BESS Container Spacing and Radiant Heat

The imposed heat-release history is a pixel digitization of the chemical HRR in FM Global's full-scale LFP rack test (2019, Fig. 6-15, attached PDF p. 68). Delayed copies are summed for the chosen rack count and scaled in power and local duration. Each rack is evaluated at $(t-(j-1)\tau)/s_t$: the duration multiplier never changes the entered physical ignition interval $\tau$. This produces a container fire scenario, not a measured container-wide burn. Original test time, including the ignition delay, is retained.

Heskestad gives the mean flame extension $L_f=\max(0,0.235\dot Q^{2/5}-1.02D)$ with $\dot Q$ in kW and $D=\sqrt{4WB/\pi}$ in m. The effective four-sided solid-flame envelope extends from grade to $H_c+L_f$. Its uniform emissive power is $E=\chi_r\dot Q/[2(W+B)(H_c+L_f)]$; all prescribed radiative power is assigned to the four vertical faces. Only the facing face is visible to the aligned neighbour.

For a differential receiver, $F=\int_A d^2/[\pi(d^2+y^2+z^2)^2]\,dA$. The width integral is evaluated analytically and the height integral by midpoint strips, automatically refined when height/separation is large. For a full-length barrier at fraction $f$ of the gap, rays below source height $(H_b-fz_r)/(1-f)$ are blocked. The largest received flux over sampled receiver heights drives the wall column. Its location may change with time.

The barrier is a 1D concrete slab with finite-difference nodes at both faces. The neighbour comprises two lumped steel skins, finite-volume insulation cells and a nearest-cell thermal-capacitance node coupled through the entered conductance. Backward Euler advances $C_j\Delta T_j/\Delta t=\sum G_{jk}(T_k-T_j)+q_j$. Interface conductances use half-cell thermal resistances. Newton iterations solve the nonlinear $\epsilon\sigma(T^4-T_a^4)$ boundary losses; ambient convection is included. The barrier's hot rear face radiates through its own rectangular view factor onto the neighbour. Both scenarios continue through the post-fire observation period.

Radiation is evaluated on a separate event timeline containing every delayed fire-history breakpoint, every crossing of a radiation-power lookup node, and additional samples no farther apart than the radiation-event step. Reported flux maxima use this timeline, independently of the thermal grid. Exact integrals of the piecewise-linear incident flux provide interval-mean thermal forcing, preserving short-pulse energy. Barrier rear temperatures are interpolated from the conduction solution onto the radiation timeline before reradiation is added.

The first cell-threshold crossing is interpolated in time. The spacing search includes the selected gap and logarithmically spaced seed distances. It then solves additional barrier-maximum candidates and brackets the last threshold crossing in each scenario. PCHIP proposes new distances; the reported spacing is the first actually solved below-threshold endpoint of the final bracket, whose width is bounded by the spacing tolerance. A 0.5 m result means every resolved search temperature is below threshold, including the selected gap. A blank spacing means the search range or six refinement rounds are insufficient; a blank threshold time means no crossing during observation.

Parallel, equal-length, aligned containers; upright, buoyancy-dominated fire without wind or flame impingement. The opaque, intact barrier spans at least the entire facing length of both containers, has no penetrations, starts at grade and occupies the entered position. Its thickness enters heat diffusion and front/rear surface distances; its centre plane defines the geometric shadow.

Time-varying peak exposure is applied to one thermal column. For the barrier, this also applies the hottest local rear-face temperature uniformly over its emitting rectangle. The sum of separate maxima for direct and rear-face radiation is conservative with respect to the chosen geometry. Lateral spreading, reflection between surfaces and mutual thermal feedback are omitted. Atmospheric transmission is unity; smoke attenuation is not credited.

Material properties are constant effective values. Illustrative stainless-steel, mineral-wool and concrete properties follow Martelletto et al. (2021), Table 2; insulation conductivity should be replaced by its hot-condition value. The cell heat capacity, coupling and 150 °C default threshold are scenario inputs, not a universal LFP ignition model. Cell self-heating and chemical reactions begin outside this passive thermal model; temperatures shown after a threshold crossing are hypothetical.

This is a radiation and passive-heating exposure calculation, not evidence that a BESS installation will avoid fire propagation. Door or vent jets, wind-tilted flames, hot gases, burning debris, enclosure failure, cable paths, internal propagation, gas accumulation and deflagration require separate assessment. A wall can block radiation while failing to control these other routes. Check the actual tested enclosure, state of charge, cell chemistry, vent arrangement and barrier configuration.

NFPA 855 2026 edition and UL 9540A 6th edition (March 13, 2026) are the comparison basis, reviewed November 2, 2026. The displayed 3 ft group-spacing benchmark is not a universal clearance for a container holding hundreds of kWh or MWh. The open NFPA committee record is development material, not the final normative text. Follow the adopted final edition, installation instructions, HMA and accepted installation-level large-scale fire-test evidence; check against the standard in force. The calculated distance never authorizes reduced spacing. UL's public guidance describes the edition-6 post-deflagration test with intentionally ignited vent gases.

The public HRR source is rack-scale. Rack-to-container scaling, ignition synchronization and power/duration multipliers are explicit extrapolations; they do not capture ventilation limitation, oxygen availability or container-specific test behaviour. The digitized history integrates to about 3.98 GJ per rack, approximately 4% above the report's independently stated 3.81 GJ. Neither the history nor the equations are normalized to match that total. Raster resolution and overlapping traces limit fidelity; some instantaneous peaks are unresolved. The separately reported 2540 kW rack peak bounds the radiation lookup. Replace the source curve with representative open or licensed test data through the builder for project work.

Defaults use 12 through-thickness cells, a maximum 120 s thermal step, 61 radiation-power samples, 17 receiver heights, 13 initial logarithmic separation samples, a 0.01 m crossing tolerance and a maximum 60 s radiation-event interval. The event and adaptive solves raised runtime above approximately 3 s with denser initial grids; these defaults retain the complete heat-transfer model. Interval means preserve incident pulse energy, but backward Euler can smooth the temperature response and shift threshold times by roughly a thermal step. Refine the advanced controls for marginal cases. Adaptive maxima are checked to approximately 0.01 K; search results remain numerical estimates over the entered range. Separation and wall-profile pictures use PCHIP interpolation between solved samples; radiation histories use linear lookup interpolation. Automated checks cover doubled layer meshes, halved thermal steps, both peak-flux outputs, refined radiation integration, open-gap and barrier-spacing convergence, ignition-delay invariance, and energy balances. Extend observation when temperatures are still rising. Concrete moisture, spalling, degradation and temperature-dependent properties are outside the model.