PhD Oral Exam - Sina Akhavan Shams, Building Engineering

When studying for a doctoral degree (PhD), candidates submit a thesis that provides a critical review of the current state of knowledge of the thesis subject as well as the student’s own contributions to the subject. The distinguishing criterion of doctoral graduate research is a significant and original contribution to knowledge.

Once accepted, the candidate presents the thesis orally. This oral exam is open to the public.

Abstract

Cross-laminated timber (CLT) is increasingly used in mid- and high-rise construction due to its strength, sustainability, and prefabrication benefits. However, its hygroscopic nature makes it sensitive to moisture exposure, particularly when wind-driven rain (WDR) leakage occurs. Therefore, WDR leakage must be considered in moisture-risk assessment of CLT wall assemblies. Hygrothermal simulation tools are commonly used to predict moisture-related risks in building-envelope assemblies. This assessment can be conducted using physics-based numerical models or machine-learning surrogate models. In physics-based modeling, existing tools such as WUFI and DELPHIN often simplify the moisture storage function, a key input for hygrothermal simulations, by neglecting moisture hysteresis. This simplification may lead to inaccurate estimation of moisture risks under dynamic wetting and drying conditions. In machine-learning surrogate modeling, rapid predictions can accelerate hygrothermal assessment; however, model reliability remains uncertain for predicting the nonlinear moisture response of CLT wall assemblies under WDR leakage.

Motivated by these gaps identified from the literature review, this thesis addresses two methodological components. First, a one-dimensional coupled heat and moisture transfer model incorporating moisture hysteresis was implemented in Python. The model was validated using a laboratory dataset from a date palm concrete wall and 18-month field measurements from a full-scale CLT test wall exposed to Vancouver’s climate, with and without controlled water injection. After validation, the model was used to assess CLT wall performance under historical and projected future climates using extreme Moisture Reference Years, WDR exposure, and four wall orientations. In addition to temperature, relative humidity (RH), and moisture content (MC), mold growth index and wood decay were used to evaluate the effects of hysteresis.

Including hysteresis improved agreement with measured MC in the leakage-affected section, reducing the mean absolute error by 1.42 % MC and reducing the maximum absolute difference by 3.12 % MC. Under long-term climate scenarios, hysteresis caused minor orientation-dependent variations in predicted moisture risks but did not considerably change the overall performance trends for the wall configuration and climate scenarios examined in this study.

In the second component, Random Forest (RF), Long Short-Term Memory (LSTM), and Convolutional Neural Network (CNN) surrogate models were evaluated for predicting temperature, RH, and MC at the exterior surface of CLT panels. The models were trained and tested using DELPHIN-generated data covering four CLT wall configurations, including assemblies with vapor-permeable and vapor-impermeable exterior insulation, four orientations, 17 Canadian climates, and two moisture-loading scenarios: no leakage and 1% WDR leakage applied directly to the CLT surface.

Under no-leakage conditions, all models predicted temperature accurately, with R2 values greater than 0.99. For RH and MC, LSTM and CNN outperformed RF, while LSTM achieved high accuracy with shorter training time. Under WDR leakage, LSTM and CNN reproduced the main moisture response trends for assemblies with vapor-permeable exterior insulation. However, prediction accuracy decreased substantially for the more restricted-drying configurations with vapor-impermeable exterior insulation, where leakage caused localized moisture accumulation and delayed drying. In these cases, MC predictions showed R2 values below 0.60. Overall, the results show that LSTM-based surrogate models can support efficient assessment under normal moisture loading and WDR-leakage conditions with sufficient drying potential. However, physics-informed or hybrid surrogate models may be required for restricted-drying conditions dominated by liquid-water transport, moisture accumulation, and delayed drying.