Machine learning-optimized porous thermally responsive SS-PCM with switchable transparency for adaptive building envelope coatings

Zhiying Xiao, Rajae Bousselham, Mingjiang Tao, Sergio Granados-Focil, Oren Mangoubi, Steven Van Dessel,
Machine learning-optimized porous thermally responsive SS-PCM with switchable transparency for adaptive building envelope coatings,
Energy and Buildings,
Volume 349,
2025,
116593,
ISSN 0378-7788,
https://doi.org/10.1016/j.enbuild.2025.116593.
(https://www.sciencedirect.com/science/article/pii/S0378778825013234)
Abstract: Buildings account for nearly 40% of total energy consumption, with a significant share of heating and cooling demand arising from building envelopes. Conventional passive envelopes—such as cool roofs, radiative cooling surfaces, glazing systems, and passive solar walls—cannot automatically adapt to environmental conditions by switching between heating and cooling modes. To address this limitation, we propose a passive, adaptive building envelope coating system that responds to ambient temperature changes without external energy input. The system integrates solid–solid phase change materials (SS-PCM), Polydimethylsiloxane (PDMS) and Silver (Ag), enabling switchable radiative cooling and solar heating effects. We investigated the influence of porous structures on the optical and thermal performance of the SS-PCM system and implemented a machine learning-based optimization method to enhance its optical properties. The results: (1) confirmed that PDMS/SS-PCM/Ag coating system exhibits a switchable radiative cooling effect and a solar heating effect in response to the ambient temperature, while the net power exceeds that of the reference PDMS/Ag system by ∼250 W/m2 in heating mode; (2) indicated that random forest models can capture the complex relationships between the porous features and optical properties of the PDMS/SS-PCM/Ag systems, achieving prediction R2 values greater than 0.8; and (3) identified two potentially optimized porous PDMS/SS-PCM/Ag coating systems, derived from machine learning analysis, each enhancing either the heating or the cooling power by 20–50 W/m2 compared to the non-porous PDMS/SS-PCM/Ag system.
Keywords: Phase change materials (PCMs); Radiative cooling; Machine learning; Adaptive building envelopes (ADE)