visual comfort l(AI)outs

Abstract

The interior layout of apartments is made in the early design stage of an architectural project when the decisions can significantly impact the building's performance. During the early design phase, the ability to impact an architectural project is the most important, and this phase serves as the foundation of subsequent design phases. Proper daylighting improves visual comfort and minimises the dependency on artificial lighting. Combining good natural (day)lighting with a greenery view substantially affects the health and well-being of the building occupants. Optimising the layout of apartments based on daylight and view in the early stage of the building is crucial to ensure visual comfort. In this regard, artificial intelligence presents the potential to provide valuable support for performance-based decision-making in interior zoning based on daylight and view. However, there is currently a lack of machine learning methods to support designers in making informed decisions regarding early interior design decisions that affect daylight and view quality.
The performance of daylight and view quality significantly impacts the overall quality of residential spaces. The EN17037 guideline ensures the quality of indoor spaces by providing specific requirements for residential spaces regarding the view and daylight quality. The national annexe of the UK for daylight in dwellings should be included to ensure that daylight requirements align with the specific purposes of different rooms in dwellings. Incorporating adequate daylight exposure and good views in residential spaces promotes a connection to the natural environment, contributing to overall satisfaction and a higher quality of life for occupants, and leads to lower energy consumption and a smaller carbon footprint in buildings. Significant design parameters that impact the performance of daylight and view in residential spaces include the building orientation, window fenestration and the interior layout arrangement, specifically the room type orientation. Optimising the layout for optimal use of daylight and views is crucial for creating well-designed residential spaces that promote well-being, energy efficiency, and sustainability.
A novel ML design process workflow has been proposed to integrate ML models seamlessly into the architectural design process. Designers upload their layout designs into a dedicated tool, where the layout designs are pre-processed for compatibility with the ML model. Subsequently, the ML model predicts daylight and view values, which are then translated into practical visual representations during an after-processing step. A multimodal machine learning model utilising a ResNet and fully connected network is the most effective for predicting daylight and view quality in residential spaces. An ML model is trained using one image feature and five numerical features to predict the median daylight illuminance on the 21st of March, July and December and the p80 for ground and sky view inside a room. The trained model achieved a test loss MSE of 0.0047 and a test MAE of 0.0440 for the prediction of the three daylight labels, a test loss MSE of 0.0057 and a test MAE of 0.0478 for the prediction of the two view labels. An optimisation step identifies the optimal apartment layout based on a layout evaluation method guided by EN17037 requirements. A multifaceted approach is suggested for evaluating and improving residential layouts for visual comfort, incorporating a novel assessment system that evaluates daylight, view, and room orientation quality in each room to assess the overall apartment layout's visual comfort comprehensively. Overall, this framework represents a significant advancement in integrating ML models into architectural workflows by systematically evaluating daylight, view quality, and room orientation, providing visual feedback, and offering optimisation suggestions that align with contemporary design standards and requirements.