Exploring for Inspiration: A Bayesian Inference Based Approach to Design By Analogy Retrieval

Georgia Tech Research Corporation, Atlanta GA

Investigators

Abstract

People use analogical reasoning in everyday life to understand new things by drawing parallels with familiar things. The same principle is used by engineering designers in an approach called Design by Analogy (DbA). Many groundbreaking innovations have emerged through DbA, including self-cleaning, outdoor-surface paint that was inspired by the lotus flower and the first printing press, which was inspired by the early wine press. Although analogical inspiration can be arrived at serendipitously, formal approaches for DbA enable reliable the approach to be applied more widely and reliably. To date, most analogy identification techniques have relied on a query-based approaches, such as key-word search. Instead, this research will enable visual techniques for exploring analogies for engineering design innovation and show the superiority of this visual approach over a query-based DbA. The results will advance engineering design and promote engineering innovations by enhancing the ability of engineers to search for analogical inspiration in a way that is more dynamic and exploratory than current techniques. Results will be made available in an open-source format for use, comparison, and integration by researchers across many fields. Outreach efforts will incorporate research ideas about engineering design and DbA into a mechanical engineering Girl Scout badge. The objective of this research is to test the hypothesis that exploration-based approaches are more effective than query-based approaches for retrieval of stimuli for DbA practices. By using Topic Modeling techniques, big data from the US patent database, and visual analytics techniques, the research will transform the way DbA inspiration is found. The results will give designers more intuitive, dynamic access and maneuverability for exploring the design space. Properties of analogical information and external stimuli will be extracted deductively from literature and theory of DbA. Visual, exploration-enabling, manipulable design repositories will be constructed using Latent Semantic Analysis, Latent Dirichlet Allocation, Non-negative Matrix Factorization, techniques in data visualization and Human Computer Interaction design principles; the repositories will be compared to identify which can best support the retrieval stage of DbA. Novel, seeded, hierarchical topic modeling methods will be developed and tested. These DbA properties and techniques will be tested with cognitive studies of the design process. By supporting an exploratory approach at a larger scale of repository size than ever before, transformational effects can be achieved on the path to innovation. This research will enable engineering of complex device designs that lead to products with improved performance, better material use, and fewer resources consumed.