University of Birmingham
Estimating functional performance for use in the aesthetic design process
Abstract
dc:description.abstractIn engineering fields such as automobile design, optimisation of functional performance properties often conflicts with aesthetic optimisation. Functional performance feedback into the aesthetic design software may therefore improve the convergence of the design process. Unfortunately, many functional performance scores such as aerodynamic drag require intensive computational effort. We consider the use of machine learning approaches to instead provide estimates of these functional performance scores. We study the problems encountered when developing such an estimation function. The use of a historically accumulated data set of STereoLithography-format designs and their performance scores is suggested. We first look at preparing such a data set as training data for a machine learning task. Our first major novel contribution combats this problem in a manner similar to voxelisation. We next look at generating the regression function, seeking to achieve good generalisation across a large space of possible designs and for a problem where dimensionality reduction is challenging. Our second major novel contribution deals with this problem using an ensemble regression framework incorporating multiple data representations. Finally, we look at strategies of combining these two novel systems into a complete system. Upon evaluation, we conclude that our original aims have been met by this complete system.
Degree
thesis:*- Name dc:type.qualificationname
- d_ph
- Level dc:type.qualificationlevel
- d_ph
- Grantor dc:publisher.institution
- University of Birmingham
- Year dc:date.issued
- 2015
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Kent, Matthew Paul
Subjects
dc:subject × 2Rights
- Language dc:language.dcterms_iso639_2
- eng