This study applies a forensic-science–based approach combined with three-dimensional CFD simulations to reconstruct a large-scale explosion at a chemical plant and to evaluate the plausibility of candidate accident scenarios. A joint on-site investigation and document review were first used to reconstruct a baseline scenario for butane leakage, vapor cloud formation, and ignition. FLACS simulations were then conducted to examine whether a range of assumed leak volumes could reproduce the observed damage pattern, with particular emphasis on overpressure at a reference building located 93 m from the explosion center. The National Disaster Management Research Institute (NDMI) led the investigation, which included on-site inspections and document analysis. Based on physical evidence such as major structural damage, fragment distribution, and gas detector activation data, the location and dispersion range of the leak were estimated. A 3D simulation of the incident process was performed using the FLACS software. Alongside theoretical leak estimations, a reverse-engineering simulation approach considering actual damage patterns confirmed the possibility of a butane release exceeding 11,000 kg. The simulation results quantitatively matched the measured damage radius and verified that the calculated overpressure levels met established damage thresholds. This study scientifically reconstructed the accident scenario using quantitative simulation methods, demonstrating the effectiveness and applicability of forensic-based disaster investigations. The proposed methodology may serve as foundational data not only for identifying the root causes of disasters but also for informing prevention policies, institutional reforms, and legal liability assessments. • A forensic science-based framework for explosion reconstruction is proposed. • Field evidence is systematically integrated with CFD simulations. • Simulation results constrain plausible gas leak scenarios. • Methodological limitations and uncertainties are explicitly discussed.
Lee et al. (Sun,) studied this question.