Reality Data Characterization of Recovered Construction Materials for Generative Design

Abstract

This paper explores potential AI-supported design and construction pipelines for irregular shape buildings after characterizing 3D object features of salvaged building materials. The purpose is to support researchers to achieve goals of generative design, fit-to-a-BIM design, and procurement optimization through well-identified building elements. Steps in redesigning pipelines include data capture, semantic segmentation, and object classification for 3D point clouds and 2D images of recovered materials. Potential to extend these methods to point clouds and images of parts of existing buildings is considered in future work as well. Specifically, object characterization by means of computer vision is discussed in terms of the derivation of critical dimensions, geometric properties, and element conditions. Based on the information generated from these processes, generative design approaches for primarily geometrically constrained design problems are explored first. Next, Rules-based BIM design tools are considered for the potential size fitting method. Finally, a conception for optimizing the procurement process for reused components is proposed. The application of these processing pipelines is practically demonstrated for a simple case of dividing wall material selection and generative design utilizing salvaged bricks in the platform of Visual programming design platform (Grasshopper). It is anticipated that these approaches would be extended in the future to more classes of materials and to more critical design and procurement problems, such as structural design as well as value estimation.