Daylight Solutions for Adaptive-Reuse Heritage Buildings: A Machine Learning Approach

Adaptive reuse of heritage buildings plays a vital role in advancing sustainability by reducing environmental impact and preserving cultural value. However, integrating contemporary functional needs—such as sufficient daylight for precision tasks—often conflicts with conservation principles. Inadequate daylighting in repurposed spaces compromises both energy efficiency and occupant comfort, calling for innovative yet minimally invasive solutions. This study explores ceiling-mounted reflective elements designed for compatibility with heritage conservation criteria: physical integrity, visual coherence, reversibility, and performance efficiency. The methodology integrates parametric design and machine learning (ML) to optimize daylight performance, in line with global sustainability goals. A case study is conducted on a repurposed heritage primary school, now functioning as a fashion design academy requiring elevated illuminance levels. In the first phase, existing daylight conditions are analyzed using Climate Studio, applying key performance metrics such as spatial daylight autonomy (sDA), annual sunlight exposure (ASE), Useful Daylight Illuminance (UDI), and Daylight Glare Probability (DGP). A series of reflective ceiling configurations are simulated to assess performance variation. In the second phase, ML techniques including regression models and neural network architectures are employed to analyze simulation outputs and predict optimal geometries. The research also considers the application of generative models, such as variational autoencoders (VAEs), to propose new geometry configurations informed by high-performing results. The findings demonstrate improved daylight distribution and daylight sufficiency through ML-supported design iteration, offering a replicable, computationally efficient framework for adapting heritage interiors to modern lighting needs while preserving architectural authenticity.