A methodology for supporting design grammar development and application in computational design synthesis
Author: Königseder, Corinna
Supervisor: Prof. Dr. Kristina Shea
Institution: ETH Zurich, Switzerland
Page(s): I-XIII; 1-162
In the last decades researchers have developed various methods enabling computers to solve design synthesis problems using engineering design grammars. Successfully applying grammar-based Computational Design Synthesis (CDS) methods is challenging for several reasons. Problem-specific knowledge has to be represented formally through models describing designs and grammar rules that define design transformations. For a fully automated CDS process, evaluation routines have to be provided to enable the comparison of generated designs and a search algorithm is required to guide the synthesis process. The success of CDS methods usually depends on both the development of a suitable grammar and the selection and tuning of an appropriate search algorithm since the two are strongly coupled. Nowadays, there is limited methodological support for human designers during grammar development and selection and tuning of a search algorithm for CDS. This thesis investigates a methodology to support human designers in the CDS process. The focus is on supporting grammar development and application through automatically generating and analyzing data to provide the designer with information on grammar rules, search algorithms and their interrelations. This can increase the understanding of the CDS process which in turn can give rise to improved grammars and beneficial combinations of grammar rules and search algorithms that lead to better-quality synthesis results. As a long term goal, this can help to bring the benefits of CDS methods to industrial applications. The methodology consists of four methods that allow human designers to systematically analyze the CDS process. First, the Grammar Rule Analysis Method analyzes developed rules and the influences they have on synthesized designs considering objectives and design characteristics. Second, the Network-based Rule Analysis Method supports the analysis of application conditions of rules and the analysis of sequences of rule applications, e.g., to identify beneficial or counterproductive rule sequences. Third, the Relation Visualization method enables the analysis of both the grammar itself and how search algorithms explore and exploit the design space. Finally, the Search Strategy Comparison Method supports human designers in selecting a strategy that decides whether to apply topologic or parametric rules throughout the synthesis process. For each method, a process to generate data, analyze it and provide visualizations for the human designer is defined inspired by research in visual analytics. Information on interactions among grammar rules, rule sequences, changes in objectives and design characteristics, and the search process are presented to the human designer. A software prototype is developed to evaluate the methods and show their potential on different case studies. The presented methodology is integrated in the process of grammar rule development and contributes to supporting grammar development, selection of a search algorithm and refinement of the search process. The results of the case studies show not only the potential of the methodology to support the human designer in grammar development and application, but also to increase the understanding of links between problem representation, search and the explored design space.