The role played by uncertain input data and measures on the modeling of the Viareggio LPG railway accident

Experiments involving the release and dispersion of heavier-than-air gases and models describing their behavior allow carrying out risk assessment studies and accident investigations. During experimental campaigns, the factors influencing the dense gas dispersion are either controlled or measured, e.g., the release flow-rate, the meteorological conditions, the presence of obstacles.

When a real accident occurs, there is often a degree of uncertainty on these factors, which affects the results of the models used in the accident investigation.

This manuscript investigates the dramatic outcomes of the accident event occurred on June 29th, 2009 at 11:49 PM at the railway station of Viareggio (Italy). A train for the LPG transportation running at about 90 km/h went off the rails. Five out of fourteen tank cars derailed and overturned. A 100 cm2 hole formed in the first car, probably due to the impact with a signaling stake. The pressurized LPG was released as a two-phase jet: the liquid phase formed a boiling pool on the ballast while the dense gas dispersed in the atmosphere. The wind blew the cloud towards the neighboring houses. After approximately 2-5 minutes, the cloud was ignited and gave rise to a flash fire in the open air and to a VCE inside some houses, which were destroyed by the overpressure. Thirty-two people died and several residents were injured.

The manuscript assesses and discusses the complexity of modeling a real accident event subject to the uncertainties related to both input data and measurements. A sensitivity analysis is carried out to ascertain the role played by those uncertainties on the accident outcomes.