was triggered over 30 minutes (1768 seconds) before the actual safety vent opening occurred in the reference overcharge test (at 2418s).
- At the alarm moment, external indicators were still benign: voltage was at 4.46V (still rising), temperature was mild at 21.9°C, and no H2 was detected. The primary indicators of abnormality were the ultrasonic signals ($R_U$ and $R_T$) just beginning their characteristic decay.
This validates the algorithm’s capability to fuse multi-parameter data, with a strong emphasis on internal ultrasonic features, to provide a reliable, early warning of overcharge abuse well before catastrophic failure, enabling protective actions (like disconnection) to be taken to secure the cell energy storage system.
4. Conclusion
This study successfully demonstrates the application of in-situ ultrasonic through-transmission testing as a powerful diagnostic tool for large-format, prismatic LFP batteries, directly addressing a key safety need in stationary cell energy storage system. Our integrated multi-parameter testing system reveals the fundamental correlations between ultrasonic wave propagation features (Urms and TOF) and the internal state of the battery under static, dynamic, and abusive conditions.
The key findings and contributions are:
- Sensitive Internal State Monitoring: Ultrasonic signal intensity ($R_U$) is highly sensitive to both temperature and State of Charge (SOC), providing a means to monitor internal electrochemical and physical changes in real-time, complementing external measurements.
- Earliest Indicator of Overcharge Failure: During overcharge leading to thermal runaway, the attenuation of the ultrasonic signal (drop in $R_U$ to noise level) is the earliest detectable abnormal event among all monitored parameters. It occurs tens of minutes before the safety vent opens, which is the trigger for most conventional gas-based alarms in a cell energy storage system.
- Mechanistic Understanding: The evolution of ultrasonic signals is interpreted through the lens of changing material properties (elastic moduli, density), interfacial impedance, and, most critically, gas generation which causes severe scattering. This provides a physical basis for the warning signal.
- Proactive Safety Algorithm: We developed and validated an Improved Mahalanobis-Taguchi System-based early-warning algorithm that fuses ultrasonic data with traditional voltage, temperature, gas, and strain measurements. The algorithm assigned the highest weight to ultrasonic features and successfully predicted venting with a 30-minute lead time, demonstrating a viable path towards proactive safety management for battery energy storage systems.
The implementation of such internal sensing and analytics can significantly enhance the safety paradigm for grid-scale cell energy storage system, moving from passive containment and late-stage fire suppression to active, early fault detection and prevention. Future work will focus on optimizing the sensor integration for practical deployment, expanding the algorithm’s robustness to other failure modes (e.g., internal short circuit), and testing the methodology at the module or pack level within a representative cell energy storage system environment.