The headline is emphatic, but the conclusion is conditional
Lithium-ion battery thermal runaway is often understood as “a phenomenon in which temperature rises and then a fire occurs.” Accordingly, people in the field tend to think first of cell-surface temperature, rack-internal temperature, and smoke detectors. Yet, inside the cell, the electrolyte and electrode interfaces may begin to decompose and generate gases before the external surface becomes hot. If a cell remains sealed until its safety valve opens or a minute leakage path forms, electrolyte vapour and hydrogen, carbon monoxide, carbon dioxide, and hydrocarbon gases may enter the module or rack space. Properly located sensors may be able to detect this “off-gas” before a temperature alarm.
However, the headline must not be generalized literally. When gases are generated and when they reach an external sensor depend on cell chemistry, state of charge, ageing, failure cause, enclosure and vent configuration, cooling method, airflow, and sensor location. In incidents with very rapid localized heating, such as an internal short circuit, temperature or voltage changes may be effective first; and for a highly sealed cell, an external gas sensor may not see gases even when they have accumulated inside. In configurations such as immersion cooling, gases may also dissolve in or become trapped by the fluid. The accurate statement is therefore: “Under some cell and system conditions, gas signals may precede external surface temperature or smoke.” Gas detection is not a technology that replaces temperature detection; it is an additional layer that reduces gaps in the time available to existing monitoring. What matters is distinguishing normal operation from hazardous degradation with sufficient lead time. No single sensor can make that judgment on its own.
Why surface temperature can lag
Thermal runaway is a self-accelerating process in which heat-generating reactions inside a cell exceed its ability to reject heat. The issue is that sensors are usually attached not inside the cell but to the enclosure, module, or rack air. Heat generated in a small internal reaction zone must travel through the electrode stack, enclosure, and contact interfaces before reaching the external sensor. When a cooling plate is operating normally, the average surface temperature may change more slowly, and control logic that aggregates temperatures across several cells may interpret a small anomaly in one cell as noise.
The U.S. National Renewable Energy Laboratory (NREL) Automotive Battery Safety Roadmap notes that by the time an external surface sensor identifies a statistically significant temperature rise, the rate of rise may already have increased enough to progress to thermal runaway. This does not mean temperature sensors are useless. Rather, it means that the spatial and temporal gap between internal cell temperature and external measurements must be recognized, and final judgment should not be entrusted to surface temperature alone. When interpreted together with the rate of temperature change, cell-to-cell variation, cooling performance, and voltage anomalies, temperature remains one of the most important safety signals.
What gas can tell us
When thermal, electrical, or mechanical abuse begins, electrolyte volatilization and decomposition, breakdown of the solid electrolyte interphase, and side reactions between the electrodes and electrolyte may occur. Depending on the conditions, carbon dioxide and hydrogen, carbon monoxide, combustible components such as methane, ethane, and ethylene, and electrolyte-solvent vapours may appear. The presence of gas is not merely an after-the-fact signal that “a fire has occurred”; it can indicate that chemical reactions different from normal operation are proceeding inside the cell.
A 2024 Journal of The Electrochemical Society experimental study of early thermal-runaway detection using commercially available diagnostic technologies compared VOC and hydrogen sensors with high-speed impedance measurement under thermal-abuse and overcharge conditions. Under thermal abuse, high-speed impedance measurement was often the first to respond; under overcharge, the VOC sensor responded slightly before impedance. In the same paper, real-time FTIR identified gases associated with electrolyte decomposition before visible venting indications. The key point is not that “gas is always No. 1,” but that the leading signal differs by failure mode.
A study that tested the feasibility of CO2 detection, published in experimental study in eTransportation, also presented cases in which venting serves as a precursor to thermal runaway and methods for setting pack-level alarm thresholds. However, CO2 can also vary for other reasons, including people, outside air, and combustion equipment. Concluding that thermal runaway is occurring from a single concentration point can produce false alarms; the increase above background concentration, the rate of rise, and concurrent anomalies in other sensors must therefore be considered together.
Not all gas sensors observe the same thing
The phrase “one gas detector” is not sufficient for design. VOC sensors may be sensitive to electrolyte-solvent vapours, but they can also respond to humidity, cleaning agents, adhesives, and plastic emissions. Hydrogen sensors can be selective indicators of certain degradation reactions, but the amount generated varies by cell type and incident conditions, and semiconductor sensors are affected by temperature and humidity. CO and CO2 sensors likewise each have limitations related to cross-sensitivity, baseline drift, response time, and measurement range. Sensor selection must account not only for “which gases are emitted” but also for “which interfering gases are present during normal operation.”
NREL’s 2024 comparison experiment on thermal-runaway initiation methods reported that the initiation method affects heat release and gas composition when comparing external heating with a thermally triggered internal-short-circuit device. This is why a threshold obtained by externally heating one cell type with a heater in a laboratory cannot simply be transferred to a different chemistry, capacity, or actual internal fault. Before selecting a product, test data on released gases should be confirmed for, at a minimum, the cell chemistry, state-of-charge range, ageing condition, and expected failure mode. Where possible, alarm time should be validated through module- and rack-level testing in the actual installation configuration.
