Hydrogen: the first alarm for lithium-ion thermal runaway

How trace H₂ detection delivers early warning, prevention and aftermath mitigation across cells, packs and battery rooms.

Abstract. Thermal runaway is the defining safety hazard of lithium-ion batteries. Because cells almost always vent gas before they fail catastrophically, detecting that vent is the key to intervening in time. This review synthesizes recent peer-reviewed evidence — principally Gardner et al. (Cell Reports Physical Science, 2025) and its cited literature — with battery-room safety and compliance guidance to show why trace hydrogen is the most actionable early-warning marker for cell failure, and how hydrogen sensing mitigates both the onset and the aftermath of thermal runaway across electric vehicles and stationary energy storage.

Why thermal runaway is the problem to solve

Lithium-ion batteries now anchor electric mobility and grid-scale energy storage, but their dominant failure mode — thermal runaway — remains the central safety challenge of the technology. A single defective or abused cell can self-heat, vent, ignite and cascade to its neighbors in a propagation event that destroys packs, battery rooms and buildings. As deployments scale into hundreds of gigawatt-hours, regulators have responded: UNECE Global Technical Regulation No. 20 and the United States FMVSS No. 305a now require electric vehicles to warn occupants before thermal runaway propagation reaches the cabin. Meeting that mandate — and protecting stationary battery energy storage systems (BESS) — depends on detecting trouble early enough to act.

A growing body of evidence points to one signal that arrives earlier and more reliably than temperature, voltage or pressure: trace hydrogen. This article explains the chemistry behind that signal, the experimental proof that it is both early and actionable, and how the same marker underpins battery-room safety codes and mitigation strategy.

The anatomy of thermal runaway

Thermal runaway is a self-accelerating chain of exothermic reactions. A cell stressed by high temperature, overcharge, mechanical deformation or an internal defect begins to degrade: the solid-electrolyte interphase breaks down, the separator softens and can short, the charged cathode decomposes and releases oxygen, and the electrolyte reacts and decomposes. Each step adds heat, which accelerates the next. Internal pressure climbs until the cell housing fails in a first venting event. If nothing intervenes, the temperature continues to rise until the cell goes into full runaway and can ignite or rupture, propagating to adjacent cells in a dangerous cascade.

Two factors raise the stakes. Aged cells are more vulnerable because cumulative degradation lowers the threshold for failure, and cells at higher state of charge (SoC) are both more reactive and more energetic when they fail — delithiated NMC cathodes, in particular, become markedly more reactive with the electrolyte above roughly 80 % SoC. Crucially, between the onset of abuse and catastrophic failure there is a window. The first venting event is a near-ubiquitous feature of cells that will eventually enter runaway, and it is the opportunity that gas detection is designed to seize — especially for latent failures that build up over time and are hard to catch with conventional diagnostics.

The vent-gas signature — and why hydrogen

The first gases released at venting are primarily carbon dioxide, carbon monoxide, ethylene, vaporized electrolyte and hydrogen. Among these, hydrogen is the standout early-warning marker for three reasons set out by Gardner et al.: it is distinctive to cell venting and essentially absent from ambient air, so false positives are rare; it is a fundamental product generated across a wide variety of battery chemistries; and it has a high diffusion coefficient, so it travels quickly to a sensor and rises buoyantly toward detectors mounted overhead.

Hydrogen forms through several routes that do not depend on the cathode: electrolysis of trace water and of water produced by electrolyte decomposition; reaction of lithium dendrites with the polyvinylidene fluoride binder; and thermal decomposition of carbonate solvents. Using graphite-graphite symmetric cells, Xueqin and Lei showed hydrogen is generated regardless of cathode material, pointing to an electrolyte- and anode-driven origin. This chemistry-agnostic mechanism is exactly why hydrogen appears across NCA, NMC and LFP cells. Notably, Jin et al. demonstrated that capturing hydrogen can flag micro-scale lithium dendrite formation as an early safety warning.

There is one catch: hydrogen is present at relatively low concentration compared with other vented species. Conventional thermal-conductivity hydrogen sensors have detection limits of thousands of parts per million and can report “no hydrogen” after a genuine vent; only with more sensitive tools is hydrogen invariably detected. The practical requirement, therefore, is trace-level detection — a sensitivity floor well below 100 ppm.

Why conventional sensors fall short

Existing battery-health sensors offer limited prognostic power. Voltage and current sensing degrade in the parallel and series configurations used in real packs: a single failing cell among many in parallel barely moves the measured pack voltage, and in cylindrical cells the current-interrupt device (CID) suppresses the voltage signature of venting entirely. Temperature sensing is only reliable if every cell is individually instrumented, which is impractical at scale, and case temperature lags the internal event. Pressure sensing is often blind to the vent altogether.

