EV Battery Thermal Runaway: Why Early Warning Must Start Before the Temperature Spike

01Intake reality under uncertainty

Upon arrival, a pack may be physically compromised, deeply discharged, undocumented, or entirely isolated from a trustworthy battery-management system. Its operating history is frequently incomplete; its chemistry label missing. What appears stable at the gate can still harbor stored energy, internal damage, or a hidden manufacturing defect. The right question is therefore not simply, “Is it hot?” It is, “What combination of evidence tells us this battery is changing state, and what decision should follow?” ^[11]

Thermal runaway is best managed as an engineered risk, not an unpredictable fire. Its primary triggers are well-established: mechanical abuse, electrical abuse, external heating (including propagation from a neighboring cell), and internal defects that cause a short circuit. While factors like aging, over-discharge, and lithium plating steadily erode safety margins, the exact initiating cause in a real-world incident often remains hidden. This inherent uncertainty is precisely why relying on a single threshold or a single sensor is never enough.^[1]

Thermal runaway is a chain, not a single moment

At the cell level, thermal runaway begins when internal heat generation outpaces heat dissipation. This triggers a cascading series of exothermic reactions: solid electrolyte interphase (SEI) breakdown, electrolyte decomposition, gas generation, separator degradation, internal short-circuiting, and—depending on the cathode chemistry—oxygen release. This sequence is highly dynamic; cell chemistry, form factor, state of charge (SOC), aging, and the specific abuse mechanism all dictate the timing and severity. Crucially, a cell can undergo much of this internal destruction with little or no external flame. ^[1][2]

The absence of visible fire does not guarantee safety. Early indicators like off-gassing, venting, acoustic changes, voltage anomalies, and thermal spikes must be assessed together as linked signals of a progressive hazard rather than treated as independent alerts.

The three hazards that operators must not conflate

A failing lithium-ion cell presents at least three distinct downstream hazards. Because each requires entirely different engineering controls, none can be reliably evaluated by observing flame intensity alone. Flame intensity is a dangerously poor proxy for overall hazard severity. A low-intensity or smoldering fire can frequently mask catastrophic gas accumulation or high-consequence toxic risks.

1. External fire and heat — vented gases or particulate ejecta ignite, exposing adjacent cells, equipment, and personnel to intense heat flux and direct flame contact. This thermal exposure is the primary driver of cascade propagation across a battery module or pack.
2. Vent-gas deflagration — flammable gases accumulate inside an enclosure prior to ignition, creating a rapid, destructive overpressure spike when a spark or flame finally triggers them. In modeling this hazard from published cell-level vent data, Baird et al. emphasized that the lower flammability limit (LFL), flame speed, and maximum overpressure are the core design inputs required to engineer safe venting or suppression systems. ^[4]
3. Toxic exposure — venting and subsequent combustion release a hazardous cocktail of carbon monoxide (CO), hydrogen fluoride (HF), phosphoryl fluoride (POF₃), and other dangerous decomposition products. In confined or semi-confined spaces, the toxic dose can easily dominate the overall consequences of the failure event—even if the visible fire appears strictly limited or well-contained. ^[3]

The numbers need careful boundaries. Larsson et al. ran 39 externally ignited, propane-burner fire tests across seven commercial battery types. Burning cells emitted approximately 20–200 mg HF per Wh of nominal battery capacity. That is a measured study range—not a universal emission factor for every cell, pack or thermal-runaway event. The same paper reported 15–22 mg/Wh of POF₃, but only for one battery type at 0% SOC; no POF₃ was detected in the other tests. ^[3]

This distinction is operationally critical. While emergency planning requires conservative baselines to ensure safety, it equally demands a transparent accounting of uncertainty.

Reported gas values are highly variable and cannot be treated as fixed constants, as they are heavily influenced by a wide range of factors:

• Cell characteristics — cell format, electrolyte inventory, and state of charge (SOC).
• Environmental and test controls — the specific thermal runaway trigger method and ventilation rates.
• Data collection logistics — sampling position and the type of measurement technology utilized.

Chemistry matters—but it is not a safety verdict

Cathode chemistry influences thermal stability and heat release. High-energy layered oxides, including nickel-rich NMC and NCA families, can release oxygen during decomposition and can produce more severe thermal behaviour than LFP under comparable test conditions. In controlled overheating tests on three commercial 18650 cell types, Golubkov et al. measured cell temperatures up to 850 °C and found markedly different behaviour between metal-oxide and LFP cells. ^[2]

But a chemistry label does not determine the site decision. SOC, cell size and format, electrode loading, electrolyte, pack architecture, cooling, barriers, vent paths, ageing and physical damage can reverse a simplistic ranking. LFP’s lower thermal severity in many studies does not eliminate toxic or flammable gas. Recent near-field laser (TDLAS) measurements tracked HF directly in the vent plume of single NMC and LFP 18650 cells at 50% and 100% SOC under controlled external heating, and found chemistry- and SOC-dependent HF dynamics between the two chemistries. It is evidence of different emission dynamics—not a universal league table. ^[8]

“Solid-state” is even less suitable as one safety category. Polymer, sulfide and oxide electrolytes; lithium-metal or intercalation anodes; cathode choice; interfacial design; and stack pressure create different failure pathways. A 2026 study of sulfide all-solid-state cells demonstrated thermal runaway driven by unstable cathode–electrolyte interfaces and showed that interface engineering could raise safety thresholds. Separate 2025 work demonstrated that lithium metal can react with LFP and ignite without a flammable liquid electrolyte. The responsible conclusion is neither “solid-state is safe” nor “solid-state safety is marketing”: it is “test the specific cell and pack architecture under the relevant abuse and propagation conditions.” ^[9][10]

02Early warning is a sequence—not a temperature threshold

Surface temperature and voltage remain valuable because they are proven, available and easy to integrate. They are also incomplete. Internal reactions can develop before heat reaches an external sensor, and a voltage signal may arrive late or be unavailable when a damaged pack is disconnected. Gas, pressure, strain and acoustic signals can reveal earlier changes, but their usefulness depends on sensor placement, airflow, cell design and the failure mechanism.