External sensors can only observe gases after venting
Even when gas has already been generated inside the cell, it must leave the enclosure and travel to the sensor before an external sensor can detect it. Safety-valve opening pressure, cell orientation, ducts, fan operation, rack tightness, filters, and space volume all create delay. Installing one sensor on the ceiling and placing aspirated sampling points along the expected discharge path of each rack provide entirely different performance. Location must not be determined solely by whether a single gas is lighter or heavier than air, because a hot gas mixture follows both buoyancy and ventilation flow.
Because of this limitation, alarm performance cannot be assessed only from the minimum detectable concentration on a sensor data sheet. The time from the start of cell venting until gas reaches the sensor, delays from filters and piping, fan operating state, and the alarm algorithm’s confirmation time must all be added together. Whether monitoring is maintained during a power or communications failure, whether the sensor recovers after saturation, and the calibration interval and service life must also be included in safety-function requirements.
Gas detection addresses both early warning and explosion prevention
Combustible gases released from a battery are both clues for early warning and a new hazard source. If they accumulate in an enclosed container or battery room, ignition can cause deflagration and overpressure. Sandia’s study of explosion hazards from lithium-ion battery vent gases shows the need to quantify gas flammability, flame speed, and maximum overpressure under conditions including cell chemistry and state of charge. A design that installs sensors without ventilation, isolation, and overpressure measures is therefore only half complete.
NFPA 855’s lithium-ion ESS safety annex and publicly available revision materials address a layered approach that reduces combustible-gas generation, prevents accumulation, and manages residual deflagration risk. Specific legal requirements and applicable editions must be confirmed according to the authority having jurisdiction, design standard, and installation scale, but the direction is clear. Alarms must be linked to predefined actions such as starting ventilation, stopping charging and discharging, isolating equipment, remote notification, and access control. However, automatically turning on a fan is not always safe; a fire and explosion risk assessment that includes release rate, discharge location, ignition sources, and hazardous-area classification must come first.
The most practical answer is multi-sensor defence
Sandia’s early-detection resource on battery safety explains that detecting trace vent gases during the self-heating stage and interrupting charging and discharging may make it possible to prevent thermal runaway, while explicitly stating that early detection alone cannot guarantee fire prevention. Applied to field design, this principle calls for overlapping sensors that observe different points of failure and physical quantities.
Voltage, current, insulation, and impedance monitoring identify electrical anomalies and cell imbalance.
Cell-surface, module, and coolant temperatures, together with the rate of temperature rise, identify heat generation and cooling failure.
Gas detection for VOCs, hydrogen, CO, CO2, and similar gases detects chemical degradation and venting.
Pressure and acoustic sensors can supplement detection of swelling, valve opening, and structural change.
Smoke, flame, and thermal imaging provide independent means of confirming a fire once an incident has progressed.
Ventilation status, door opening, fan failure, and communications and power integrity monitor the conditions required for sensors to function.
Alarm logic should use stages and combinations rather than a momentary value from one sensor. For example, Stage 1 records a VOC baseline deviation or abnormal voltage deviation as “caution” and increases the sampling rate. At Stage 2, charging and discharging are stopped and the relevant rack is isolated when two or more of the gas rate of rise, temperature rate of rise, and deviation across multiple cells occur simultaneously. At Stage 3, remote alarm, access control, designed exhaust, and fire-response measures are executed when smoke, high temperature, a rapid pressure rise, or a sharp increase in multiple gases is confirmed. Actual thresholds and time delays must be determined in accordance with manufacturer safety data, test results from representative installation configurations such as UL 9540A, and the applicable jurisdictional requirements.
Seven items to verify in the field
Define the chemistry, capacity, state of charge, and ageing range of the installed cells.
Define realistic failure scenarios, including overcharge, internal short circuit, external heating, and loss of cooling.
For each scenario, compare the order in which voltage, temperature, gas, pressure, and smoke signals arise using test data.
Confirm the vent-discharge location and normal and emergency ventilation flows through smoke testing or tracer-gas testing.
Set alarm criteria that account for interference factors in the normal environment, including humidity, dust, cleaning agents, and vehicle exhaust.
Functionally test whether stopping charging and discharging, rack isolation, notification, ventilation, and evacuation actually follow an alarm.
Assign responsibility for sensor calibration, functional testing, replacement intervals, data retention, and review of false alarms and missed alarms.
What it really means to see gas first
The value of the statement “The precursor to thermal runaway is gas, not temperature” is not that thermometers should be discarded. It is that signals sent by chemical changes inside the battery should not be missed before flames or dense smoke are visible. Under some conditions, off-gas provides warning time before a clear rise in external surface temperature. Under other conditions, electrical anomalies, internal temperature, pressure, or localized hot spots may appear first.
Good ESS safety design is not a competition to identify a single “first sensor.” It is the work of ensuring that different sensors cover one another’s blind spots and that detection results lead to verified responses such as shutdown, isolation, ventilation, and notification. Gas detection extends that protective network to an earlier chemical stage. Temperature and smoke detection independently confirm incident progression and fire. When this multi-sensor architecture is in place, the insight in the headline becomes an actionable early-warning strategy rather than an overstatement.