Hygric characterization of wood fiber insulation under uncertainty with dynamic measurements and Markov Chain Monte-Carlo algorithm

Publication date: March 2017
Source:Building and Environment, Volume 114
Author(s): Simon Rouchier, Thomas Busser, Mickaël Pailha, Amandine Piot, Monika Woloszyn
The present work is the hygric characterization of wood fibre insulation boards, using dynamic measurements of relative humidity and sample weight, analyzed in the frame of Bayesian inference for parameter identification under uncertainty. It is an attempt at identifying detailed profiles of moisture-dependent properties, and thus a relatively high number of parameters. Because of this ambition, some caution should be exercised once the outcome of the inversion algorithm is available: in addition to confidence intervals of parameters provided by the Bayesian framework, a simplified form of identifiability analysis is performed by analysing a posteriori parameter correlations and likelihood-based confidence intervals. The characterization methodology does not require for the model structure to have a differentiable analytical formulation, or for material samples to reach mass equilibrium between each RH step of the experimental process. Two separate experimental designs were used for material characterization and for validation, respectively. Results show a clear relation between available information (experimental data) and inference (confidence intervals of parameters). A single relative humidity step is not informative enough for a precise inference of moisture-dependent properties such as vapour permeability and moisture capacity. A two-step experiment however holds enough information to significantly reduce parameter uncertainty.