Statistical uncertainty in bench-scale flammability tests

Abstract

Standardized testing procedures and experimental studies in fire safety science specify repeat trials in an effort to minimize the impact of statistical variation on the presented results. Bench-scale flammability testing procedures typically require a constant number of trials – generally a total of three – across all experimental conditions. A series of experiments was conducted to quantify the influence of trial quantity on the statistical variation in Cone Calorimeter results; sets of 100 identical trials were conducted using black PMMA for three different heat flux exposures (20, 40, and 60 kW/m2). Gaussian statistics were applied to the time to ignition data and time-resolved mass loss rate data to illustrate the influence of trial quantity on predicted statistical uncertainty. Statistical variation observed for time to ignition data was found to decrease with increasing heat flux while the variation observed in the MLR data increased with increasing heat flux. The statistical uncertainty was found to decrease as a function of 1/n for n trials. The presented statistical analysis requires population characteristics to be known, or assumed, a priori; the results, however, demonstrate the importance of considering statistical uncertainty in addition to the variation within data collected under different exposure conditions.