DEVELOPMENT AND USE OF PIXEL-BY-PIXEL PYROMETRY METHODS ON SMOLDERING WOOD EMBERS AND PILES

Abstract

Wildfire, especially along the Wildland Urban Interface (WUI), presents a large threatto life and property globally. Firebrands can increase the size and spread of wildfires rapidly. More than half of spot fires and WUI fires are caused by firebrands. Firebrand generation, transport, and morphology has been studied in recent literature, but few papers have reported firebrand temperatures, even fewer on groups of firebrands across varying configurations. Those that have reported temperatures have typically relied on expensive IR and thermocouple data that may not be as accurate at determining temperature or emissivity. Color camera pyrometry presents a high resolution alternative to previous pyrometry methods and can be done using a cheap color camera, and with certain techniques can derive temperature independent of emissivity. This research builds on previous color camera pyrometry, automating the process to allow for large datasets to be analyzed as opposed to single images. Two-color, grayscale, and hybrid pyrometry [20] were used to recreate pyrometry results of previous literature. Similar average single firebrand temperatures in the range of 900-950C were reported. A novel Pixel-by-Pixel hybrid pyrometry was developed to incorporate more data into established hybrid pyrometry methods. This method introduced large amounts of noise into the temperature results, making them unreliable. Additionally, a method was developed for determining the temperature of 8-gram ember piles at various wind speeds of 1.4, 2.4, and 2.7m/s through a borosilicate glass window. Modified grayscale temperatures assuming constant emissivity were used for these experiments and were fit to firebrand temperature data from Kim and Sunderland [20]. A total of 720 ember pile images were analyzed in the final dataset at an effective emissivity of 0.76. Peak ember pile average temperatures ranged from 700-900C. Normalized temperature (T /Tmean) PDFs were produced. Data was approximated as a normal distribution with mean of 1 and standard deviation ranging from 0.048 - 0.057.