Towards a Bayesian framework for fire origin and evolution in fire forensics

Date

2020-12-04

Authors

Cabrera, Jan-Michael

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

Fire scene reconstruction and determining the fire evolution (i.e. item-to-item ignition events) using the post-fire compartment is an extremely difficult task because of the time-integrated nature of the observed damage within the compartment. Adequately quantifying the uncertainty for quantities of interest is crucial for making statistically sound inferences about the pre-fire compartment.

The work needed to implement an experimental fire compartment capable of producing repeatable fire-evolution scenarios is presented. Improvements to the University of Texas Fire Research Group (UTFRG) Burn Structure are discussed including increased automation and reduced setup time for compartment scale fire tests as well as an ability to automate fire evolution tests. Instrumented burners are configured in the compartment, each with a simple ignition model. Heat flux sensors are located around the burners to provide temporal incident heat flux measurements. The heat flux sensors are directional flame thermometers (DFTs); robust measurement devices suitable to the harsh environments found in fire scenarios.

The typical DFT is large when compared to other standard heat flux measurement devices. To better understand the uncertainties associated with heat flux measurements in these environments, a Bayesian framework is utilized to propagate uncertainties of both known and unknown parameters describing the thermal model of a modified, smaller DFT. Construction of the modified DFT is described along with a derivation of the thermal model used to predict the incident heat flux to its sensing surface. Markov Chain Monte Carlo simulations were used to obtain posterior distributions for the free parameters of the thermal model as well as the modeling uncertainty.

A Bayesian inferential framework is developed to test possible ignition scenarios given measurement uncertainties against data taken at the fire scene. The framework is developed for temporal heat flux measurements as well as for the observed damages at surrogate sensors within the compartment. The framework is first exercised on temporal heat flux measurements recorded from compartment tests in the Burn Structure to highlight the importance of error structure for making calibrated predictions. The use of computational models for making inferences on damage measurements taken in the compartment utilizing a Bayesian inferential framework is also discussed.

Description

LCSH Subject Headings

Citation