The metaheuristics have become popular as modern methods that provide near-optimum solutions to the optimization problems because of their simplicity and gradient-free movement operators; it is well known that most of the population-based metaheuristics are successful in structural optimization. Unfortunately, metaheuristics require evaluation of a significant number (mostly ten thousands) of design candidates to locate a near-optimal solution. During a structural optimization process, each candidate is sent to a structural analysis software to determine whether it satisfies the considered structural constraints. As a result, structural optimization with metaheuristics may be time consuming depending on the size and complexity of the problem even with today's advanced computers. This study presents an optimization procedure with an artificial neural network-based surrogate assistance to reduce the computational cost of the structural optimization with metaheuristics. The presented surrogate model is able to capture the structural system behavior and able to provide estimations without spending too much computational effort even for extremely difficult problems. The numerical experiments are conducted on a popular truss-sizing benchmark problem using a parameter-free metaheuristic algorithm. It is shown by the consecutive optimization attempts that the proposed surrogate assistance considerably improves the results. On the other hand, the numerical experiments revealed a few issues on surrogate-assisted structural optimization that should be discussed.
Anahtar Kelimeler: Structural Optimization, Artificial Neural Networks, Surrogate, Metaheuristic, Truss
|
