Abstract

It is generally understood that the average temperature of the earth is increasing, resulting in an increased number of extreme events. To assess the impact of climate change on the durability of the building envelope, a commonly used method is to use hygrothermal modeling tools to perform simulations. The hygrothermal response varies depending on the location, material properties, type of wall assemblies, etc., and hence proper inputs are required. In general, indicator based on simulation results indicates the moisture risk. However, to obtain this indicator and considering different situations, a large number of simulations are required. This research thus focused on developing a climate-based index that can give a range of expected performance of the wall without performing the simulations.
Firstly, different existing climate-based indices were computed and correlated with the performance indicator to quantify the risk. The purpose is to see if any existing climate-based indices can lead to accurate risk assessment and the analysis showed that none of these indices lead to reliable risk assessment. Thus, a machine learning algorithm, Partial Least Squares (PLS) regression was used to develop a new climate-based index. Three cities from different climate zones across Canada and two wall claddings were considered for model development. For each city and future projected climate, the index was calculated, and correlated with the performance indicator to quantify the risk. PLS modeling technique proves to be an effective way in predicting the hygrothermal response and to improve computational efficiency. A PLS model was developed for a brick cladding wall and the model was applied to other wall types and a larger climate range (15 runs of data with each run having 31 years of historical and future climate data). The results showed that the moisture risk increases in the future periods for all three cities and wall claddings and a similar performance was noted for different climate runs. The predicted results from the meta-model can be used as a screening measure to limit the number of simulations to cases where the predicted hygrothermal performance is above a certain threshold set by the user.