Bayesian-based optimization of concrete infill pattern for enhancing thermal insulation performance
Hanmo Wang, Sunmi Shin, Alexander Lin
Abstract
This study explores the impact of vertical and horizontal configurations on thermal insulation in cellular concrete brick design, aiming to identify optimal insulation patterns. Results indicate that under a constant volume (67%) of coconut fiber, appropriate geometric changes can reduce thermal conductivity by around 10% (from 0.198 W/(m·K) to 0.178 W/(m·K)). Bayesian inference is employed to construct a bi-directional network, providing a more intuitive understanding of variable relationships. A probabilistic-driven search space reduction approach is proposed, improving candidate selection efficiency and reducing the number of assessments. The study introduces a Bayesian Genetic Algorithm (BGA), which outperforms the genetic algorithm when the mutation rate is 0.1.