Fire safety engineering relies on engineering judgement when empirical evidence is insufficient. This challenge is particularly evident for Battery Energy Storage Systems, where limited large-scale fire data require engineers to make assumptions under epistemic uncertainty. Despite the central role of engineering judgement in design fire selection, little empirical evidence exists on how experienced engineers develop representative design fires. This study examines this process through an international Round Robin in which experienced fire safety professionals independently addressed the same BESS consultancy task using identical project information. The submissions were analysed to compare how participants identified scenario clusters, selected and represented design fires, and justified their engineering assumptions. The results show that the variability was evident in how participants interpreted the engineering task and translated incomplete knowledge into engineering assumptions. Four recurring reasoning strategies were identified: construction, transfer, analogy and constraint. These findings highlight the importance of transparent and well-justified engineering assumptions in fire safety engineering.
Quantitative risk assessment of accidental hydrogen releases requires reliable prediction of the physical effects associated with loss-of-containment scenarios. This work compares the predictive performance of PHAST v9.0, HyRAM + v5.1.1, and EFFECTS v13 against three experimental benchmarks for compressed hydrogen dispersion, jet fire flame length, and radiation. Model predictions were compared with measurements using parity plots and statistical performance metrics. For dispersion, all three tools satisfied the established acceptability criteria for pointwise comparisons in simple geometry, including mean relative bias within ±0.4, mean relative square error below 2.3, and at least 50% of predictions within a factor of two of the measurements. PHAST showed the lowest mean relative square error (0.097), and EFFECTS the highest fraction of predictions within a factor of two (0.95). For jet fire flame length, the three tools showed comparable performance and generally overpredicted the experimental values, particularly for larger nozzle diameters. PHAST and HyRAM + predicted all flame lengths within a factor of two of the measurements, while EFFECTS had only one value outside this range. For jet fire radiation, only PHAST and HyRAM + could be compared quantitatively. Both remained within a factor of two of the measurements for all points, but PHAST reproduced the experimental values more closely overall, with a total mean relative square error of 0.021. Overall, all three tools showed satisfactory performance, although accuracy depended on the phenomenon considered and the underlying modelling assumptions. PHAST provided the most consistent performance across the three benchmarks, while EFFECTS performed particularly well for hydrogen dispersion.