The experimental data make this concrete. In Gardner et al., the measured voltage did not change at first venting, and pressure spikes were frequently below 0.1 hPa — effectively invisible to those channels — while hydrogen was already rising. Gas sensing, by contrast, catches the near-ubiquitous vent event and is uniquely suited to latent, abuse-driven failures.

The evidence: trace H₂ as an early, actionable signal

Gardner et al. tested a hydrogen chemically sensitive field-effect transistor (H₂-CSFET) built on a palladium sensing layer. Hydrogen produces a work-function shift specific to the H–H bond, changing the transistor threshold voltage and channel current; a 60 °C microheater rejects humidity and other adsorbates, giving selectivity over CO₂, CO, hydrocarbons and electrolyte vapor (C–H bonds are not cleaved at the operating temperature). The reported limit of detection is below 1 ppm with roughly 10 % accuracy from 30–1000 ppm, at very low power (about 1 µW for the transistor plus a microheater that can be duty-cycled down from about 50 mW).

The sensors were challenged against commercial cells spanning formats and chemistries — a 5 Ah cylindrical NCA 21700, a 72 Ah NMC pouch and a 120 Ah LFP prismatic cell — at 10–100 % SoC under overheating and overcharge abuse. The headline results were consistent and striking:

• Early warning: in cells driven all the way to runaway, hydrogen provided 23.9 minutes of warning under 6 °C/min overheating and 3.6 minutes under 1C overcharge.
• Universally detected: hydrogen appeared in the vent gas in every test — across all chemistries, formats and states of charge — at roughly 100–400 ppm at venting for the NCA cells, with no strong SoC correlation.
• Most reliable channel: hydrogen was a better indicator of imminent runaway than pressure, voltage or case temperature in every case.
• Actionable: when the stressor was stopped upon hydrogen detection, thermal runaway was prevented in all tested cases.

That last point is the bridge from warning to mitigation: the signal is early enough that intervention actually works. The study also surfaced useful nuance. Higher-SoC cells were harder to cool once the stressor was removed, consistent with the greater reactivity of delithiated NMC cathodes; the overcharge tests showed a voltage plateau near 5.26 V indicative of lithium-dendrite formation before venting; and the LFP prismatic cell vented at the highest case temperature (142 °C) with the largest pressure spike (~60 hPa). The sensor specification that emerges — echoed by stationary-storage studies — is a detection limit below about 100 ppm, near-zero false positives to ambient and automotive gases, low key-off power, and a service life exceeding that of the battery pack.

Table 1. Early-warning behavior observed across cell types (Gardner et al.).

From cell to room: mitigating the aftermath

Early warning at the cell answers half the problem. The other half plays out at the scale of a battery room or BESS enclosure, where the “aftermath” of a vent is gas accumulation, deflagration and fire propagation. Here hydrogen serves a second, well-established safety role detailed in the guidance that hydrogen-sensor manufacturers provide for standby-power and energy-storage installations.

Hydrogen is flammable from 4 % to 74 % by volume in air; its lower flammable limit (LFL) is 4 vol % (40,000 ppm). Codes typically require keeping room concentration below 25 % of the LFL — a 1 vol % alarm and ventilation threshold — to preserve a wide safety margin. Detecting trace hydrogen long before that threshold lets a facility ventilate, de-energize, suppress or evacuate while the atmosphere is still inert. That early margin is the practical value of trace sensitivity.

The standards landscape

Technical readers will recognize the framework that governs these installations. IEC 62485-2 sets safety requirements for stationary secondary batteries, including ventilation sized to dilute hydrogen evolution, and is the backbone of European battery-room compliance. EN 50604 covers lithium battery safety and is applied alongside it in EU risk assessments. In North America, NFPA 855 governs the installation of stationary energy storage and requires flammable-gas detection and explosion control in certain indoor installations, while UL 9540A is the test method that characterizes how thermal runaway propagates from cell to module to unit; UL’s Fire Safety Research Institute identifies off-gas sensing (hydrogen, CO, HF, hydrocarbons) as a recognized early-warning approach. For vehicles, UNECE GTR 20 and FMVSS 305a mandate occupant warning before propagation.

European compliance carries an added structural nuance. Under the principle of subsidiarity, local jurisdictions set the specifics — ventilation rates, spill containment — within EU-wide guidelines, so requirements are layered across local, national and EU levels, and Authorities Having Jurisdiction commonly fold IEC 62485-2 into their inspections. Insurers add a further layer: loss-prevention requirements that frequently exceed code and condition coverage on compliance.

