Ensuring the operational safety of large-scale cell energy storage systems remains one of the most pressing technical challenges today. Among the various battery chemistries deployed, lithium iron phosphate (LFP) batteries are a dominant choice due to their inherent stability and cost-effectiveness. However, our experience and field data confirm that these systems are not immune to failure. Performance degradation over long-term use, inconsistencies between cells, and potential protection system failures can lead to abusive conditions like overcharging, which may culminate in thermal runaway—a catastrophic event involving uncontrolled temperature rise, gas venting, and potentially fire.
Current monitoring strategies in LFP battery energy storage plants primarily focus on external parameters. At the module level, we routinely monitor voltage, current, and surface temperature. At the cluster or container level, we deploy gas and smoke detectors. While these methods provide valuable data, they possess an inherent limitation: they detect consequences of internal faults, not the initiation of the faults themselves. There is a critical time lag between an internal anomaly occurring and its manifestation as a measurable change in voltage, temperature, or gas emission. For instance, previous studies, including our own, have shown that surface deformation or specific gas concentrations can serve as warning signs, but these often signal an already advanced stage of internal failure. To truly prevent disaster, we need a diagnostic window into the battery’s internal state, allowing for intervention before irreversible damage propagates.
This pursuit led our team to explore in-situ ultrasonic testing (UT) technology. As a non-destructive evaluation technique, UT offers high sensitivity to changes in material density, elasticity, and internal structure. Prior research has successfully applied it to smaller-format, pouch-type lithium-ion cells. However, translating this success to the large-format, prismatic aluminum-cased LFP batteries used in grid-scale cell energy storage systems presents significant challenges. The thicker casing (often >20 mm) and the highly reflective aluminum shell, compared to the thin, acoustically compliant aluminum-laminate pouch, severely attenuate and reflect ultrasonic signals. Our innovation lies in adapting the ultrasonic transmission method to overcome these barriers, selecting an optimal frequency that balances penetration depth with resolution to effectively probe the internal state of commercial-grade LFP cells.
In this article, we present our integrated, multi-parameter testing system designed to correlate internal ultrasonic signatures with external thermal, electrical, and mechanical responses. We detail our findings on how ultrasonic signals evolve during normal operation (static and cycling) and during the critical overcharge-to-thermal-runaway process. Crucially, we demonstrate that ultrasonic signal decay serves as a highly sensitive early-warning indicator, preceding traditional signs of failure by a significant margin. Building on this insight, we developed and validated a novel prognostic algorithm based on an Improved Mahalanobis-Taguchi System (IMTS), enabling the early prediction of venting events. This work advances the fundamental understanding of spatiotemporal evolution within failing batteries and provides a tangible technological pathway for enhancing safety protocols in cell energy storage systems.
1. Our Methodology and Experimental Framework
Our investigation centered on commercial 20 Ah prismatic aluminum-cased LFP batteries, the workhorse of contemporary cell energy storage systems. To ensure consistent initial conditions, all cells underwent three formation cycles (0.5C charge/discharge) followed by a 12-hour rest period prior to any testing.
The core of our approach was the development of an in-situ ultrasonic transmission system integrated into a comprehensive thermal runaway characterization platform. The system schematic is conceptualized below, highlighting the simultaneous acquisition of internal and external parameters.
| System Component | Measurement Target | Key Parameters |
|---|---|---|
| Ultrasonic Core | Internal Structural State | Signal Intensity (Urms), Time-of-Flight (TOF) |
| Electrical Monitor | Cell Electrical Condition | Terminal Voltage, Current |
| Thermal Array | Surface Temperature Distribution | Temperature at 5 critical locations |
| Mechanical Sensor | Case Deformation | Micro-strain at cell center and upper corner |
| Gas Analysis Unit | Emitted Gases (Post-venting) | Concentrations of H2, CO, CO2 |
For ultrasonic interrogation, we employed a through-transmission method. A function generator produced a Gaussian-windowed, 5-cycle sine pulse, which was amplified to drive a transmitting transducer coupled to one side of the battery. The signal propagating through the cell was received by an identical transducer on the opposite side and captured by a digital oscilloscope. We focused on two primary ultrasonic features extracted from the received signal:
1. Root-Mean-Square Voltage (Urms): Representing the signal’s energy or intensity, sensitive to attenuation and interfacial reflections.
2. Time-of-Flight (TOF): The time taken for the ultrasonic wave to traverse the cell, sensitive to wave velocity and propagation path length.
