Back to search

Georgia Institute of Technology

Generative Design Using Deep Learning Methods for Functionality and Manufacturability

Abstract

dc:description.abstract

Digital manufacturing aims to optimize the entire design-to-manufacturing workflow, seamlessly integrating cutting-edge technologies to boost efficiency, precision, and agility. To enable digital manufacturing, one must achieve the seamless implementation and integration of multiple critical functions, and two of them, i.e., manufacturing process selection and generative design are the focus of this research. To aid the initial design process, automated manufacturing process selection capabilities should be developed that make recommendations early before all part shape details including quality specifications are added. The manufacturing process selection procedure will always consider shape information and may or may not include material information and other manufacturing information such as part quality requirements. Selection of manufacturing processes can be achieved through either direct classification of processes based on part shape and other requirements, or by searching for similar designs and extracting manufacturing insights from previously manufactured similar designs. We have trained neural networks on different representations of 3D designs, i.e., triangle mesh, point cloud, multi-view images and voxel, to extract shape features. These extracted shape features could be used directly for manufacturing process classification, or they could be combined with material information to select similar designs and determine manufacturing processes. Our results show that these trained models can accurately predict manufacturing processes based solely on shape information, or they can select suitable manufacturing processes through similarity search using shape and material information. During the process of adding part shape details, two design procedures: design for functionality (DFF) and design for manufacturing (DFM) are relevant. During DFF, the objective is to improve the functionality of the designed part. In DFM, functional design is utilized as input to improve its manufacturability. In the end, DFF and DFM can form a loop to continuously improve design functionality and manufacturability till the solution converges. Conditional generative adversarial networks (cGANs) were trained in these two modules, DFF and DFM, to improve the functionality and manufacturability of designs. DFF directly generated functional designs from random noise and the trained cGAN could be extended to different load and boundary conditions, while DFM modified existing unmanufacturable designs to generate their corresponding counterparts. The two modules were combined in an iterative loop, i.e., the output of DFF would be the input of DFM and the output of DFM would utilized to train DFF, to directly generate designs with both good functionality and manufacturability. This research is proposed to help automate the design and manufacturing processes in digital manufacturing so that an automatic generative design method for manufacturing can be realized.

Degree

thesis:*
Level thesis:degree_level
Doctoral
Department dc:contributor.department
Mechanical Engineering
Grantor dc:publisher
Georgia Institute of Technology
Year dc:date.issued
2023

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Wang, Zhichao
Advisors dc:contributor.advisor
  • Rosen, David W.
  • Melkote, Shreyes N.
Committee members dc:contributor.committeemember
  • Wang, Yang
  • Choi, Seung-Kyum
  • Gombolay, Matthew

Subjects

dc:subject × 5

Rights

Language dc:language.iso
en_US

Identifiers

dc:identifier.*
Handle dc:identifier.uri
https://hdl.handle.net/1853/72983

Chain of custody

source
Harvested from
Georgia Tech
Base URL
repository.gatech.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Wang, Zhichao. Generative Design Using Deep Learning Methods for Functionality and Manufacturability. Doctoral thesis, Georgia Institute of Technology, 2023. https://hdl.handle.net/1853/72983