Strategies for generative configuration designs: A knowledge foundation and generative designer assistance tool architecture.

摘要

Design theory researchers agree that conceptual design is an important part of the mechanical design process; the more design alternatives generated, the more opportunities for designers. Today's technology delivers more computational power for desktop generative design than ever before. Research on generating designs provides a platform to use this computational power to create effective designer assistance tools.; Generative design is any automated design process that outputs a number of feasible designs. Generative configuration design (GCD) is an automated process that uses functional requirements as inputs and outputs solid models of feasible designs, integrating traditional conceptual, component selection, and configuration design. Understanding the underlying knowledge foundation of the generative design process is the key to doing broader research on the topic than just building one-of-a-kind prototype systems.; This work presents a partition of the design knowledge space according to function, form, behavior and domain knowledge boundaries. Common design methods and GCD systems developed by the author are mapped to the partitioned knowledge foundation. This mapping leads to an understanding of design process backtracking and suggests a preferred, modular GCD system architecture.; Structuring generative designer assistance tools (GDATs) on the proposed knowledge framework, allows us to control algorithm backtracking processes to improve overall tool efficiency and effectiveness. The efficiency of a generative design algorithm with a constant set of design factors can be increased by implementing rule and constraint application strategies that save CPU time. Guidelines for application strategies and deployment of rules and constraints during GCD are presented. These guidelines maintain the function and form neutrality of the generative design process while and improving its efficiency. These claims are demonstrated with GD-CHAIR, a GCD algorithm built with the preferred GCD architecture.; Contributions of the research include: knowledge structuring guidelines and generating strategies for GCD, a partitioned knowledge model for the conceptual design process, development of improved configuration space size estimates, and condition handling strategies that improve the efficiency and effectiveness of the generative design process.