Charlie Shields and Yuxiang Chen
i MSc Student, University of Alberta, Edmonton, Canada, cshields@ualberta.ca
ii Associate Professor, University of Alberta, Edmonton, Canada, ychen5@ualberta.ca
ABSTRACT
The pursuit of energy-efficient building design has intensified with the recent National Energy Code of Canada for Buildings (NECB), which mandates comprehensive accounting of thermal bridges in building envelope assessments. This study addresses the challenge of cataloging linear transmittance values for brick veneer envelopes with concrete masonry unit (CMU) backup walls. To capture the extensive variability in wall configurations, parameterized models are needed, enabling systematic exploration of thermal performance across diverse scenarios. However, the computational burden of simulating every possible variation remains prohibitive. To overcome this, we integrated machine learning models trained on a subset of parameterized simulations, allowing the prediction of thermal performance for unmodeled configurations by learning the influence of key design parameters. This method reduces modeling time by over 99% while maintaining high accuracy, enabling rapid, informed design decisions and supporting the development of high-performance, NECB-compliant building envelopes.
KEYWORDS: Thermal performance, brick veneer envelopes, machine learning, parametric modeling, surrogate modelling.
184-Shields.pdf