Novelty-Driven Binary Particle Swarm Optimisation for Truss Optimisation Problems
This work addresses truss optimization for structural designers by providing a method to find distinct optimal designs, though it is incremental as it builds on bilevel optimization approaches.
The paper tackles the problem of truss topology and sizing optimization by introducing a novelty-driven binary particle swarm optimization method to discover multiple high-quality designs, outperforming state-of-the-art methods.
Topology optimisation of trusses can be formulated as a combinatorial and multi-modal problem in which locating distinct optimal designs allows practitioners to choose the best design based on their preferences. Bilevel optimisation has been successfully applied to truss optimisation to consider topology and sizing in upper and lower levels, respectively. We introduce exact enumeration to rigorously analyse the topology search space and remove randomness for small problems. We also propose novelty-driven binary particle swarm optimisation for bigger problems to discover new designs at the upper level by maximising novelty. For the lower level, we employ a reliable evolutionary optimiser to tackle the layout configuration aspect of the problem. We consider truss optimisation problem instances where designers need to select the size of bars from a discrete set with respect to practice code constraints. Our experimental investigations show that our approach outperforms the current state-of-the-art methods and it obtains multiple high-quality solutions.