Implications of data smoothing on experimental mass loss rates

Abstract

Time resolved mass loss rate (MLR) is a commonly reported metric in fire safety science and is a measurement often characteristic of a high degree of noise. Such noise encourages the use of various smoothing techniques to illustrate average MLR behavior. A study was conducted to observe the influence of data smoothing techniques on experimentally determined mass loss rates. Methodologies including numerical differentiation techniques, the standardized ASTM E1354/ISO 5660 method, simple moving averages, and the Savitzky-Golay filter were applied to MLR data from charring (timber) and non-charring (PMMA and PA6) solids. Transient behavior such as charring and determining the MLR at ignition are particularly influenced by the choice of processing scheme. The peak MLR for charring materials was truncated up to 30% when using high-order filters (50-Pt Moving Average or 51-Frame Length Savitzky-Golay). This work outlines a novel integral method through which the user can quantify the degree to which a given smoothed MLR curve fits the true observed mass loss behavior. No single approach to processing MLR data can fully capture the variation associated with the material and instrumentation used; therefore careful consideration should be taken when smoothing MLR data.