Thermal Runaway — The 5 Stages Your BMS Is Racing Against
You know thermal runaway is bad. But knowing it's a five-stage cascade with specific temperature windows, gas signatures, and shrinking intervention opportunities changes how you think about BMS design, pack architecture, and chemistry selection entirely.
Table of Contents
- Why This Is a Race, Not an Event
- The 5 Stages — Overview
- Stage 1 — SEI Breakdown Begins
- Stage 2 — The Anode Starts Reacting
- Stage 3 — The Point of No Return
- Stage 4 — It's on Fire
- Stage 5 — Propagation
- What the BMS Can and Cannot Do
- LFP vs NMC — Why Chemistry Decides the Outcome
- The Three Triggers
- Early Warning Signals Worth Watching
- SOC and Risk — The Underappreciated Link
- Key Takeaways
This is the Intermediate level of the EVPulse Thermal Runaway series. It assumes you understand basic BMS architecture, know what Li-ion cells consist of, and are familiar with terms like SOC, SEI, and electrode chemistry at a conceptual level. For the deep electrochemical breakdown with temperature windows and heat generation rates, read the Expert article.
Why This Is a Race, Not an Event
Most people — including many engineers early in their careers — think about thermal runaway as a binary: the battery is fine, or it's on fire. This framing leads to protection systems that only trigger at the last possible moment, right when it's already too late to do anything useful.
The reality is a sequence. Five distinct stages. Each has its own temperature window. Each has its own characteristic signals. And critically, each has a different intervention opportunity — ranging from "plenty of time" in Stage 1 to "absolutely nothing" in Stage 4.
The job of a well-designed BMS is not to detect thermal runaway. It's to detect the conditions that precede thermal runaway — and act while there's still something to act on.
If your BMS only responds to absolute temperature thresholds, it's fundamentally designed wrong. Here's why.
The 5 Stages — Overview
Notice how the temperature windows overlap slightly — Stages 2 and 3 can start almost simultaneously in a rapid-onset scenario like a nail penetration or severe overcharge. The stages are analytically distinct but not always temporally separated.
Stage 1 — SEI Breakdown Begins
The SEI — solid-electrolyte interphase — is a thin film that forms naturally on the graphite anode during the first few charge cycles. It's essential. Without it, the electrolyte would react directly and continuously with the graphite. The SEI lets lithium ions pass through while blocking that direct reaction.
The problem: the SEI is metastable. It's not a permanent material. At around 90–120°C, it starts to decompose exothermically — meaning the decomposition itself generates heat. Not a lot — maybe 40–120 joules per gram of anode material — but enough to push the cell slightly hotter, which speeds up the decomposition, which generates more heat.
This is the first self-reinforcing loop. It's slow. It's subtle. And it's the last point where the system can realistically be pulled back.
The hardest part of Stage 1 detection is sensor physics. Cell cores experiencing SEI decomposition can be at 110°C while the nearest surface NTC temperature sensor reads 65–75°C. By the time the sensor hits a typical 80°C protection threshold, Stage 2 may already be starting. Temperature sensor placement inside modules is not a cosmetic decision — it directly determines whether Stage 1 is detectable in time.
What intervention looks like at Stage 1:
- Ramp cooling to maximum immediately
- Cut charge/discharge current to zero
- Alert operator/driver to the anomaly
- Monitor for dT/dt stabilisation
If cooling is effective and the heat source is removed, Stage 1 can arrest. The SEI will be partially damaged and the cell will be degraded, but the cascade stops.
Stage 2 — The Anode Starts Reacting
Now things start accelerating. The temperature has passed the SEI decomposition window and the lithiated graphite anode is hot enough to react directly with the liquid electrolyte — specifically with the carbonate solvents (ethylene carbonate, dimethyl carbonate) that make up the electrolyte.
These reactions produce ethylene gas and significantly more heat — roughly 100–400 J/g, several times the Stage 1 output. The rate is strongly temperature-dependent; every 10°C increase roughly doubles the reaction rate.
Gas generation becomes meaningful at Stage 2. Pressure inside the cell rises. On well-designed prismatic and cylindrical cells, pressure relief vents open at this stage — releasing hot flammable gas into the module enclosure. If there are no gas sensors in the pack, this venting goes completely undetected.
