A BAYESIAN METHOD FOR DETERMINING THE FIRE EVOLUTION WITHIN A COMPARTMENT

Fire forensics, which aims to narrow the hypothesis space for a fire's area of origin and cause, is a complex endeavor that is known to be significantly error prone. While the post fire compartment is generally data rich in terms of the signatures the fire has left on objects within the compartment, there are few rigorous tools and techniques to connect these data to fire origin determination. Additionally, a challenge with relying on these data is that the fire damages the contents in the compartment through which it has propagated, obfuscating signatures produced early on in the fire's evolution. In this work, we present a rational framework for determining the first item ignited and most probable compartment fire evolution given post fire observations. The framework relies on Bayesian techniques to distinguish the first item ignited within a compartment from many different potential ignition sources given the post fire observations. This work evaluates a low-order, fire propagation model and low-order damage models to demonstrate the potential for determining the first item ignited within a compartment and the subsequent fire evolution. The inversion framework was tested on synthetic data produced using the fire CFD tool Fire Dynamics Simulator (FDS).