Typogenetic Design is an interactive computational design system combining generative design, evolutionary search and architectural optimisation technology. The active tool for supporting design decisions during architectural shape generation uses an aesthetic system to guide the search process. This aesthetic system directs the search process toward preferences expressed interactively by the designer. An image input as design reference is integrated by means of shape comparison to provide direction to the exploratory search. During the shape generation process, the designer can choose solutions interactively in a graphical user interface. Those choices are then used to support the selection process as part of the fitness function by online classification.
Enhancing human decision making capabilities in human-in-the-loop design systems addresses the complexity of architecture in respect to aesthetic requirements. On the strength of machine learning, the integral performance trade-off during multi-criteria optimisation was extended to address aesthetic preferences. The tacit knowledge and subjective understanding of designers can be used in the shape generation process based on interactive mechanisms. As a result, an integrated support system for performance-based design was developed and tested.
Closing the loop from design to construction using design optimisation of structural nodes in a set of case studies confirmed the need for intuitive design systems, interfaces and mechanisms to make architectural optimisation more accessible and intuitive to handle. This dissertation investigated Typogenetic Design as a tool for initial morphological search. Novel instruments for human interaction with design systems were developed using mixed-method research. The present investigation consists of an in-depth technological enquiry into the use of interactive generative design for exploratory search as an integrated support system for performance-based design. Associated project-based research on the design potential of Typogenetic Design showcases the application of the design system for architecture.
Generative design as an expressive tool to produce architectural geometries was investigated in regard to its ability to drive initial morphological search of complex geometries. The reinterpretation of processes and boosting of productivity by artificial intelligence was instrumental in exploring a holistic approach combining quantitative and qualitative criteria in a human-in-the-loop system. The shift in focus from an objective to a subjective understanding of computational design processes indicates a perspective change from optimisation to learning as a computational paradigm.
Integrating learning capabilities in architectural optimisation enhances the capability of architects to explore large design spaces of emergent representations using evolutionary search. The shift from design automation to interactive generative design introduces the possibility for designers to evaluate shape solutions based on their knowledge and expertise to the computational system. At the same time, the aesthetic system is trained in adaptation to the choices made by the designer. Furthermore, an initial image input allows the designer to add a design reference to the Typogenetic Design process. Shape comparison using a similarity measure provides additional guidance to the architectural shape generation using grammar evolution.
Finally, a software prototype was built and tested by means of user-experience evaluation. These participant experiments led to the specification of custom software requirements for the software implementation of a parametric Typogenetic tool. I explored semi-automated design in application to different design cases using the software prototype of Typogenetic Design. Interactive mass-customisation is a promising application of Typogenetic Design to interactively specify product structure and component composition. The semi-automated design paradigm is one step on the way to moderating the balance between automation and control of computational design systems.
Examiners: Prof Jonas Runberger, Prof Leon Sterling Supervisors: Dr phil. habil. Marcelo Stamm, Prof Jane Burry, Dr Andy Song