What the BMS sees at Stage 2:
- Surface temperature rising rapidly, dT/dt clearly above predicted I²R baseline
- Cell voltage beginning to drop as internal resistance changes
- Pressure vent opening (detectable with pressure or acoustic sensors, rare in production packs)
- Gas sensors (if fitted): CO₂ and hydrocarbon detection
Stage 3 — The Point of No Return
This is the stage that determines whether you have a venting event or a fire. The cathode material — the positive electrode — begins thermally decomposing. The critical consequence is oxygen release.
For NMC chemistry, this decomposition starts as low as 170°C. The cathode releases oxygen directly into the cell environment, where it meets the flammable electrolyte vapours that have been accumulating since Stage 2.
The cell is now producing its own oxidiser. It no longer needs atmospheric oxygen to sustain combustion. This is the fundamental reason EV fires behave so differently from conventional fires — and why you cannot smother them.
Stage 3 is the point of no return because it fundamentally changes what is burning. Before Stage 3, you have an overheated battery. After Stage 3, you have an internally oxygenated combustion reactor. External intervention cannot stop it — only contain it.
What the BMS can do at Stage 3 (which is very little):
- Contactors should already be open from Stage 2 detection
- Broadcast highest-priority fault on CAN
- Log the event with full timestamp and cell-level data for investigation
- That's it. Electrical intervention is irrelevant — the cell is generating heat through pure chemistry, not electrical current.
Stage 4 — It's on Fire
The electrolyte vapours ignite. The first visible flame appears. At this temperature, the reactions are fully self-sustaining and generating heat faster than any conceivable cooling system could remove.
The combustion products at Stage 4 include CO₂, CO, unburned hydrocarbons, smoke particulate — and critically, HF (hydrogen fluoride).
HF is produced from the decomposition of LiPF₆, the salt used in most liquid electrolytes. It is toxic at concentrations above 3 ppm, and at high concentrations it causes chemical burns that penetrate deeply into tissue. A single 100 Ah prismatic cell can release 15–30 mg of HF during full combustion. First responders approaching a burning EV pack without specialised respiratory protection face genuine acute risk.
HF cannot be detected by smell reliably at dangerous concentrations. Standard N95 and FFP2 masks provide no protection against HF gas. This is not a theoretical risk — it is the primary reason EV fires are hazardous beyond the visual drama of the flames. Pack enclosure design should direct vent gases away from vehicle occupants and external observers.
Stage 5 — Propagation
The cell reaches peak temperature. The separator has long since melted (PE/PP separators melt at 130–170°C — they were gone in Stage 2). The cell case ruptures or vents catastrophically, ejecting hot gas, electrolyte, and burning material.
The question for pack design now is: does this stop at one cell, or does it spread?
Propagation from one cell to its neighbours happens through three mechanisms simultaneously:
Conduction — Heat travelling through busbars, cell holders, and any direct cell-to-cell contact. The fastest path in tightly packed modules.
Convection — Hot gases venting from the failed cell flow through the module enclosure and pre-heat adjacent cells.
Radiation — At 600–900°C, a cell radiates enormous amounts of thermal energy. Neighbouring cells within a few centimetres absorb this directly.
What the BMS Can and Cannot Do
This is the most important section for anyone designing protection systems.
| What the BMS CAN prevent | What the BMS CANNOT do |
|---|---|
| Overcharge (voltage cutoff) | Stop Stage 3+ chemical reactions |
| Overtemperature charging (current reduction) | Generate oxygen inside a cell |
| Overcurrent (current limiting) | Cool a cell below 500°C |
| External short circuit (contactor open) | Prevent propagation through mechanical design |
| Identify anomalous dT/dt before Stage 2 | Detect Stage 1 with standard NTC placement |
The pattern is clear: the BMS is preventive up to Stage 2 and strictly reactive (isolation + alert) from Stage 3 onwards. Any BMS firmware that attempts "management" logic during Stage 3+ is dangerous — it creates false confidence and introduces failure modes without adding safety.
