Robust probabilistic modelling of mould growth in building envelopes using random forests machine learning algorithm

Abstract

Probabilistic methods can be used to account for uncertainties in hygrothermal analysis of building envelopes. This paper presents methods for robust mould reliability analysis and identification of critical parameters. Mould indices are calculated by probabilistic hygrothermal analysis, followed by the application of the "Finnish mould growth model." To increase the robustness of the mould growth analysis, a random forests metamodel is first trained on the dataset and then used to expand the number of simulations. Finally, the reliability is calculated based on the probability of exceeding a given maximum mould index limit state. Critical parameters are identified through a sensitivity analysis based on linear and non-linear dependencies between inputs and maximum mould index. The methods are demonstrated by analysing three external wall assemblies. In conclusion, the mould reliability analysis method helps to assess the robustness of the hygrothermal analysis and mould assessment by investigating the influence of hygrothermal variables' uncertainties on the maximum mould index. By combining a metamodel with probabilistic analysis, it is possible to significantly reduce the amount of time required to evaluate a large number of scenarios.