AI-supported form-finding approach to enhance multi-performance efficiency in Voronoi-based parametric modeling

Purpose Parametric modeling enables complex shape creation in architecture, engineering and construction, improving building performance. With advanced manufacturing, form-finding has become crucial for identifying efficient, high-performance topologies. This paper aims to present an innovative form-finding approach supported by machine learning techniques embedded in parametric modeling. This approach overcomes the traditional limitations of shape generation, which typically focuses on adjusting one parameter at a time. Design/methodology/approach The proposed approach involves training an artificial neural network (ANN) system to support parametric modeling. To this end, different 3D models are generated using parametric modeling and Voronoi-based algorithms. Subsequently, the ANN system is used to generate new configurations that enhance thermal and mechanical performance. Findings The study demonstrates that the proposed hybrid system, which combines ANN training with the parametric modeling process, can support form-finding for a building component by improving thermal and mechanical performance while providing information on material consumption. Practical implications The successful demonstration of the system’s feasibility for the novel ANN-supported form-finding approach opens up new opportunities for efficiently modeling building components, benefiting both scientific research and industry applications. Originality/value The existing literature primarily focuses on mechanical topology optimization or form-finding using simplified linear analysis. To the best of the authors’ knowledge, this proposed work is the first to integrate a machine learning system with the Voronoi algorithm, allowing complex cellular configurations to be controlled through a limited set of seed points. This synergy establishes a novel form-finding approach, distinct from existing methods, enabling the enhancement of multiple, highly diverse performance criteria.