The effect of uncertainty in calibration data on burning rate predictions

• NCJ Number: 308427
• Journal: Fire Safety Journal Volume: 143 Dated: February 2024
• Author(s): Mark B. McKinnon; Holli Knight
• Date Published: February 2024
• Length: 10 pages
• Grant Number(s): 2019-DU-BX-0018
• Agencies: NIJ-Sponsored
• Publication Type: Research (Applied/Empirical)

Annotation

This paper reports on a numerical investigation of the effect of repeatability of thermogravimetric analysis data on the kinetics calibrated from the data; it notes that uncertainty in calibration TGA data causes shifts in kinetic parameters and is directly correlated to uncertainty in burning rate predictions; and provides guidance on acceptable upper threshold of dispersion in replicate TGA data.

Abstract

Comprehensive pyrolysis models describe solid phase reaction rates with respect to the material temperature and the concentrations of components. In theory, this aspect of comprehensive pyrolysis models allows for true predictive capabilities for pyrolysis models and, more generally, fire models. To utilize these capabilities, the reaction kinetics must be defined for the material of interest and there is a lack of publicly available data from which these kinetic parameters may be determined. There is also a lack of guidance for quality and repeatability for thermal analysis experiments, which complicates the use of mean thermal analysis data found in the literature to estimate the thermal degradation kinetics. This study details a numerical investigation on the effect of repeatability of thermogravimetric analysis data on the kinetics calibrated from the data. The uncertainty in the kinetic parameters that corresponded to target repeatability values imposed on synthetic calibration data was determined through a latin hypercube sampling procedure. The generalized polynomial chaos expansions method allowed propagation of the uncertainty of the kinetic parameters through a one-dimensional model of a bench-scale experiment. The relationship between uncertainty in calibration data and uncertainty in bench-scale model predictions was linear. (Published Abstract Provided)