This project studies design space exploration through the use of an interactive evolutionary algorithm. Unlike standard optimization approaches, interactive evolutionary algorithms can incorproate qualitative designer input, guiding but not forcing the user toward high-performing design options that also meet unformulated goals. This is a compelling way to enable designers to integrate structural performance along with architectural goals into the conceptual design process.

Specifically, this project proposes a flexible and generalized framework for a broad range of conceptual structural design problems. This has been implemented in a prototype design tool, structureFIT, which illustrates the use of the framework using two-dimensional truss structures with nodal coordinate design variables. The framework expands the user experience around the interactive evolutionary algorithm by including a project setup mode prior to interactive evolutionary exploration, and a design refinement mode with real-time analysis to be used afterwards.

This project has been integrated along with the Trans-typology Structural Grammar and Automatic Surrogate Modeling approaches into a new and unified computational design approach. For more about the integration of these methods, see the Integrated Conceptual Design Approach project. Additionally, this framework has been studied for effectiveness and use patterns through a usability test of structureFIT. To read more about this, see the Tool Testing: Architects and Engineers project.

This project proposes defining rich and diverse design spaces for conceptual structural design through systems of rules, or grammars. The goal of this approach is to create design spaces that contain structural options of more than one typology, or standard structural type such as an arch or beam. By spanning across multiple typologies, these grammatical design spaces can be explored computationally to consider and compare a wide variety of conceptual design alternatives. This allows designers to brainstorm digitally and use quantitative and qualitative design evaluation methods. Furthermore, by exposing the space between standard typologies, this method can potentially help us discover new and exciting structural forms.

This project specifically prescribes a method for defining grammars that lead to designs that are structurally feasible and broadly varying. The prescription includes requirements for rules that are flexible in how they are applied, that are able to apply multiple times, and that incorporate structural logic. Additionally, the grammar must include a numerical evaluation method for determining structural performance in a quantitative way. Designs are generated using the grammar by applying a series of rules in succession, which can be done manually by a designer, or automatically and randomly by a computer.

To illustrate the power of this approach, this project also introduces an example grammar for generating structural designs for cable-stayed and suspension pedestrian bridges. The resulting designs are shown in the project images. The grammatical design space approach has been combined with the Interactive Evolutionary Framework for design space exploration and the Automatic Surrogate Modeling strategy for design space approximation, and integrated into a unified approach in the Integrated Conceptual Design Approach project.

This project addresses the need to approximate the design space in order to improve the computational speed of performance evaluation during conceptual structural design. While structural analysis for individual structures takes only a few seconds, computing the structural performance in real-time (i.e. continuously) in rapid feedback tools or for hundreds of design options simultaneously in guidance tools can be prohibitively slow. To increase speed, this project proposes the use of statistics-based approximation to predict rather than fully evaluate structural performance. Expanding on an existing approach called surrogate modeling, this project introduces adaptations to make the approach more appropriate for conceptual design.

Existing surrogate modeling strategies are generally carried out by experts to be used in detailed and complex engineering optimization procedures. In contrast, in conceptual structural design, it is important the the approximation process be fast, robust, and easy or automatic, and there is a higher tolerance for approximation error due to the inherent uncertainty early in the design process. This project therefore proposes a strategy for surrogate modeling that uses robust regression methods that don't require much tuning: Random Forests and ensemble neural networks. Additionally, the proposed strategy focuses on reducing error among high performing design options, and weights rank error higher than value error. Finally, this project implements the proposed strategy in a simple, easy-to-use dashboard that builds surrogate models automatically and can be connected to an interactive evolutionary tool or other conceptual structural design tools that require fast structural analysis.

As one example of such a connection, the surrogate modeling strategy is integrated with the Interactive Evolutionary Framework as part of the unified Integrated Conceptual Design Approach. While this strategy was originally formulated for parametric design spaces, a method to use automatic surrogate modeling for the grammatical design spaces defined using the Trans-typology Structural Grammar approach.

This project integrates the computational strategies developed in the Interactive Evolutionary Framework, Trans-typology Structural Grammar, and Automatic Surrogate Modeling into a unified design approach that offers designers performance-based guidance, creative freedom, and rapid interactivity. Combining all three of these approaches is powerful and has not yet been achieved in existing research, both because the approaches come from disparate disciplines and because of the challenges inherent in integration.

A pairwise approach is used to integrate the three strategies, individually addressing each of the three overlapping regions in the Venn diagram to the left. Combining the interactive evolutionary framework with trans-typology structural grammars involves resolving crossover and mutation for a nonparametric design formulation, integrating a grammar-specific analysis engine, and generalizing the user experience to support grammar-based design spaces. Integrating the automatic surrogate modeling approach into the evolutionary framework requires a strategy for when to use the approximation, how to update and adapt it, and how to use it for real-time analysis in design refinement. Finally, combining performance-based surrogate modeling with the structural grammar approach requires deriving common variables or features from grammar-based designs that can be used to build regression models.

This project studies Louis Kahn's celebrated Kimbell Art Museum (1972) in Forth Worth, Texas from a structural perspective, and investigates the role of structural engineer August Komendant, who was Kahn's closest structural consultant for the last 18 years of his career. The Kimbell's dominant feature is a series of 100-ft post-tensioned concrete shells that are an impressive feat of both architecture and structural engineering. The shells have an unusual cycloid-shaped cross section which has led to confusion about their structural behavior (for example, Kahn famously often referred to them as compressive vaults, although their behavior is closer to that of a simply supported beam).

The goal of this project is to clarify the structural behavior of the Kimbell's shells and the structural significance of the celebrated cycloid shape. Additionally, this project aims to shed light on how Komendant's technical contributions impacted and integrated into the final result, which is often heralded as a successful harmony of architectural and structural goals.

This project tested the effectiveness of structureFIT as a conceptual design tool for two populations of designers: architects and structural engineers (in graduate programs at MIT). The study included a brief tutorial, and then the users were given three structural design problems to complete. In addition to evaluating the tool, the project aimed to identify differences in the ways that the two groups used structureFIT; would the engineers focus more on optimization while the architects concentrated more on creative exploration?

In fact, both groups used the tool in similar ways, and both managed to discover new design options that saved an average of 30% of the material of common initial designs. This result confirmed that structureFIT can be an effective design tool for both architects and engineers. Additionally, the study found that compared to a control group using traditional software, both architects and engineers found a diversity of designs that was statistically signficant; in other words, structureFIT helped users to find a wider range of design alternatives. Finally, the study polled users in both groups to determine that they enjoyed using structureFIT more than existing software tools.

This project developed, fabricated, and tested new designs for braced frame lateral systems for tall buildings. The aim of the project was to test out structureFIT on a relatively complex problem, to explore the design space of lateral systems for tall buildings, and to develop a methodology for connecting digital and physical models through testing. Six new braced frame geometries of constant volume were designed using structureFIT, and along with a control design, were then digitally fabricated in polycarbonate using a waterjet cutter. The designs were then load tested to failure by applying a linearly varying force simulating wind load.

The load testing confirmed that the new designs performed similarly to each other and to the control design, while offering significant variation in aesthetic character. Four of the designs outperformed the control in ultimate load, and all six new designs were initially stiffer than the control. The uniformity in results verified what was found in the design stage of this project: that the design space for braced frame structures is shallow, meaning that large changes in design variables has limited impact on structural performance. This means that "optimal" designs can't offer significant savings over conventional designs, but it also means that designers have considerable freedom that can be exploited for architectural reasons.