Explanations-based support for early design informed by a design study

Noel Titus, Purdue University

Abstract

Conceptual design is regarded as one of the most critical phases of the product design process since as much as 70 percent or more of the product cost is committed during the early design phase viz. conceptual and embodiment design. The decisions taken during early design are the most influential of any design stage. Despite these facts, tools to support the early design stages of conceptual and embodiment design are lacking. This work endeavors to support aspects of early design using a two pronged approach. The first aspect involves studying the activities of designers throughout the entire product development lifecycle in order to understand how they affect design outcome and using the results of the study to identify opportunities for computer based support in early design. The operations performed by designer during early design are translated to operations that a computer model of the design should be able to perform using the results from the afore mentioned design study and results from design literature. The second aspect involves implementing the findings from the design study to computationally support the decision making activities occurring during early design using an interval-constraints based model with feedback, called "explanations", regarding the implications of the decisions taken by the designer. A design study that was conducted on US high school robotics design teams participating in a national competition, involving both experts from industry and novices in high schools, solving electromechanical problems of moderate complexity. The study covers sixteen design activities previously identified as commonly occurring in design practice and seeks to understand their influence on design success. The study provides a broad, integrated perspective of how design activities individually and collectively affect the design outcome. Of the sixteen activities, aspects related to nine activities were found to have statistically significant influence on the design outcome. The reliability of using retrospective protocols to study design process is also studied, drawing on Simon's problem-solving model to help understand the validity of design information provided by subjects along the dimensions of time, memory, cognitive process type and information class. The study indicated that using mathematical models during early design resulted in better design outcomes. Using mathematical models during early design reduces the iterations during design, provides an effective framework for comparing design alternatives and provides insight into design decisions. In order to computationally support early design, the design problem is formalized as an interval-constraints satisfaction problem and constraint propagation is used to reason on the constraint network in order to inform the designer of the implications of design decisions in terms of changes to the domains of design variables. Furthermore, when incompatible decisions result in inconsistencies in the design, "explanations" are provided to the designer in terms of the smallest set of incompatible design variable instantiations, called minimal conflict explanations, and in terms of alternate design variable instantiations that will result in a restoration of consistency to the constraint network, called corrective explanations. The interval-constraint based representation and explanations algorithms are implemented in an interactive prototype to support early design.

Degree

Ph.D.

Advisors

Clark, Purdue University.

Subject Area

Mechanical engineering

Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server
.

Share

COinS