To enable comparison across different cells and tests, we normalized these values to their initial readings, defining Relative Urms (RU) and Relative TOF (RT):
$$ RU = \frac{U_{rms}(t)}{U_{rms}(t_0)}, \quad RT = \frac{TOF(t)}{TOF(t_0)} $$
where \(t_0\) denotes the initial measurement time.
Experiments were conducted in three phases: 1) Characterizing ultrasonic response to temperature and State-of-Charge (SOC) under static conditions; 2) Observing signal behavior during normal charge-discharge cycles at various rates (0.25C, 0.5C, 0.75C); and 3) Inducing and monitoring overcharge thermal runaway at the same rates until venting occurred.
2. Ultrasonic Signatures of Battery Internal State
2.1 The Static Battery: Temperature and SOC Dependence
We first established baselines by examining how ultrasonic features change with core internal states. The relationship between signal parameters and temperature for cells at different SOCs is summarized by the following empirical trends:
$$ RU(T) \approx RU_0 \cdot e^{-\alpha (T – T_0)} $$
$$ RT(T) \approx RT_0 \cdot [1 + \beta (T – T_0)] $$
where \(\alpha\) and \(\beta\) are positive temperature coefficients. Our data showed that \(RU\) decreased by approximately 80% when temperature rose from 20°C to 45°C, while \(RT\) increased by only about 10%. This demonstrates that Ultrasonic Signal Intensity is far more sensitive to thermal changes than Time-of-Flight. A distinct inflection point near 35°C was observed, which we attribute to the phase change of the ethylene carbonate solvent in the electrolyte, causing a sudden shift in its acoustic impedance.
The sensitivity to SOC was equally pronounced. As lithium ions shuttle between electrodes, the mechanical properties of the active materials change. We found \(RU\) exhibited a non-linear relationship with SOC, initially increasing as the impedance mismatch between electrodes decreased, then slightly dropping at very high SOC. Conversely, \(RT\) decreased almost linearly with increasing SOC, primarily governed by the increasing wave speed in the graphite anode as it lithiates. The magnitude of change further confirmed the superior sensitivity of \(RU\) for SOC estimation.
| State Variable | Primary Effect on Urms (RU) | Primary Effect on TOF (RT) | Relative Sensitivity (RU vs. RT) |
|---|---|---|---|
| Temperature Increase | Strong Decrease (High Sensitivity) | Moderate Increase (Lower Sensitivity) | RU >> RT |
| State-of-Charge (SOC) Increase | Non-linear Increase (~11.5% max) | Linear Decrease (~0.7% max) | RU > RT |
2.2 Dynamic Cycling: Decoupling Effects
Under charge/discharge cycles, the ultrasonic signals superimposed the effects of SOC change, concentration polarization, and Joule heating. Our key finding was that while \(RT\) variations closely tracked the cell’s surface temperature curve (dominated by thermal expansion), the \(RU\) curve maintained its characteristic shape related to SOC, as observed in static tests. This confirms that RU can robustly indicate the internal lithium distribution state even during dynamic operation of a cell energy storage system, making it a reliable real-time metric.
3. The Overcharge Thermal Runaway Process: A Multi-Parameter Chronicle
We now dissect the sequence of events during an overcharge-induced thermal runaway, using a 0.5C overcharge test as a representative case. The process prior to venting can be demarcated into three distinct phases based on ultrasonic signal evolution.
| Phase | Key Internal Processes | Ultrasonic Signal (RU) | External Parameters |
|---|---|---|---|
| I: Initial Overcharge | Li-plating on anode; SEI layer decomposition begins; minor gas generation. | Steady, accelerated decrease. | Voltage rises; temperature slowly increases; strain stable. |
| II: Accelerated Degradation | Severe structural distortion at cathode; massive Li-dendrite growth; significant gas accumulation. | Rapid decay to noise floor (signal lost). | Voltage peaks then drops; temperature rise accelerates; case strain increases noticeably. |
| III: Pre-Venting | Internal pressure builds from gas; reactions become self-sustaining. | Signal submerged in noise. | Strain reaches maximum; voltage falling; temperature rising rapidly. |
The most critical observation is the precocious decay of the ultrasonic signal. In the 0.5C test, the \(RU\) signal attenuated to the noise level at approximately 586 seconds after the start of overcharge. In stark contrast, the safety vent opened at 2,418 seconds. The ultrasonic warning preceded the venting event by over 30 minutes. This trend was consistent across rates: the higher the overcharge rate, the faster the ultrasonic signal was lost, but it always occurred significantly earlier than venting.
