Show simple item record

dc.contributor.advisorGoel, Ashok K.
dc.contributor.authorWiltgen, Bryan Joseph
dc.date.accessioned2019-01-16T17:25:16Z
dc.date.available2019-01-16T17:25:16Z
dc.date.created2018-12
dc.date.issued2018-11-09
dc.date.submittedDecember 2018
dc.identifier.urihttp://hdl.handle.net/1853/60805
dc.description.abstractDesign is the process by which solutions get developed to solve social challenges, and its products can be seen across our world from toothbrushes to computers to spaceships. Conceptual design is an early phase of design where an initial candidate solution gets developed. Analogical design is a form of design where knowledge from some known source case is transferred into the design solution, leveraging knowledge not directly related to solving the problem at hand. Biologically inspired design is a real-world form of analogical design where source cases come from nature. In all of this, there is a need for design revision. The output (the candidate design solution) likely begins in an imperfect state and needs to be revised to ensure it properly and completely solves the design problem. Additionally, the designer's understanding of any source cases or prior solutions used could also exist in an imperfect state and should too be revised since they are used in reasoning. Thus, any flaws in this knowledge could negatively impact the process and/or outcome. This is especially important in an interdisciplinary context like biologically inspired design where a given design practitioner may have expertise in only some of the involved domains. To support design revision, this dissertation presents an AI agent that can help a design practitioner engage with a process of incremental design revision. In this hypothesized, iterative process, a practitioner externalizes their knowledge of a design concept in a functional model, gives this model to the AI agent that evaluates the model, and then receives feedback from the agent. The practitioner can then decide to revise their model based on this feedback, and then can restart the cycle or conclude it. Specifically, this dissertation addresses the task of model evaluation to support incremental design revision in the context of the conceptual phase of biologically inspired design. It presents the Design Evaluation through Simulation and Comparison (DESC) AI agent, a computational approach to evaluate the candidate design, biological source cases, and prior designs in this context. More precisely, DESC evaluates functional model articulations of these concepts in the form of Structure-Behavior-Function Star (SBF*) models. These functional models act as proxies for the designer's understanding or vision of the modeled designs. DESC does its evaluation through two techniques: Simulation and Comparison. Simulation evaluates the behaviors (processes or mechanisms) and functions (intended or perceived purposes) of a model by simulating the behaviors and determining if those results conflict with claims made in the model. This checks the internal consistency of the model--the extent to which the model's parts agree with each other--for the language rules understood by the simulator. Comparison evaluates the complete model (i.e., the structure submodel, the behaviors, and the functions) by comparing it to another model of the same topic and identifying differences, which are potential areas of misunderstanding or misrepresentation. This checks the external consistency of the model--the extent to which it accurately represents a concept in the world or (in the case of the candidate design) a proposed vision. As part of the Comparison technique, this dissertation also explores a novel analogical mapping algorithm called Compositional Mapping for matching elements between two models. This dissertation evaluates DESC in two ways. First, experimentation was done in the form of a computational ablation-like study to illustrate DESC's ability to evaluate models using Simulation and Comparison. The results support the hypotheses that these techniques can evaluate the internal and external consistency of models and that the SBF* modeling language has syntax and semantics that support automated simulation. This computational experimentation also partially supported Compositional Mapping's advantage over a more conventional approach to mapping models for the purposes of Comparison. The second form of evaluation was a pilot study with human participants that tested the usefulness of DESC in helping humans construct better models and improve their understanding of a design concept thematically related to biologically inspired design. In other words, it explored at a high level to what extent DESC can support the hypothesized incremental design revision cycle. The results of this study partially supported the hypothesis that DESC can support incremental design revision, showing that participants with DESC tend to produce superior models compared to those without it. However, this did not translate to improved understanding. To recap, this dissertation presents an AI agent called Design Evaluation through Simulation and Comparison (DESC) for evaluating designs in the form of SBF* functional models, contextualized in a hypothesized incremental design revision process in the conceptual design phase of biologically inspired design. Evaluation of this AI agent supports that it can indeed evaluate models and partially supports its usefulness in supporting incremental design revision, with analysis that suggests people with an implementation of DESC produce better models than those without it. The overall evaluation of DESC through the computational experimentation and the usefulness study presents promising results towards the broader incremental design revision goal.
dc.format.mimetypeapplication/pdf
dc.language.isoen_US
dc.publisherGeorgia Institute of Technology
dc.subjectDesign
dc.subjectConceptual design
dc.subjectBiologically inspired design
dc.subjectArtificial intelligence
dc.subjectAnalogical mapping
dc.subjectSimulation
dc.subjectEvaluation
dc.subjectIncremental design revision
dc.titleIncremental design revision in biologically inspired design
dc.typeDissertation
dc.description.degreePh.D.
dc.contributor.departmentInteractive Computing
thesis.degree.levelDoctoral
dc.contributor.committeeMemberNersessian, Nancy
dc.contributor.committeeMemberRiedl, Mark
dc.contributor.committeeMemberRugaber, Spencer
dc.contributor.committeeMemberYen, Jeannette
dc.date.updated2019-01-16T17:25:16Z


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record