Investigation of genetic algorithm design representation for multi-objective truss optimization

by Pathi, Soumya Sundar

Abstract (Summary)
The objective of this research is to develop a flexible design grammar and

genetic algorithm representation to be used in a multi-objective optimization method to

design efficient steel roof trusses given space dimensions and loading requirements by

the user. The goal of implementing the method as a multi-objective problem is to obtain

a set of near-optimal trusses for the defined unstructured problem domain, not just a

single near-optimal design. The method developed was required to support the

exploration of a broad range of conceptual designs before making design decisions.

Therefore, a method was developed that could define numerous design variables, support

techniques to locate global or near-global optimal designs, and improve the efficiency of

the computational procedures implemented. This research effort was motivated by the

need to consider structural designs that may be beyond the established conventions of

designers in the search for cost-efficient, structurally-sound designs.

An effective design grammar that is capable of generating stable trusses is

defined in this research. The design grammar supports the optimization of member size,

in addition to truss geometry and topology. Multi-objective genetic algorithms were used

to evolve sets of Pareto-optimal trusses that had varying topology, geometry, and

member sizes. The Pareto-optimal curves provided design engineers with a range of

near-optimal design alternatives that showed the tradeoffs that occur in meeting the

stated objectives. Designers can select their final design from this set based on their own

individual weighting of the design objectives. Trials are performed using a multiobjective

genetic algorithm that works with the design grammar to evolve trusses for different span lengths. In addition to evaluate the performance of the developed

optimization method further, trials were performed on a benchmark truss problem

domain and the results obtained were compared with results obtained by other


The results of the performance evaluation trials for the proposed method, in

which the sizing, shape and topology were simultaneously performed, indicated that the

method was effective in evolving a variety of truss topologies compared to previous

published results, which evolved from a ground structure. The diverse topologies,

however, were obtained over several trials instead of being found in a Pareto-optimal set

found by a single trial. In addition, the proposed method was not able to locally

optimize the member section sizes. Additional trials were performed to determine the

benefit of applying local optimization to the member section sizes for a given truss

topology or geometry provided by the method. The results indicate that significant

weight reduction could be achieved by performing local optimization to the truss designs

obtained by the proposed multi-objective optimization method.

Bibliographical Information:

Advisor:Raich, Anne M.; Anshelevich, Michael; Bracci, Joseph

School:Texas A&M University

School Location:USA - Texas

Source Type:Master's Thesis

Keywords:optimization roof trusses


Date of Publication:08/01/2006

© 2009 All Rights Reserved.