Use this URL to cite or link to this record in EThOS:
Title: Generative part design for additive manufacturing
Author: Goguelin, Steven
ISNI:       0000 0004 7961 408X
Awarding Body: University of Bath
Current Institution: University of Bath
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Access from Institution:
Additive manufacturing has transformed from a technology primarily focused on the creation of small-scale prototypes and models to a process for the manufacture of end-use components. Additive manufacturing has a number of inherent advantages over traditional manufacturing processes. These include the ability to fabricate complex geometries at no extra production cost, creating mass customised parts without additional tooling and the ability to consolidate assemblies into single parts. However, additive manufacturing is not without limitations. There are a number of geometric constraints that limit design freedom when designing parts, for example the maximum unsupported angle that material can be printed. There are also limitations in current CAD software, which prevent designers from maximising the quality of additively manufactured parts. Design for additive manufacturing seeks to improve the design methods or tools to help improve the functional performance, reliability, manufacturability or cost of parts produced using additive manufacturing technologies. Generative design is an emerging form of computational design in which the user provides goals and constraints to a system and generative synthesis algorithms produce a series of optimised solutions based on the input criteria. There are many limitations with current generative design systems preventing the mass adoption of the technology. These include, the lack of integration between topology optimisation synthesis algorithms and the part build orientation and, additionally, the ability to design for goals, such as part cost or build time. To overcome these challenges, this research applies two generative design methods to design an additively-manufactured cantilever beam. The optimised beams are created by integrating a ground-structure topology optimisation with manufacturing constraints and build orientation angle information. Design performance metrics of varying degrees of abstraction are then derived from the mesh file data. These represent two common additive manufacturing business scenarios; maximum part performance and high production quantity. A data-driven generative design approach is then used to locate the top performing solutions within a solution space. This space is searched using a parametric grid-search that alters the build orientation and the overhang angle constraint. Component performance is related to the abstracted design objectives using a TOPSIS multi-criteria decision analysis. There is an ongoing challenge associated with running many model evaluations on large mesh files. This can make generative design prohibitive in terms of computational resource. To overcome this challenge, a goal-driven generative design method is developed to solve the inverse problem of finding the optimal input parameters to the cantilever beam problem by using a Bayesian optimisation surrogate model. The data-driven and goal-driven generative design approaches are then compared for their efficiency and ability to locate optimal design solutions. The contributions to knowledge, borne from this research are a data-driven generative design method demonstrated to be suitable for locating high-performing solutions in complex multidimensional solution spaces providing the number of design space dimensions is small. The novel use of Bayesian optimisation is shown to be to be 17 times more efficient than a conventional grid search for locating the top performing build orientations of two additively manufactured test parts. Finally, goal-driven generative design methods are demonstrated to locate the optimised build orientation and manufacturing constraints for the cantilever beam within 20 optimisation iterations. These outcomes demonstrate potential for future generative design CAD systems.
Supervisor: Flynn, Joseph ; Dhokia, Vimal ; Newman, Stephen Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available