Predicting moisture-related degradation risks across facades

Licentiatavhandling, 2025

Predicting facade deterioration due to microclimate effects is crucial for sustainable building management. Traditional hygrothermal assessments often neglect these microclimate effects, leading to underestimated moisture risks, localized degradation and increased maintenance costs. To address these limitations, research advocates for the transition from a one-dimensional to a multi-dimensional approach, using Computational Fluid Dynamics (CFD) to capture microclimate effects, though this method is complex and computationally intensive.

In this work two methodologies were developed—Engineering and Data-driven approaches—to assess moisture-induced degradation risks across facades, section by section. The Engineering approach combines semi-empirical wind-driven rain methods with solar irradiation analysis to model hygrothermal loads and risks. The Data-driven approach uses image-based analysis, 3D model data and climate data to train a machine learning algorithm for predicting degradation ratios.

                The framework for Degradation Analysis over Facades was introduced to validate these methodologies and to provide degradation data for the Data-driven approach. Using drones and computer vision, the framework was applied to 16 case study facades in Gothenburg, Sweden. The analysis of degradation data reveals that south-facing facades are more damaged, while degradation patterns vary.

                The Engineering approach, applied to a single, yet representative, facade due to the laborious simulation process, revealed discrepancies between the modeled and observable degradation. A better incorporation of the roof overhang into the chosen WDR method (Straube and Burnett method) would increase the accuracy of the approach. The Data-driven approach, applied to all 16 facades, performed well in predicting degradation using the test dataset, achieving R-squared (R2) value of 0.978, but deviated when predicting degradation on the validation facade (R2=-1.152), which was initially excluded from the analysis. This was attributed to the distribution shift that arises due to limited training data and unknowns introduced from prior maintenance.

                Despite these limitations, both methodologies offer promising alternatives to full-facade moisture risk assessment using CFD. Although tested on a case from Sweden, these methodologies are broadly applicable. Ultimately, the Data-driven approach could offer a computationally inexpensive way to assess facades in entire neighborhoods.