Probabilistic Risk Assessment of Grid-Scale Lithium-Ion Battery Energy Storage System Fire Hazards: Hydrogen Fluoride (HF) Toxicity, Suppression Effectiveness, and Comparative Compartment Design Analysis

Overview

The study starts from the problem that qualitative frameworks such as NFPA 855's 5 × 5 consequence-likelihood matrix are too coarse to discriminate BESS fire hazards, and proposes a quantitative approach that separates design alternatives on risk grounds. The case installation has 2 × 242.76 kWh compartments (485.52 kWh total), 14 NMC cabinets of 34.68 kWh, a compartment volume of 116 m³, 9 ACH emergency purge and ambient 30-34 °C at 75-85% RH. Monte Carlo (N = 10,000) characterises uncertainty in HF dose, IDLH time, propagation probability and suppression effectiveness.

HF yield was sampled as a triangular distribution of 0.3-0.8 g/kWh (mode 0.5), whereas the source it cites (Larsson et al., Sci. Rep. 2017) reports 20-200 mg/Wh (20-200 g/kWh), so the input appears understated by one to two orders of magnitude and absolute concentrations should be read with caution. Ventilation activation delay is log-normal (median 90 s) and suppression effectiveness a piecewise function of water application delay. The pre-action sprinkler timeline is about 2-4 min two-zone detection confirmation, 1-3 min solenoid and pipe fill and 0.5-1 min head operation, so 3.5-8 min from thermal runaway to water (8-15 min with manual fire-service intervention).

Results: with 9 ACH, mean IDLH (25 mg/m³) clearance time was 599 min for one compartment and 301 min for two; the probability of clearing within 5 min was 0% for both. Mean 10-min occupant HF dose was 1161 mg/m³ (one) and 580 mg/m³ (two), 46× and 23× IDLH; IDLH was exceeded in 100% of simulations for both designs. Mean water-only suppression effectiveness was 37.9%, against a nominal 78% if water arrives within 3 min, because the median delay was 7.9 min (95th percentile 21.6 min). Per compartment-year, P(multi-cabinet TR) = 0.01 × 0.15 (BMS fail) × 0.08 (UL 9540A containment fail) = 1.2 × 10^-4; expected risk to life was 1.22 × 10^-4/yr with water-only suppression and 2.40 × 10^-5/yr with gas + water (an 80.3% reduction).

Conclusions: because HF dose exceeds IDLH in full thermal runaway under both designs, runaway prevention (BMS integrity, UL 9540A containment, emergency power off) is the only effective life-safety control; two compartments halve HF dose and clearance time and move residual risk from ALARP-tolerable to broadly acceptable. Limits: a single installation and author-chosen distributions; absolute risk levels carry credible intervals spanning more than an order of magnitude, while the relative benefit is stable under parameter uncertainty.

Key points

1. Case installation: 2 × 242.76 kWh compartments (485.52 kWh total), 14 NMC cabinets of 34.68 kWh, compartment volume 116 m³, 9 ACH emergency purge, ambient 30-34 °C and 75-85% RH.
2. HF yield was sampled as a triangular distribution of 0.3-0.8 g/kWh (mode 0.5); the cited Larsson et al. (Sci. Rep. 2017) reports 20-200 mg/Wh, so the input appears understated by one to two orders of magnitude.
3. BESS pre-action sprinkler timeline: 2-4 min two-zone detection confirmation, 1-3 min solenoid and pipe fill, 0.5-1 min head operation, so 3.5-8 min from thermal runaway to water (8-15 min with manual fire-service intervention).
4. Mean IDLH (25 mg/m³) clearance time with 9 ACH was 599 min for one compartment and 301 min for two; the probability of clearing within 5 min was 0% for both.
5. Mean 10-min occupant HF dose was 1161 mg/m³ (one compartment) and 580 mg/m³ (two), 46× and 23× IDLH; IDLH was exceeded in 100% of simulations for both designs.
6. Mean water-only suppression effectiveness was 37.9%, against a nominal 78% if water arrives within 3 min, because the median delay was 7.9 min (95th percentile 21.6 min).
7. Event tree: P(multi-cabinet TR) = 0.01 × 0.15 (BMS fail) × 0.08 (UL 9540A containment fail) = 1.2 × 10^-4; expected risk to life was 1.22 × 10^-4/yr with water-only suppression and 2.40 × 10^-5/yr with gas + water.

Using it in design

• Compare BESS compartmentation quantitatively: two compartments halve HF dose and IDLH clearance time, but IDLH is still exceeded for anyone in the room in full thermal runaway, so runaway prevention (BMS integrity, UL 9540A containment, emergency power off) must stay the primary control.
• Treat water-only suppression as weak because of long pre-action delays, and justify a dual system with gas as first stage and water as second by expected risk to life.
• In HMA studies use a Monte Carlo approach that shows probability distributions rather than a 5 × 5 matrix, verify the HF yield input against its source, and present absolute concentrations with sensitivity analysis.

Who should read it

Engineers and risk analysts working on BESS and battery-room fire safety, HMA and suppression system design.

Document contents

1. 1. Introduction (1.1 Problem Statement, 1.2 Gap Analysis, 1.3 Research Questions, 1.4 Contributions to Knowledge)
2. 2. Background (2.1 The EQIX SG4-4A BESS Installation, 2.2 Thermal Runaway Chemistry, 2.3 HF Generation Chemistry, 2.4 HF Toxicological Reference Values, 2.6 NMC Mediated Lithium-Ion Chemistry in Regulatory and Comparative Contexts)
3. 3. Methods (3.1 Monte Carlo Simulation Framework, 3.2 Installation Parameters, 3.3 Probability Distributions, 3.4 HF Dose Model, 3.5 Suppression Delay Justification, 3.6 Suppression Effectiveness Model, 3.7 Verification and Validation Strategy)
4. 4. Results (4.1 HF Concentration and IDLH Clearance Time, 4.2 HF Dose to Occupant, 4.3 Propagation Probability and Annual Risk, 4.4 Suppression Effectiveness, 4.5 Dual Suppression System, 4.7 Spatial Dispersion Results: CFD Simulation Using FDS, 4.8 Comparative Risk Summary, 4.9 Sensitivity Analysis)
5. 5. Discussion (5.2 The Case for Two-Stage Suppression, 5.3 The HF Toxicity Finding, 5.5 Tropical Climate Effects)
6. 6. Main Conclusions

Related standards

• NFPA 855
• UL 9540A
• NFPA 2001
• NFPA 13

Read the document

Publication details

Authors

Samson Tan; Teik Toe Teoh; Paul Joseph; Khalid Moinuddin

Publisher

Fire (MDPI AG)

Year

2026

Volume

9

Issue

8

Pages

319

Document language

English

Source type

Scientific

Access

Open licence — downloadable

DOI

10.3390/fire9080319

Licence

CC BY 4.0 — share with attribution

Licence evidence

www.mdpi.com

Citation