A layered mitigation architecture

Effective battery-room safety is best framed as a multi-layered system in which hydrogen detection fires first, because it gives the earliest actionable signal:

• Gas detection: trace hydrogen monitoring to flag a vent or off-gassing before accumulation becomes hazardous.
• Ventilation: extraction sized to dilute hydrogen and prevent it from reaching flammable concentrations.
• Fire suppression: typically water-based systems that cool cells and limit propagation.
• Containment and thermal management: electrolyte-resistant flooring and spill containment, plus early thermal-runaway detection and cell isolation.
• Human and procedural controls: PPE, training, signage and disciplined record-keeping for compliance and continuous improvement.

Designing a hydrogen-based safety strategy

For implementers, several design choices determine whether a hydrogen-based layer actually delivers. Sensitivity should target a detection limit below roughly 100 ppm so the system catches first-vent hydrogen rather than the thousands-of-ppm floor of legacy thermal-conductivity units that can miss a real event. Selectivity is equally important: immunity to CO₂, CO, hydrocarbons, electrolyte vapor and humidity prevents the nuisance trips that erode operator trust and trigger costly shutdowns. Placement should exploit hydrogen’s buoyancy and high diffusivity — detectors mounted high and near likely vent paths respond fastest, and modeling work has optimized detector siting in both EV packs and storage cabins.

Finally, integration is what converts a reading into protection. The prevention result in Gardner et al. depended on automatically removing the stressor the moment hydrogen was detected; tying detection into the battery management system (BMS) or facility controls to cut charging, start ventilation and raise an alarm is therefore essential. In vehicles the sensor must also survive key-off without draining auxiliary batteries and outlast the pack, while in hard-to-access battery rooms maintenance-free operation is a practical necessity.

Outlook

Hydrogen’s appeal is a rare combination among candidate markers: it is early, near-universal across chemistries and distinctive against ambient air. The CSFET platform is also extensible — the same architecture can target electrolyte volatile organics for corroboration, or hydrogen sulfide for emerging lithium-sulfur chemistries. Open questions remain, including behavior in fully assembled commercial packs, the generality of prismatic-cell results, and the absence of prescribed detection setpoints in standards such as NFPA 855. But the direction is clear. As fire codes push toward pre-ignition warning and vehicle regulations mandate occupant alerts, trace hydrogen detection is moving from laboratory demonstration to a practical, scalable layer of lithium-ion safety — catching failure at the first breath of gas, when there is still time to act.

Hydrogen is, in fact, already a key gas in electrical diagnostics: in dissolved gas analysis of transformers it is one of the earliest signs of an incipient fault. Amperis supports its customers in selecting measurement and diagnostic equipment, as well as in the maintenance and testing of batteries for industry and energy storage. Working on a battery safety or monitoring project? Contact our engineers and we will advise you on the most suitable equipment.

References

The following sources informed this review:

