“What is more likely to be associated with a reduction in emissions: switching from concrete to timber, or shortening the spans throughout the building?” While such insights are valuable for mitigating emissions from structural systems during early stages of design, it is difficult to answer these types of questions in current paradigms of performance-driven design. This dissertation makes several original contributions to the system-level design of low-carbon structures.

First, a literature-supported network of strategies available to reduce emissions during early-stage structural design is established and evaluated on the bases of literature availability, impact, implementability, and compatibility. Material efficiency and material choice represent two key levers for reducing emissions in structural design, but it is difficult to navigate trade-offs between these strategies at a system level of structural design. Holistic design strategies can help achieve this, but these current paradigms of performance-driven design (e.g. deploying rules of thumb, comparing a few design options, and optimization) are limited in their capacity to inform decision-making towards higher performing designs. There is a particular opportunity to produce these insights using data-driven approaches given the growing quality and quantity of data in the field of low-carbon structural design.

In response, this dissertation analyzes both types of data that are available in the field: wild data (measured from the industry) and synthetic data (produced from bottom-up parametric structural models). Data from over 200 fully designed structural systems from a structural engineering firm are analyzed. This analysis is the first to 1) provide empirical evidence for floors and foundations representing the largest opportunities for carbon reductions and 2) evaluate the relationship between structural material quantities and embodied carbon in structural systems (many analyses evaluate the latter without the former). In a field where material choice is a predominant impression for reducing emissions, these new insights importantly affirm the prominent role of material efficiency in reducing a structural system’s emissions.

While the design space of wild data includes a diverse variety of projects, leveraging a synthetic dataset computed from a bottom-up parametric model helps produce insights specific to the design problem at hand. The final contribution of this dissertation is to propose a computational framework that leverages data to empower decision-making in design. The framework addresses two challenges: 1) the challenge of extracting decision-making insights from design data, and 2) the challenge of comparing decision-making across continuous (numerical) and categorical variables, which are typical in most design problems. In this framework, a machine learning model is trained on a provided set of design data to compute gradients across the design space. These gradients are distilled into “influence metrics”, which offer a novel, accessible way to build and supplement intuition on low-carbon design decisions. A few case studies in low-carbon structural design are presented to demonstrate the use of the proposed method with synthetic datasets. By striking a meaningful balance between applying rules of thumb and optimization, the method empowers a paradigm shift from performance-driven design to performance-informed, human-driven design.

THESIS COMMITTEE

• Caitlin T. Mueller, MIT (advisor)
• Josephine V. Carstensen, MIT
• John A. Ochsendorf, MIT
• Kate Simonen, University of Washington

Fang, Demi. “System-Level Design of Low-Carbon Structures.” PhD, Massachusetts Institute of Technology, 2024.