LFP vs NMC — Why Chemistry Decides the Outcome
The chemistry choice is arguably the single most impactful safety decision in pack design. Not the BMS. Not the cooling system. The chemistry.
| LFP | NMC 811 |
|---|---|
| Thermal runaway onset: 270–310°C | Thermal runaway onset: 170–210°C |
| Heat of reaction: ~100–200 J/g | Heat of reaction: ~500–900 J/g |
| Oxygen release: Minimal | Oxygen release: Significant |
| Propagation risk: Lower | Propagation risk: Higher |
| Energy density: ~150 Wh/kg | Energy density: ~250 Wh/kg |
| Cycle life at 80% DOD: 3,000+ | Cycle life at 80% DOD: 1,200–1,500 |
The 100°C difference in onset temperature is not cosmetic. At 270°C, a properly designed BTMS (Battery Thermal Management System) still has a realistic chance of pulling heat away fast enough to arrest the cascade. At 170°C, the window is much narrower.
The 5× difference in heat of reaction means that when an NMC cell fails, it dumps roughly 5× the energy into neighbouring cells compared to LFP. This is why NMC propagation is faster and harder to contain.
The Three Triggers
External heat source drives the cell into Stage 1 directly. Common scenarios: fire exposure from external source, adjacent module already in runaway, BTMS failure in 45°C+ ambient during DC fast charging, cell group thermal isolation failure. Detection: dT/dt divergence from load model, differential temperature across module.
Overcharge is the most preventable trigger — and still the most common in field failures involving third-party or modified charging equipment. Overcurrent generates I²R heat at a rate that can reach Stage 2 in seconds on severe overcurrent events. Detection: voltage above HVC threshold, current above limit, impedance rise during charging. Protection: robust voltage sensing with redundancy, current sensor accuracy verification.
Internal short circuit from mechanical abuse (crash, road debris impact, crush) or manufacturing defect (metallic particle contamination, separator damage) is the most dangerous trigger because it bypasses all electrical protection. A hard internal short can dump the full cell energy as heat in under a second. BMS cannot detect it in advance — only respond after the thermal event begins. Mitigation: mechanical protection (underbody armor, structural frame), post-accident inspection mandates.
Early Warning Signals Worth Watching
These four signals, monitored together, give the best chance of detecting thermal runaway precursors before Stage 3:
1. dT/dt — Rate of Temperature Rise Absolute temperature is a lagging indicator. Rate of temperature rise above what the current load predicts is an early indicator. A cell rising 2°C/minute above its neighbours during rest is anomalous. Good BMS firmware monitors this continuously.
2. Voltage Divergence at Rest An internal short causes self-discharge faster than neighbouring cells. A cell that drops 50–100 mV below pack average during an overnight park (no load, no charge) is telling you something is wrong internally. This can provide hours to days of warning for slow-developing failures.
3. Impedance Anomaly AC impedance measurement (electrochemical impedance spectroscopy at a simplified level) can detect changes in charge-transfer resistance that precede temperature rise. Not standard in production BMSs today, but increasingly feasible with modern MCUs.
4. Gas Detection A CO₂ or VOC sensor inside the pack enclosure costs under ₹500. It detects Stage 2 gas venting before any temperature sensor has triggered. Lead time of 2–8 minutes before visible fire. This is the most underdeployed safety technology in Indian EV packs right now.
Of these four signals, gas detection has the best cost-to-warning-time ratio by a significant margin. The reason it's not standard has nothing to do with engineering logic — it's inertia. Pack designers who have been building packs for ICE-adjacent industries simply haven't standardised it yet.
SOC and Risk — The Underappreciated Link
High State of Charge is consistently the highest-risk state for thermal runaway initiation and severity. This has direct practical implications:
Set your charge limit to 80–90% unless you specifically need full range. Most drivers don't. The risk reduction from avoiding the 90–100% window is real.
Park at 30–50% SOC if leaving the vehicle for more than a week. High SOC + long time = maximum period for any slow-developing internal fault to progress.
Do not charge immediately after any collision, even minor. Get the battery inspected first.
Most EVs taper charge rate above 80% automatically. This is thermal management, not just battery longevity. Don't try to override it.
Reader Poll
Poll: Which thermal runaway precursor technology should be mandatory in new EV packs?
