Bayesian Inference of Fire Evolution Within a Compartment Using Heat Flux Measurements

The dynamical evolution of a multiple fuel package fire leaves thermal signatures. For practical and theoretical reasons, it is important to determine conditions in which one can identify the path the system took by conducting a set of experiments that cover the space of all possible paths. An experimental fire compartment capable of producing repeatable and highly customizable fire-evolution scenarios is presented. Instrumented propane burners are configured in the compartment each with a simple critical heat flux ignition model. Heat flux sensors are located around the burner configuration to provide temporal incident heat flux measurements. Data from several hypotheses representing possible scenarios are compared to data generated using some true configuration using a Bayesian methodology. The Bayesian methodology is able to illicit the correct fire-evolution scenario from the set of hypotheses with a high degree of confidence. The information content provided by each sensor is analyzed to highlight the importance of sensor location in determining the fire-evolution. Posteriors for the hypotheses using two different error structures are also compared over the sensors to highlight the importance of choosing the correct error structure.