A genetic algorithm based topology optimisation approach for exploiting rapid manufacturing's design freedom

Current product structures are designed to meet predefined specifications and are thereforer arely optimal. Instead, they are frequently over-engineered to ensure fitness for purpose, which results in excess weight and uneven stress distributions throughout their structures. Designs are then often compromised further in terms of optimality by the inherent process limitations of conventional manufacturing. Rapid manufacturing (RM), due to its vastly increased design freedom, can overcome these restrictions and become the enabling technology for fabricating uncompromised, optimal products. This thesis describes the design, creation, testing and evaluation of a new design optimisation system capable of exploiting the high design freedom afforded by RM technologies. Inspired by the design rules that Nature follows, the system combines the stochastic search behaviour of Genetic Algorithm (GA) with finite element analysis in order to evolve optimal topological structures via a survival of the fittest process. The novelty of this approach is that 3D unit cell structures varying in volume fraction are used to simulate different densities of a single material, which are then efficiently distributed throughout the problem domain by the GA to yield an improved stress distribution. Furthermore, the system can consider different unit cells mixed together within the same problem, thereby substantially expanding the topology optimisation research field. Following a series of experiments of increasing complexity, the performance, stability and computational demands of the system are discussed. Experimental results indicate the system works, however, the presence of unit cells was found to cause localised stress concentrations to occur, which tend to inadvertently steer the overall optimisation process. Suggestions to address this issue were made. In addition, the mixing of different unit cells together was found to improve trade-offs between system objectives but did not always improve stress distribution.