BESS Thermal Runaway Cascade and Enclosure Vent Gas

A lumped thermal network follows the cell-to-cell propagation framework of Feng et al., Applied Energy 154 (2015), 74–91, DOI 10.1016/j.apenergy.2015.04.118. Each cell has its own temperature, thermal capacitance and irreversible runaway event. Adjacent interfaces conduct heat; exposed surfaces exchange heat with the ambient and a shared hot vent path.

The hot-path balance is $C_h\dot T_h=\dot n C_{p,g}(T_v-T_h)+\sum_i H_h(T_i-T_h)-U_h(T_h-T_a)$. Temperatures in radiation and gas laws are absolute. The initiating cell starts at the trigger temperature at time zero; the energy needed to preheat it is included in the energy ledger.

Each threshold crossing latches a finite heat pulse and a gas pulse. Their cumulative shape is $S=3u^2-2u^3$, where $u$ is elapsed time divided by pulse duration, clipped to 0–1. Heat is $Q_c=C_c\Delta T_{ad}$; source integrals are evaluated exactly over each numerical interval. Heun integration advances the thermal ODEs, with thermal-stability and pulse-resolution limits and 14 bisections to locate the next onset. Interface and hot-path exchanges cancel in the energy balance.

Cooled gas is converted between reference and enclosure conditions by the ideal-gas law. In a pressure-relieved, well-mixed enclosure, raw-gas fraction $c$ obeys $V\dot c=q-(q+Q)c$. Here $q$ is the gas source and $Q$ is clean-air exhaust after the specified start time; the $qc$ term accounts for source-induced displacement. The constant-average-source exponential solution advances concentration within each interval and locates 25% LFL crossings.

For undiluted vent-gas fractions $y_i$, $L_{raw}=[\sum_i y_i/L_i]^{-1}$. The fuel-only mixture limit is $L_f=(\sum_i y_i)L_{raw}$. The comparison is $c/L_{raw}=0.25$. CO₂/inert gas is excluded from the combustible sum; its detailed suppression chemistry is outside Le Chatelier's rule.

The vent-source chart evaluates the analytical pulse derivative directly from the latched onset times. A separate event-based grid resolves every observed pulse, including any partial pulse at the simulation horizon. Its plotted integral is checked against the released-volume ledger.

• One homogeneous thermal node per cell in a linear chain, with identical cells and interfaces and one common hot-gas node. A folded schematic preserves chain numbering; its dimensions are symbolic.
• Default trigger 132.7 °C and adiabatic rise 687.1 K use Feng et al. (2018), Figure 12B: trigger 132.7 °C, peak 819.8 °C. The measured 25 Ah NCM cell is at 100% state of charge. The alternative Figure 12A trigger is 250.2 °C.
• Thermal capacitance 900 J/K, 20 s heat duration, 5 s gas duration, interface resistance, cooling, exposed area, emissivity and hot-path parameters are editable scenario assumptions. They are not a fitted reproduction of the 2015 experimental pack or its reaction kinetics.
• Default gas fractions are Golubkov et al. (2014), Figure 6, NMC 18650 data at full charge. They are an illustrative composition for the larger-cell scenario. The 50 L source represents 2 L/Ah for an illustrative 25 Ah cell; Amano et al. (2023), pp. 12–13, reports substantial variation in normalized yields. Replace yield, composition and durations with tests on the actual cell.
• Cell retained calorimetric heat and sensible enthalpy of the departing vent stream are separate energy inputs. Do not enter total fire-calorimeter energy as retained cell heat and then add the same vent enthalpy again.
• The hot-path temperature describes heat exchange near the module. Enclosure concentration assumes the gas subsequently cools to the entered enclosure temperature. The exhaust does not directly cool the local hot path; its effect on that path belongs in the effective cooling conductance.

This calculates a pre-ignition propagation and gas-accumulation scenario. It does not predict flame jets, combustion, explosion pressure, deflagration vent sizing, HF/CO exposure limits, oxygen depletion, rack-to-rack fire spread, busbar faults, internal cell gradients, mass ejection or evolving contact resistance. A lower peak enclosure average cannot exclude a locally flammable plume. Neither a concentration below 25% LFL nor an untriggered cell is a safety or compliance verdict.

Le Chatelier uses ambient-air component LFLs of 4%, 12.5%, 5%, 2.7% and 3% for H₂, CO, CH₄, C₂H₄ and C₂H₆, respectively. Gas temperature, pressure, inert content, test apparatus and unmeasured vapours can change the real limit. The raw-gas equivalent treats inert gas as dilution, without its thermal or chemical inhibition. Use measured mixture limits for design.

NFPA 855:2026 concerns stationary energy-storage installations; NFPA 69:2024 concerns explosion-prevention systems. The 25% LFL comparison here is one concentration-control benchmark, not an implementation of every provision or alternative in either standard. Check against the standard in force and the edition adopted by the authority having jurisdiction. Standards status checked 1 November 2026.

The solver uses at least 12 intervals per gas pulse and 20 per heat pulse, and refines further for stability. The default 0.42 s maximum actual step keeps browser work near 3 s; a 0.25 s trial changed final onset by less than 0.01 s and peak gas by less than 0.001 percentage point. At the same cost, very short pulses or stiff interfaces need a shorter horizon; a 24,000-step estimate is enforced. Temperature and enclosure-concentration histories have 901 samples, with 601 concentrated in the initial transient and 300 in the tail; where the entire run fits in that transient they are uniform. Playback interpolates between saved states. Narrow features in these histories can be visually undersampled, although peak and threshold metrics use every integration step. Reduce the horizon to inspect them. The vent-release chart uses a separate grid containing every observed pulse's start and end, clipped at the simulation horizon. Each pulse has 64 intervals for up to 24 observed cells, or 32 intervals for larger cascades. Flow is evaluated analytically from onset times, without additional thermal-solver work. The displayed window ends just after the last observed release or at the horizon. Its trapezoidal integral agrees with generated gas volume to within 0.1%.

Blank onset/crossing results mean the event was not observed within the horizon. “Initiator only” does not prove propagation will never occur. Time above the comparison is truncated at the horizon, and gas/heat generated counts only released amounts within it. Extend the horizon to check delayed events and clearance.

Source-library correction: the attachment headed “Sturk et al. 2023” contains Amano et al., Batteries 2023, 9, 300; the reference below gives its correct authorship. The server protects attachment metadata.