Design Engine for 3D-Printed Sand Mould and Metal Casting
Neuronal Stool investigates the role of computation in the integration of fabrication constraints into generative design. It used metal casting stools for exploration and developed a design engine, which operation is divided into the following steps: 1. Assigning parameters for human and machine to set up, 2. Categorizing visually similar designs to better navigate the design space, 3. Visualizing the design that best fits with the user-specified criteria. Contrary to the traditional seating design that distinguishes between seating surface and structure, our approach promotes heterogeneity and differentiation of material properties, therefore exploring novel aesthetic possibilities through computational design.
Computation Design and 3D Printing
Neuronal Stool is an open-ended exploration of the mathematics and logic behind Nature with the inspiration of nervous systems. It pushes the limit of geometrical complexity by generating synthetic and organic forms through customized algorithms of a shape design engine without any manual intervention. These forms support lightweight construction and create a seating surface that follows the natural curves of the human body to provide comfort and flexibility.
To generate feasible designs, the engine uses agent-based modeling that incorporates fabrication constraints (the maximum aluminum traveling distance, the minimum diameter of the mould, etc.). Once users set the specification of the stool, the engine generates forms with random parameters and clustered them by their visual similarities and used design images as the input for unsupervised learning. Users then choose their favourite categories and define performance criteria (combination of castability, stability, and cost) based on what the engine finds optimal. This way, the design engine allows users to balance between design freedom and fabrication constraints in reaching their favourite designs with the desired properties. As fabrication is integrated into the design process, these design outcomes are ensured to be castable.
For the actual casting, the bottoms of the stool legs were designed as the sprues. The stools were cast upside down so the aluminum was poured into the 3D-printed sand mould through the stool legs as hollow channels. The metal and mould were then cooled, and the metal part (the casting) was extracted. Results of this project are the shape design engine and prototypes of two cast stools it generated. The dimensions of the cast stools are 450*450*450mm and 350*350*600mm, each weighted 5.6kg and 6.2kg, respectively. We will continue generating designs for stools with more advanced functions and greater geometrical complexity in the near future.
In perspective, the next step of developing the project to optimize the design engine to produce stools with better performance by adding the step of performance evaluation to the form-generating process, allowing the engine to achieve better design results with timely feedback. The engine could be trained with reinforcement learning to narrow down the design space that better fit users’ performance criteria. It could also be trained to create the design from users’ rough sketches of the objects using GAN (generative adversarial network.)
Neuronal Stool aims to meet the need for designing seating objects with effective, customized algorithms. Its design engine integrates topology optimization and deep generative models in an iterative manner to explore new design options with limited initial design settings and data. As shown by the resulted prototypes, the project manifests greater diversity and robustness of generated designs in new aesthetics.
Project Credits: ZongRu WU and Haruna Okawa.
Supervisors: Mania Aghaei Meibodi (DBT, Senior Research), Prof. Dr. Benjamin Dillenburger (DBT, Principal Investigator).
Acknowledgement: The casting of the stools was made possible by support from Christenguss AG.
We would like to especially thank Florian Christians (Geschäftsführer) and Milot Shala (Betriebsleiter).
Photo Credits: Jetana Ruangjun, ZongRu WU, Haruna Okawa.