Predicting heat release rate from fire video data:

Abstract

A novel approach for estimating transient heat release rate from fire images and video using deep learning techniques is presented. The heat release rate (HRR) is a critical parameter in characterizing the fire hazard and thermal effects of a burning item. It is an effective indicator of the fire growth rate and fire hazard that is used extensively in building fire safety design. However, for outdoor fires, heat release rate measurements are usually not available due to lack of measurement equipment.The goal of this work is to develop and demonstrate a novel technique based on "image calorimetry" for predicting heat release rate using video data and recurrent neural network models. The proposed methodology only requires video camera data and can be extended to outdoor fire experiments. Results of the trained neural network model on the investigated set of experiments conducted in a laboratory setting, show excellent comparison between predicted and temporally evolving heat release rate measurements, with an overall accuracy score of 0.94.