• circular economy
• CO2 emissions
• computational design
• emission reductions
• form finding
• freeform surface
• geometric deep learning
• grasshopper
• gridshells
• hops plug-in
• life cycle assessment
• machine learning
• mechanical behavior
• optimization
• recycling
• reuse
• shape optimization
• steel
• stock-constrained design
• structural-based approaches
• sustainability
• sustainable construction

In recent years, the optimization of gridshells has emerged as a prominent topic in research on computational design. The possibility of altering the shape and topology of a gridshell presents a challenging opportunity to maximize material efficiency, as the double curvature and grid of beams enable the spanning of large areas with minimal structural mass. However, the mechanical behavior of a gridshell is influenced by various factors, making its design and optimization tasks that demand complex methods and advanced tools. Recently, there has been growing interest in exploring the potentials of machine learning methodologies to explore such a multi-variable space.
Nowadays, the construction industry stands out as a major contributor to global CO2 emissions, accounting for approximately 37% of total emissions, estimated at 10 GT in 2021. Addressing this issue necessitates a shift towards maximizing the efficient use of existing materials, aligning with the principles of a circular economy. Strategies like repair, reuse, and recycling play crucial roles in minimizing waste and environmental impact. While recycling steel has become common, the practice of reusing steel remains less widespread. However, the reuse of structural steel offers significant advantages by circumventing the environmentally intensive manufacturing phase. The suitability of steel for reuse is underscored by its reversible connection principles, standardization, and availability of certification. Combining the mechanical performance of a gridshell with the reuse of steel components represents a promising avenue for sustainable construction practices.

This thesis aims to explore a novel design paradigm that enhances the design of gridshell structures by combining stock-constrained steel beam members from a dismantled building to improve eco-performance, assessed through Life Cycle Assessment with a focus on embodied carbon emissions. Member sizing is constrained to stock availability, while shape optimization through geometric deep learning explores rationality and structural-based approaches. The computational tool developed for this purpose utilizes Grasshopper to interface with an external Python-coded optimizer via the Hops plug-in. This setup enables the geometric deep learning algorithm to adjust the shape of the gridshell by shifting the nodes, considering the best-fit allocation of stock to minimize embodied carbon emissions and cutting processes for each beam member.
Testing the computational tool across different stock scenarios, both ideal and real-world case studies reveal significant reductions in emissions of beam members under optimal reuse conditions. In particular, dismantled trusses from an industrial building owned by the Piaggio company are used to build the elements’ stock. The results underscore the importance of factors such as stock availability and reuse rates in achieving emission reductions.
Furthermore, the study incorporates an estimation of emissions for nodes. Accordingly, while steel beam members play a substantial role in reducing carbon emissions, nodes can sometimes offset these reductions due to increased mass resulting from larger cross-section configurations.
In conclusion, the development and application of computational deep learning optimization hold immense potential for enhancing the eco-performance of gridshell structures through the strategic reuse of stock-constrained steel beam members, ultimately contributing to more sustainable practices within the construction industry.