皇家华人

Research

Subject

Genetic Algorithms and Machine Learning with Spatially Varying Micro Properties of Materials and Fabrication Constraints for Digital Design and Making听

First and second supervisors听

• Prof Marjan Colletti
• Prof Stephen Gage

Abstract

This PhD research investigates ways of creating a co-creative and generative design and making environment, based on a strategy similar to biological construction as a series of events enabling a final design, while co-operating with algorithms, materials, and fabrication constraints from an earlier stage for maximised integration. It proposes using algorithms, including genetic algorithms (GAs) and machine learning (ML) with spatially varying micro properties of heterogeneous materials, as a way to design and build, particularly using GAs to design/optimise a solution and ML to learn and generate materials.

It is structured around three chapters: File for algorithms, Forest for materials, and Factory for fabrication constraints. The File chapter focuses on instructions and their contemporary version of algorithms. The Forest chapter focuses on materials, in particular wood as an anisotropic material and marble as an isotropic in two different scales of human and microscopic, with discussion of spatial autocorrelation in natural materials and the designing of synthetic microstructures. The Factory chapter concentrates on the selected fabrication method of three-axis computer numeric controlled (CNC) milling machines and investigates ways of integrating and enhancing the fabrication constraints into a co-creative, generative design and making process.

The case studies presented throughout the research听first investigate methods to maximise the integration among the elements of design, including instructions, microstructures of materials, and fabrication constraints. Secondly, they explore ways of amplifying the morphological involvement of these elements to maximise co-creativity. While transforming the entire production process into one refined, sophisticated single phase, the thesis creates integrated design pipelines that address the gap and challenge pre-existing forms. In the conclusion chapter, the case studies are analysed qualitatively and quantitatively across eleven specific categories. The performance of each case study is measured and evaluated using Ashby diagrams to assess the trade-offs between these categories and position them in a multi-dimensional space.听

Biography

Seda Zirek is a designer, researcher, and lecturer specialising in design computation, co-creative digital design tools, digital fabrication, complex system definitions, and new methods of modelling and simulating using machine learning and evolutionary algorithms. She pursued her master's degree in architecture at GSAPP, Columbia University (MSc AAD) and earned her PhD in Design from the Bartlett School of Architecture, UCL. Prior to that, she worked as a computational advisor and senior designer for Rafael Vinoly Architects and Zaha Hadid Architects. In 2012, she founded SZD, a London-based design studio focusing on residential projects and product design. She has taught architecture at various institutions, including Yale University, the Bartlett, Columbia University, and London South Bank University.听

Publications

• Zirek, Seda. (2023) 鈥淪ynthesising 3D Solid Models of Natural Heterogeneous Materials from Single Sample Image, Using Encoding Deep Convolutional Generative Networks鈥�. Systems and Soft Computing, 200051, Elsevier.听
  听

• Zirek, Seda. (2023) 鈥淏ottom-up Generative Up-Cycling: A Part Based Design Study with Genetic Algorithms鈥�. Results in Engineering 18: 101099.听
  

• 1.3.5 Ziggy Table Iteration 2 was presented as a poster in the Advances in Architectural Geometry Conference in 2014. 听

Image:听Material Synthesis from a Single Sample Image through Deep Convolutional Generative Adversarial Networks by Seda Zirek听