Envoltórios arquitetônicos complexos : avaliação sistêmica de construtibilidade com aprendizado de máquina

Abstract

The design of architectural envelopes with double curvature presents significant challenges related to constructability analysis. Although several studies address the generation, fabrication, and assembly of such forms, a gap remains in understanding how geometric characteristics impact the constructability of architectural envelopes. This research develops a method for investigating the constructability of architectural envelopes with complex surfaces during the design phase, considering geometric characteristics, fabrication methods, and structural materials through systematic exploration of the solution space and surrogate modeling using machine learning. An architectural envelope with double curvature was defined, featuring a quadrilateral mesh structured by arches and purlins made of plywood, with polycarbonate panel infill. Algorithmically generated parametric models were used to obtain a sample space of envelope solutions, which served as the basis for analyzing fabrication and structural performance and for identifying constructively feasible instances, as well as for training machine learning models. The results indicate that envelopes with continuous arches exhibit superior structural and fabrication performance, and the study also yields performance indicators, correlations between geometric parameters, fabrication characteristics, and structural performance, in addition to machine learning models.