In structural optimization, for both Deterministic Design Optimization (DDO) or Reliability-Based De- sign Optimization (RBDO) approaches, the nature of the objective function remains the same, as minimizing the weight, for example. However, RBDO formula

Resumo

In structural optimization, for both Deterministic Design Optimization (DDO) or Reliability-Based Design Optimization (RBDO) approaches, the nature of the objective function remains the same, as minimizing the weight, for example. However, RBDO formulation differs from DDO by the possibility of finding the optimal solution considering failure probabilities limits or target reliability indices as design constraints. Classical methods found in the literature can do reliability assessment. Nonetheless, due to convergence problems and the considerable computational effort required, it is interesting to employ other techniques like surrogate models in order to reduce the processing time. Thus, this work intents to compare a traditional double-loop RBDO analysis and a
machine learning based RBDO model, by using Artificial Neural Networks (ANNs), in a single floor steel frame example. First order structural analysis is considered, and Genetic Algorithms perform the optimization. The First Order Reliability Method (FORM) calculates the reliability index. The numerical example shows how the ANN performance and accuracy are quite dependent on its architecture and on the available number of training samples.