The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #9710

Title:Evolutionary Shape Optimisation Using a Voxel Based Representation
Date:Sep 1997
Abstract:Shape optimisation within constraint limits is a hard problem from the field of Mechanical Engineering. The objective is to design a shape which best satisfies some predetermined goal whilst at the same time observing some constraints. Much work has been done on various problems within this domain [Bentley 96][Chapman et al. 94][Smith 95a][Watabe and Okino 93], however the majority to-date has focused on parametric and other highly restricted representations of the shapes being optimised. This thesis describes experiments performed using a voxel based representation and an evolutionary optimisation algorithm. As any attempt to perform shape optimisation using an unrestricted objective representation must use a highly effective evaluation function in order to be able to deal with the varieties of design that an evolutionary optimisation algorithm may produce, the work detailed here was intended to first design and test suitable operators to deal with the extremely large chromosomes required by a voxel representation, and by then extending the work to permit evaluation from a commercial Finite Element analysis package, it was expected that the program would be able to perform real-world shape optimisation tasks with very few restrictions on the shapes being tested. The results obtained indicate that it is possible to circumvent the difficulties presented by a voxel representation [Watabe and Okino 93], and that with suitable operators, useful and unexpected results can be gained from a real-world optimisation task.

[Search These Pages] [DAI Home Page] [Comment]