Georgia Institute of Technology
Product design customization via interactive optimization
Abstract
dc:description.abstractThis thesis presents the development and evaluation of an interactive, human-in-the-loop design framework intended for public-facing, sustainable manufacturing environments. This work was completed as part of the ReCreateIt collaboration with Austin Habitat for Humanity ReStores and Re:3D. The long-term vision of this effort is to enable localized production of household products from recycled plastics using large-format pellet-extrusion 3D printing. In this ecosystem, customers interact with a kiosk-based design tool inside a ReStore, customize or explore product designs, and then select final designs to be fabricated on-site using recycled feedstock. The central challenge addressed in this thesis is that product selection in this context is often driven more by subjective aesthetic preference than quantitative engineering metrics. As a result, fully automated optimization is not sufficient without interaction. At the same time, non-expert users require ease of use in any interaction design framework, and the manufacturing process demands that outputs from the framework to remain feasible and manufacturable. The interactive design framework represents product families using parametric procedural geometry, exposing the customer/user small set of interpretable design variables through slider controls. A geometry-generation pipeline converts parameter values into watertight mesh models suitable for fabrication, including inner and outer surfaces as well as closed top and bottom regions, and this conversion is completed quickly enough for real-time interactions with the customer. Manufacturability is embedded directly in the workflow through an overhang checking method that evaluates surface normals and flags downward-facing regions that exceed allowable limits of the printer being used. Because manufacturability failures emerge from nonlinear parameter interactions, the system includes automated refinement strategies that adjust parameters to recover feasible designs with minimal deviation from the intended design. To help users discover designs aligned with their aesthetic preference, the tool combines direct editing with two preference-guided discovery mechanisms: (1) “See Similar Designs” mode, which uses genetic-algorithm-inspired mutation and crossover on favorited designs to generate related variations, and (2) an Explore mode, based partially on a Bayesian optimization training algorithm to suggest new designs with a high likelihood of improvement. The two modes present candidate designs to balance exploration and exploitation based on user feedback. A user study with 20 participants found strong overall usability, high engagement, and generally positive satisfaction with outcomes. Results indicate that algorithmic assistance can expose users to appealing designs they may not create manually, while users still strongly value authorship, often selecting designs after making their own adjustments. Overall, the thesis demonstrates that combining parametric modeling, embedded manufacturability constraints, and interactive preference-driven exploration is an effective and usable approach for subjective, non-expert consumer design in a real manufacturing context.
Degree
thesis:*- Name thesis:degree_name
- Mechanical Engineering, MS
- Department thesis:degree_department
- Mechanical Engineering
- Grantor
- Georgia Institute of Technology
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Beyer, Andrew
- Advisor dc:contributor.advisor
-
- Seepersad, Carolyn
- Committee members dc:contributor.committeemember
-
- Linsey, Julie
- Wang, Yan
Rights
- Language dc:language.iso
- English
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/81605