Bayesian inversion for the calibration of fire experiments on insulation panels

Abstract

In the setting of structural design for fire hazards the simulation of thermal transport in insulation panels is of great interest. The use of such simulation tools initially requires the calibration of the underlying heat transport models and the temperature-dependent specific heat capacity, conductivity and density. We herein outline an approach for the calibration of a finite element model (FEM), describing heat transfer in insulation panels when exposed to fire, by using a Markov chain Monte Carlo (MCMC) based Bayesian inference approach. The procedure is enhanced by employing a surrogate modelling technique that couples polynomial chaos expansion (PCE) with principal component analysis (PCA). This allows the model parameters in question to be inferred in an efficient way. We showcase the adequacy of this approach by calibrating the temperature-dependent effective material parameters for a set of real measurements. We further show the superiority of the presented approach to a widely used conventional manual calibration method in timber engineering.