Object-Oriented Graph Grammars for Computational Design Synthesis
Author: Helms, Bergen
Supervisor: Prof. Kristina Shea
Institution: Technische Universität München
Conceptual design is an early design phase that is acknowledged as particularly critical. Its goal is the determination of the product’s essential characteristics that meet the given requirements. Conceptual design is characterized by high uncertainty resulting from lacking knowledge about the future product that makes it di cult to evaluate design quality and to systematically explore the set of solutions. Computational Design Synthesis (CDS) aims at supporting conceptual design through formalization and automation of knowledge-intensive design tasks. However, CDS still has little acceptance in industry due to the high e ort required for knowledge and task formalization, the limited scope of application, the lack of reuse of existing paper-based design knowledge, the lack of modeling standards and tool integration, and the low maturity of software tools. The potential of CDS to increase development e ciency and innovative power motivates addressing these problems. This thesis contributes to the goal of increasing the applicability of CDS in every day design practice with the development and implementation of a hybrid knowledge representation, termed object-oriented graph grammar, and an approach to automatically formalize engineering knowledge from design catalogs.
Object-oriented graph grammars enable the computational synthesis of product architectures on multiple levels of abstraction, i. e. Function, Behavior and Structure. This hybrid knowledge representation allows to capture declarative knowledge in a port-based metamodel and to formulate generic, procedural design rules in a graph grammar. The object-oriented graph grammar approach is implemented in the modular, open-source and platform-independent software booggie. A formal language definition represents the foundation for tool integration through model transformation. A complementary approach is developed to characterize and formalize physical e ects contained in design catalogs. Through an automated analysis of the equation structure, abstraction ports are assigned to physical effects and represent valid mappings between functions and physical effects.
Through the hybrid knowledge representation of object-oriented graph grammars, advances in terms of efficiency and effectiveness of the knowledge formalization are achieved. These contributions are validated through the synthesis of a solution space of automotive hybrid powertrains. The impact of evolving engineering knowledge on the solution space is shown and noteable solutions are identified through the search for specific solution characteristics. The computational generation of aircraft cabin layouts validates the practical usability of object-oriented graph grammars as implemented in booggie in an industrial case study. The automated, equation-based formalization of physical effects makes paper-based engineering knowledge available for CDS. The advantage to computationally reusing this knowledge is validated with the formalization of the physical e ects of two design catalogs and a separate software prototype that searches suitable physical e ects for a given function.
The contributions achieved in this work support the efforts to bring CDS approaches into use in every
day design practice. Besides an increase of the software maturity, future work should include the integration of object-oriented graph grammars and the automated assignment of abstraction ports in one implementation. The extension of the synthesis of product architectures towards parametric synthesis and design evaluation using simulation should be addressed as well. Further, the incorporation of logical reasoners could enable the solution of logical port-matching problems and using model transformation, the transformation to other modeling languages, e. g. SysML, could be realized.