Enhancing performance-based generative architectural design with sketch-based image retrieval: a pilot study on designing building facade fenestrations

By coupling parametric modelling, building performance (like energy efficiency) simulation, and algorithmic optimization, performance-based generative architectural design (PGAD) can automatically generate lots of high-performance architectural design solutions. Although it is ‘performance-based’, the final selection of a real design project still needs to consider the aesthetics of design choices. However, due to the overwhelming number of design choices generated by PGAD, it is difficult for designers to choose the most favourable one from them. Therefore, the current study tries to integrate the technology of sketch-based image retrieval (SBIR) into the selecting stage of PGAD. Rather than navigating alternatives one from another and getting lost, designers can directly find the most aesthetically preferred one by inputting his/her hand-drawn design. A design project of fenestrating a multiple-floor office building is used to demonstrate this method and test three SBIR algorithms: Angular radial partitioning (ARP), Angular radial orientation partitioning (AROP), and Sketch-A-Net model (SAN). Test results show that AROP performs the best among these three algorithms. Its retrievals are most similar to inquiry images drawn by architects. Meanwhile, performances of AROP with different template combinations are also rated. After that, AROP with the best template is also tested with incompletely drawn inquiry images. In the end, investigation results are validated by another building façade design case. The current study automates the PGAD process stepwise, making it more applicable to real design projects.