Development of predictive models for the probabilistic moisture risk assessment of internal wall insulation

This paper presents a method for developing predictive meta-models that can be used for a fast probabilistic moisture risk assessment of IWI, considering both the uncertainty and variability of input variables. First, in a Monte Carlo analysis, the uncertainty and variability of inputs were propagated through hygrothermal simulations. Then, generalised additive models for location, scale and shape (GAMLSS) were used to describe the relationship between inputs and response variables of the Monte Carlo analysis. The key input variables were identified by a global sensitivity analysis - using the elementary effects method - and in model building. Two types of response variable were considered for the models: variables based on percentage values (e.g. maximum relative humidity) and dose-response relationships (e.g. mould index). The paper shows that both risk assessment models had a good predictive power, highlighting the suitability of the developed method for the moisture risk assessment of the internal insulation of solid walls.