This paper presents an adaptive radial basis functions (RBFs) metamodeling method for design optimization of structures. Various numerical techniques have been developed and adopted in structural and multidisciplinary optimization. To evaluate responses of a structural or mechanical engineering system, finite element (FE) analyses are routinely used. An FE code shall be integrated with an optimization algorithm in a nested analysis and design of structures. Therefore, software input/output programming is required. A metamodeling method, on the contrary, expresses structural responses using an approximate function, so that the FE software is not directly coupled in the numerical optimization loop. Any optimization algorithm can be applied to find the optimal design, based on the explicit response functions. In this study, numerical examples were created and FE analyses were first performed at sample points. Subsequently, metamodels were constructed and a gradient-based optimization algorithm was applied. At the optimal point of one adaptive iteration, accuracy of the RBF metamodel was checked, and additional sample points were added to the sample pool to improve the model accuracy. The adaptive iterations continued, until the convergence of the objective function was achieved. The proposed optimization method worked well for a numerical example, and the optimal result was found within a few adaptive iterations.
Full Text (PDF)