Tap to vote
Key Takeaways
- Thermal runaway is five stages, each with distinct temperature windows and intervention windows. Stages 1–2 are recoverable. Stage 3 is the point of no return.
- The BMS is a prevention tool, not a runaway management tool. Isolation + alert is all it can do at Stage 3+.
- LFP is meaningfully safer than NMC — 100°C higher onset, 5× lower heat of reaction, slower propagation.
- High SOC is the highest-risk state. The combination of high SOC + mechanical damage is the most dangerous scenario.
- Gas detection inside the pack enclosure is the highest-value early warning technology that is systematically underused.
- Propagation is a pack mechanical design problem. No BMS logic compensates for cells packed with no thermal barriers and no vent path management.
Resources and References
All references verified as of May 2025. Includes peer-reviewed papers and standards documents. Paywalled papers include DOI for library access.
Standards and Regulations
- AIS-156 Phase 2 (2023) — Automotive Industry Standard for Electric Power Train Vehicles. Ministry of Road Transport and Highways / BIS. https://morth.nic.in
- UN ECE R100 Revision 3 (2021) — Uniform provisions for EV approval, including propagation and thermal runaway warning requirements. https://unece.org/transport/documents/2021/03/standards/un-regulation-no-100-rev3
- IEC 62660-2 — Reliability and abuse testing for secondary lithium-ion cells in electric road vehicles.
- ISO 12405-4:2018 — Electrically propelled road vehicles: Test specification for lithium-ion traction battery packs and systems — Part 4: Performance testing.
Research Papers
- Feng, X., Ouyang, M., Liu, X., Lu, L., Xia, Y., & He, X. (2018). Thermal runaway mechanism of lithium-ion battery for electric vehicles: A review. Energy Storage Materials, 10, 246–267. DOI: 10.1016/j.ensm.2017.05.013
- Spotnitz, R., & Franklin, J. (2003). Abuse behavior of high-power, lithium-ion cells. Journal of Power Sources, 113(1), 81–100. DOI: 10.1016/S0378-7753(02)00488-300488-3)
- Ren, D., Hsu, H., Li, R., Feng, X., Guo, D., Han, X., Lu, L., He, X., Gao, S., Hou, J., Li, Y., Wang, Y., & Ouyang, M. (2019). A comparative investigation of aging effects on thermal runaway behavior of lithium-ion cells. eTransportation, 2, 100034. DOI: 10.1016/j.etran.2019.100034
- Liu, X., Ren, D., Hsu, H., Feng, X., Xu, G., Zhuang, M., Gao, H., Lu, L., Han, X., Chu, Z., Li, J., He, X., Amine, K., & Ouyang, M. (2018). Thermal Runaway of Lithium-Ion Batteries without Internal Short Circuit. Joule, 2(10), 2047–2064. DOI: 10.1016/j.joule.2018.06.015
- Lamb, J., Orendorff, C. J., Steele, L. A. M., & Spangler, S. W. (2015). Failure propagation in multi-cell lithium-ion batteries. Journal of Power Sources, 283, 517–523. DOI: 10.1016/j.jpowsour.2014.10.081
Technical Reports
- EUCAR (2019). Battery Hazard Levels and Safety Testing Standards for Li-ion Batteries in Automotive Applications. European Council for Automotive R&D.
- ARAI (2022). Testing Protocol for AIS-156 Phase 2 Compliance: Nail Penetration and Propagation Assessment. Automotive Research Association of India, Pune.
- ICAT (2023). EV Battery Safety Certification Guide for Indian Market Entry. International Centre for Automotive Technology, Manesar.
Further Reading — EVPulse Series
- ← Beginner: What Is Thermal Runaway? (And Why Your EV Won't Randomly Explode)
- → Expert: Thermal Runaway — What Actually Happens Inside a Cell Before It Catches Fire
- → Master: Thermal Runaway — ARC Testing, Propagation Modelling, and Pack Certification
This is the Intermediate level of the EVPulse Thermal Runaway series.
← Beginner: What Is Thermal Runaway? (And Why Your EV Won't Randomly Explode)
→ Expert: Thermal Runaway — What Actually Happens Inside a Cell Before It Catches Fire
Published on EVPulse — India's most technically rigorous source for battery technology and EV engineering coverage.