The strongest recent evidence supports a qualified—not absolute—case for earlier gas warning. Gardner et al. tested commercial NCA, NMC and LFP cells from 10% to 100% SOC and detected trace hydrogen before thermal runaway across the tested formats and abuse conditions. Stopping the applied abuse at hydrogen detection prevented runaway in those experiments. ^[5] In an energy-storage-cabin study, detector placement materially affected response; a top-mounted H₂ detector warned 145 seconds before thermal runaway in the reported setup. ^[6]

Pressure and strain add another layer. An implanted optical-fibre sensor recorded internal pressure features associated with safety venting and runaway initiation and produced an alert before venting in controlled 18650 tests. ^[7] These are promising results, not proof that every facility can buy a sensor and obtain a guaranteed number of minutes. Industrial deployment must still address drift, poisoning, false alarms, survivability, calibration, airflow and maintenance over the equipment life.

HF itself may also provide a warning signal after safety venting. Near-field TDLAS experiments detected HF immediately after vent opening and before runaway onset in the tested NMC and LFP cells. The same study explains why conventional sampling can underestimate HF: the gas is highly adsorptive and corrosive, and transfer lines can lose it. ^[8] That makes HF valuable for research and hazard awareness, but it also argues for careful engineering rather than an unqualified “HF sensor equals earliest alarm” claim.

03From detection to a defensible intake decision

A warning signal only creates value if it changes what the facility does. For a recycler, the closed-loop response is different from an in-vehicle design. A damaged incoming pack may not have functioning contactors, a pyrofuse command path or reliable auxiliary power. The system therefore needs to connect evidence to site actions that remain available when the battery cannot help protect itself.

1. Connect — Bring together thermal, gas/HF, acoustic, vision, 3D inspection, battery diagnostics, BMS data when available, operator observations and battery history.
2. Fuse — Build one assessment case with signal quality, timing, trend, context and uncertainty—not a row of disconnected alarms.
3. Decide — Translate the case into a defined route: proceed to manual or automated processing, hold for specialist inspection, stop handling, or move to an approved quarantine path.
4. Act — Trigger the facility’s approved controls—such as stopping movement or processing, ventilation, an exclusion zone and escalation—at the appropriate alert tier.
5. Learn — Record the event, operator decision and downstream outcome so thresholds and routing rules can be reviewed against real process evidence.

This is the operating logic behind DEXOLYTA Guardian and Intake: use existing data sources where possible, combine physical signals with battery and process context, and support consistent hazard and routing decisions across shifts and sites. DEXOLYTA is a decision-support and early-notification layer. It should complement—not replace—certified safety systems, emergency procedures, competent risk assessment and regulatory obligations.

04Closing implication

The most defensible thermal-runaway strategy is not “find the perfect sensor” or “choose the safest chemistry.” It is to build a chain of evidence that survives uncertainty: complementary sensing, signal fusion, explicit alert tiers, pre-agreed actions and an event record that improves future decisions.

References

[1]Feng, X. et al. “Thermal runaway mechanism of lithium ion battery for electric vehicles: A review.” Energy Storage Materials 10, 246–267 (2018). doi:10.1016/j.ensm.2017.05.013

[2]Golubkov, A. W. et al. “Thermal-runaway experiments on consumer Li-ion batteries with metal-oxide and olivin-type cathodes.” RSC Advances 4, 3633–3642 (2014). doi:10.1039/C3RA45748F

[3]Larsson, F. et al. “Toxic fluoride gas emissions from lithium-ion battery fires.” Scientific Reports 7, 10018 (2017). doi:10.1038/s41598-017-09784-z

[4]Baird, A. R. et al. “Explosion hazards from lithium-ion battery vent gas.” Journal of Power Sources 446, 227257 (2020). doi:10.1016/j.jpowsour.2019.227257

[5]Gardner, D. W. et al. “Mitigating lithium-ion cell thermal runaway via selective trace H₂ sensing.” Cell Reports Physical Science 6(10), 102859 (2025). doi:10.1016/j.xcrp.2025.102859

[6]Shi, S. et al. “Hydrogen gas diffusion behavior and detector installation optimization of lithium ion battery energy-storage cabin.” Journal of Energy Storage 67, 107510 (2023). doi:10.1016/j.est.2023.107510

[7]Mei, W. et al. “Operando monitoring of thermal runaway in commercial lithium-ion cells via advanced lab-on-fiber technologies.” Nature Communications 14, 5251 (2023). doi:10.1038/s41467-023-40995-3

[8]Alsewailem, A. F. et al. “Near-field quantification of hydrogen fluoride emissions during Li-ion battery thermal runaway using laser absorption spectroscopy.” Journal of Power Sources 667, 239221 (2026). doi:10.1016/j.jpowsour.2025.239221

[9]Wu, Y. et al. “Electrochemical initiation and chemical reaction cascades in dual-stage thermal runaway in sulfide-based all-solid-state batteries.” Nature Communications 17, 2928 (2026). doi:10.1038/s41467-026-69472-3

[10]Bertrand, M. et al. “Unveiling the thermite-driven lithium fire ignition in solid-state batteries.” Joule 9(5), 101953 (2025). doi:10.1016/j.joule.2025.101953

[11]U.S. Environmental Protection Agency. “Lithium-Ion Battery Recycling Frequently Asked Questions.” Updated 6 August 2025. EPA guidance