1. Gardner, D.W., Charles, G., Nguyen, T., Javey, A., and Fahad, H.M. (2025). Mitigating lithium-ion cell thermal runaway via selective trace H₂ sensing. Cell Reports Physical Science 6, 102859. https://doi.org/10.1016/j.xcrp.2025.102859
2. Golubkov, A.W., et al. (2018). Thermal runaway of large automotive Li-ion batteries. RSC Advances 8, 40172–40186. https://doi.org/10.1039/C8RA06458J
3. Feng, X., Ouyang, M., Liu, X., Lu, L., Xia, Y., and He, X. (2018). Thermal runaway mechanism of lithium ion battery for electric vehicles: A review. Energy Storage Materials 10, 246–267. https://doi.org/10.1016/j.ensm.2017.05.013
4. Larsson, F., and Mellander, B.-E. (2014). Abuse by external heating, overcharge and short circuiting of commercial Li-ion battery cells. J. Electrochem. Soc. 161, A1611–A1617. https://doi.org/10.1149/2.0311410jes
5. Galushkin, N.E., Yazvinskaya, N.N., and Galushkin, D.N. (2019). Mechanism of gases generation during lithium-ion batteries cycling. J. Electrochem. Soc. 166, A897–A908. https://doi.org/10.1149/2.0041906jes
6. Jin, Y., et al. (2020). Detection of micro-scale Li dendrite via H₂ gas capture for early safety warning. Joule 4, 1714–1729. https://doi.org/10.1016/j.joule.2020.05.016
7. Xueqin, Y., and Lei, Y. (2021). Exploration of the cause of hydrogen generation in NCM lithium-ion batteries. Energy Storage Science and Technology 10, 1693–1698. https://doi.org/10.19799/j.cnki.2095-4239.2020.0222
8. Spinner, N.S., Hinnant, K.M., Tuttle, S.G., and Rose-Pehrsson, S.L. (2016). Lithium-ion battery failure: effects of state of charge and packing configuration. Naval Research Laboratory. https://apps.dtic.mil/sti/pdfs/AD1016095.pdf
9. Rowden, B., and Garcia-Araez, N. (2020). A review of gas evolution in lithium ion batteries. Energy Reports 6, 10–18. https://doi.org/10.1016/j.egyr.2020.02.022
10. Nilsson, E.J.K., and Ahlberg Tidblad, A. (2024). Gas emissions from lithium-ion batteries: a review of experimental results and methodologies. Batteries 10, 443. https://doi.org/10.3390/batteries10120443
11. Cui, Y., et al. (2023). Thermal runaway early warning and risk estimation based on gas production characteristics of different types of lithium-ion batteries. Batteries 9, 438. https://doi.org/10.3390/batteries9090438
12. Preger, Y., Torres-Castro, L., Rauhala, T., and Jeevarajan, J. (2022). Perspective — On the safety of aged lithium-ion batteries. J. Electrochem. Soc. 169, 030507. https://doi.org/10.1149/1945-7111/ac53cc
13. Baird, A.R., Archibald, E.J., Marr, K.C., and Ezekoye, O.A. (2020). Explosion hazards from lithium-ion battery vent gas. J. Power Sources 446, 227257. https://doi.org/10.1016/j.jpowsour.2019.227257
14. Gachot, G., et al. (2012). Thermal behaviour of the lithiated-graphite/electrolyte interface through GC/MS analysis. Electrochimica Acta 83, 402–409. https://doi.org/10.1016/j.electacta.2012.08.016
15. Jung, R., Metzger, M., Maglia, F., Stinner, C., and Gasteiger, H.A. (2017). Oxygen release and its effect on cycling stability of NMC cathode materials. J. Electrochem. Soc. 164, A1361–A1377. https://doi.org/10.1149/2.0021707jes
16. Lyu, N., et al. (2023). Hydrogen gas diffusion behavior under fault conditions and detector installation optimization of electric vehicles. Process Safety and Environmental Protection 175, 565–574. https://doi.org/10.1016/j.psep.2023.05.075
17. Shi, S., et al. (2023). Hydrogen gas diffusion behavior and detector installation optimization of lithium-ion battery energy-storage cabin. J. Energy Storage 67, 107510. https://doi.org/10.1016/j.est.2023.107510
18. Huang, W., et al. (2023). Early warning of battery failure based on venting signal. J. Energy Storage 59, 106536. https://doi.org/10.1016/j.est.2022.106536
19. Torres-Castro, L., et al. (2024). Early detection of Li-ion battery thermal runaway using commercial diagnostic technologies. J. Electrochem. Soc. 171, 020520. https://doi.org/10.1149/1945-7111/ad2440
20. Wang, Z., Zhu, L., Liu, J., Wang, J., and Yan, W. (2022). Gas sensing technology for the detection and early warning of battery thermal runaway: a review. Energy & Fuels 36, 6038–6057. https://doi.org/10.1021/acs.energyfuels.2c01121
21. Jiang, L., et al. (2019). Overcharge behavior and early warning analysis of NMC/C lithium-ion battery with high capacity. J. Electrochem. Soc. 166, A1055–A1062. https://doi.org/10.1149/2.0571906jes
22. Fahad, H.M., Gupta, N., Han, R., Desai, S.B., and Javey, A. (2018). Highly sensitive bulk silicon chemical sensors with sub-5 nm thin charge inversion layers. ACS Nano 12, 2948–2954. https://doi.org/10.1021/acsnano.8b00580
23. Gupta, N., Fahad, H.M., Amani, M., Song, X., Scott, M., and Javey, A. (2019). Elimination of response to relative humidity changes in chemical-sensitive field-effect transistors. ACS Sensors 4, 1857–1863. https://doi.org/10.1021/acssensors.9b00637
24. UNECE (2018). Global Technical Regulation No. 20: Electric Vehicle Safety (EVS). United Nations Economic Commission for Europe; cf. FMVSS No. 305a. https://unece.org/fileadmin/DAM/trans/main/wp29/wp29wgs/wp29gen/wp29registry/ECE-TRANS-180a20e.pdf
25. Industry technical guidance (2023). Best practices and guidelines for battery safety and compliance. https://h2scan.com/best-practices-and-guidelines-for-battery-safety-and-compliance/
26. Industry technical guidance (2024). Ensuring battery room safety in Europe: navigating standards and best practices. https://h2scan.com/ensuring-battery-room-safety-in-europe-navigating-standards-and-b