The physical mechanism is clear. The generation and accumulation of gas (H2, CO2, CO) within the sealed cell creates countless microscopic gas-solid interfaces. These interfaces act as potent scatterers for ultrasonic waves, causing severe signal attenuation long before the internal pressure is sufficient to mechanically deform the case or trigger the vent. This makes ultrasonic attenuation a direct and sensitive probe for the onset of internal failure mechanisms, rather than their final mechanical consequences.
4. Developing an Early-Warning Prognostic Algorithm
Recognizing that no single parameter is foolproof, we developed a data fusion algorithm for robust early warning. We selected six features that collectively capture the cell’s state: Voltage, Center Temperature, Ultrasonic \(U_{rms}\), Ultrasonic \(TOF\), H2 Concentration, and Center Strain. The challenge was to create a model that identifies the subtle, multivariate shift from a normal operating state to an incipient fault state.
We chose the Improved Mahalanobis-Taguchi System (IMTS) for this pattern recognition task. Its advantages include robustness with small sample sizes and the ability to handle correlated variables. Our implementation workflow was as follows:
1. Construct a Baseline Space: Using data from normal charge-discharge cycles as the “healthy” reference population.
2. Calculate Weighted Mahalanobis Distance (WMD): This distance metric quantifies how far any given sample’s feature vector deviates from the healthy baseline. We used Fisher’s criterion to assign optimal weights to each feature, enhancing the contribution of more discriminative parameters like \(U_{rms}\).
3. Define a Health Index (HI): We transformed the WMD into a normalized Health Index, where a value near 1 indicates health and a declining value indicates anomaly.
4. Set a Failure Threshold: Using statistical process control principles (3σ limits), we established a threshold \(HI_t = 0.9222\). A consecutive deviation below this threshold triggers an alarm.
We validated the algorithm using data from a separate overcharge test. The results were compelling. The algorithm’s HI index began a sharp decline well before any external parameter appeared critically abnormal. It triggered a fault warning when the cell’s SOC reached approximately 104.2% during overcharge, which corresponded to only 650 seconds into the test. As established, this was approximately 30 minutes before the safety vent opened. The warning was issued while the cell was still in Phase I/early Phase II, providing a crucial window for system intervention (e.g., selective disconnection).
The algorithm’s decision logic can be summarized as:
$$ \text{Alarm Triggered if: } HI(t) < HI_t \text{ for } t, t-1, t-2 $$
$$ \text{where } HI(t) = f_{Box-Cox}(WMD(t)), \quad WMD(t) = \sqrt{\sum_{i=1}^{n} w_i \cdot \left( \frac{x_i(t) – \mu_i}{\sigma_i} \right)^2 } $$
Here, \(x_i\) are the normalized features, \(\mu_i\) and \(\sigma_i\) are their means and standard deviations from the healthy baseline, and \(w_i\) are the Fisher-optimized weights.
5. Conclusion and Implications for Cell Energy Storage Systems
In this work, we have successfully demonstrated that in-situ ultrasonic transmission is a viable and powerful technique for internal state monitoring of large-format, prismatic LFP batteries. Our integrated multi-parameter analysis reveals clear, stage-wise progression during overcharge thermal runaway.
The core findings and their significance are:
1. Ultrasonic Signal Intensity (\(U_{rms}\) or \(RU\)) is a highly sensitive intrinsic indicator of both thermal and electrochemical state within the battery, more so than the Time-of-Flight.
2. The catastrophic attenuation of the ultrasonic signal serves as the earliest warning sign of overcharge-induced failure, preceding the opening of the safety vent by tens of minutes. This early warning is attributable to the ultrasonic sensitivity to internal gas generation, which occurs early in the failure chain.
3. Our developed IMTS-based prognostic algorithm effectively fuses ultrasonic data with conventional external measurements to provide a robust, early-warning signal for thermal runaway, enabling proactive safety management.
For operators and engineers of cell energy storage systems, this research opens a new pathway toward predictive safety. Moving beyond monitoring external symptoms to diagnosing internal health allows for a shift from reactive to preventive strategies. Implementing such acoustic sensing modules, possibly at the module or rack level, could form the backbone of a next-generation Battery Management System (BMS) capable of early fault isolation and prevention of cascading failures. Future work will focus on miniaturizing the sensing hardware, optimizing the algorithm for real-time, multi-cell processing, and validating the approach across different battery formats and under other abuse conditions to further solidify the foundation for safer grid-scale energy storage.