Solving engineering design problems through a combination of generative grammars and simulations

In this thesis a framework is proposed which considers the use and the applicable different knowledge levels at various abstractions within an automated design process. For an effective utilization of available design information and knowledge, the computational synthesis process is divided into three main phases: search, optimization and modification. The generative graph grammar for representing design knowledge is used but some aspects are applicable to other representations as well. The generality and flexibility of the proposed mechanism is demonstrated by automating the synthesis of three engineering design problems in two domains. For all cases the graph grammar interpreter, GraphSynth, is used to carry out graph transformations, which define different topologies for a problem. The proposed method combines generative design synthesis methods with conventional simulation models, leading to a significant reduction in the numerical operations in all three design problems. The effectiveness of the proposed method in all of the cases is checked by solving and comparing a variety of available test problems found in the literature. Furthermore by solving complex large scale three dimensional problems, the robustness of the method is tested. The results show that the approach not only creates the existing solutions for available test problems, it creates new structures that have never been seen before. The contribution achieved in this work provides a mechanism for designers to utilize design information and knowledge at all abstraction levels. Besides applying and testing the framework in other domains and design cases, future work may include the improvement of search strategies, which are used for exploring the design space to achieve faster results in larger design spaces.