Constraint-Based Methods for Automated Computational Design Synthesis of Solution Spaces

Year: 2015
Author: Münzer, Clemens
Supervisor: Prof. Dr. Kristina Shea;
Institution: ETH Zurich, Switzerland
Page(s): I-XII; 1-131

Abstract

Computers have the capability to support human designers in a variety of tasks. This
includes not only releasing the human designer from routine tasks by design automation
but also sparking and supporting innovation and creativity in development processes. In
order to support the concept phase effectively, a wide range of possible concepts, which
are quantitatively evaluated, should be considered to enable designers to explore the solution
space and to make advantageous decisions towards concepts to be considered in
consecutive development phases. To enable an automated systematic solution space exploration
and evaluation, this research presents different approaches based on a graphbased
object-oriented knowledge representation. This representation is combined with
first-order logic and Boolean satisfiability as foundation for a generic automated approach
for requirement-driven computational design synthesis of solution spaces. To enable the
evaluation of the generated solution spaces, a generic approach to automatically translate
the generated graph-based product concepts into Bond graph-based simulation models is
described. Finally, a method is presented to parametrically optimize the generated concepts
using simulated annealing. Here, parameterizations are generated by automatically
setting up and solving constraint satisfaction problems and evaluated using the generated
simulation models. The methods are validated on the case studies of chemical process
engineering, automotive powertrains and 3D-Printer kinematic mechanisms. The main
contributions of this research are a continuous and generic approach starting with task
definitions and ending with a valid, parameterized product concept, a method which is
able to determine if an engineering task is solvable for a given set of synthesis building
blocks, and an approach for a generic transformation of the generated product concepts
to Bond graph-based simulation models. Thus, this research provides new knowledge
in terms of generic transformations between different knowledge representations in order
to generate, explore and evaluate large solution spaces with an, until now, unreached
expressiveness.