Representing Knowledge for Data-Driven Design
Name
akay-haluk-phd-meche-2022-thesis.pdf
Description
Thesis PDF
Size
19.89 MB
Format
Adobe PDF
Checksum (MD5)
4bea41bc8e87be7964846deb512c4d12
Author(s)
Akay, Haluk John
Advisor(s)
Kim, Sang-Gook
Date Issued
May 2022
Publisher
Massachusetts Institute of Technology
Abstract
Data-driven models have proven to be a transformative alternative to rule-based methods of the past. A data-driven transformation of design is necessary to guide engineers through complexity to develop next-generation products and production systems. Data is abundant from digitally documented early-stage design through final production processes, but this data is often unstructured, informal, and can be qualitative or textual in nature. For data-driven design, data must be computationally interpretable for past documented knowledge to guide future engineering decision-making. This thesis research leverages deep neural network-based language modeling to represent design data; specifically, textually described knowledge. Quantitative representation models make possible a wide range of applied AI methods for performing tasks such as evaluating functional interdependencies and extracting functional information from past design documentation. By learning from past engineering failures and achievements, Big Data and Artificial Intelligence can be used to assist human designers’ decision-making for meeting the needs of society and the environment through data-driven design.
MIT Department
Massachusetts Institute of Technology. Department of Mechanical Engineering
Terms of Use
In Copyright - Educational Use Permitted
Copyright MIT
Persistent DSpace Link
