Date of Graduation

12-2018

Document Type

Thesis

Degree Name

Master of Science in Mechanical Engineering (MSME)

Degree Level

Graduate

Department

Mechanical Engineering

Advisor/Mentor

Jensen, David C.

Committee Member

Couvillion, Rick J.

Second Committee Member

Sha, Zhenghui

Keywords

Adaptation; Biologically inspired design; Design; Latent dirichlet allocation; LDA; Perplexity

Abstract

The objective of the research this thesis describes is to find a way to classify text-based descriptions of biological adaption to support Biologically Inspired design. Biologically inspired design is a fairly new field with ongoing research. There are different tools to assist designers and biologists in bio-inspired design. Some of the most common are BioTRIZ and AskNature. In recent years, more tools have been proposed to aid and make research in the field easier, for example, the Biologically Inspired Adaptive System Design (BIASD) tool. This tool was designed with the goal of helping designers in early design stages generate more robust and innovative designs. Even though this tool offers a vast database of biological examples, many limitations have been encountered in the tool. The most noticeable is the order in which the biological examples are distributed within the tool. The process used to classify them was very subjective and does not follow a pattern. Another challenge is the way in which the user of the tool reaches the biological examples. By addressing these issues, we provide a more objective way to classify the biological adaptive strategies. To do this, we needed a meta classification in order for the questions to be rationally organized within the tool. Then, approaches such as k-means and Latent Dirichlet Allocation (LDA) techniques from machine learning were employed to minimize the randomness and increase the objectivity of the tool. Out of the two, the LDA model provided a more useful classification. A validation of the LDA model was needed, so we used perplexity, which is used in statistical models to measure the accuracy of a language model and better understand datasets. At the end, a rational classification for the analogues of the BIASD tool was generated.

Share

COinS