An adaptive augmented radial basis function–high-dimensional model representation method for structural engineering optimization. Issue 15 (November 2020)

• Record Type: Journal Article
• Authors: Wang, Qian; Kim, Yongwook; Nafash, Joseph; Catala, Javier
• Is Part Of: Advances in structural engineering. Volume 23:Issue 15(2020)
• Page Start: 3278
• Page End: 3294
• Publication Date: 2020-11
• Subjects: adaptive metamodels -- augmented radial basis function -- building structures -- high-dimensional model representation -- structural engineering optimization; Structural engineering -- Periodicals; Construction, Technique de la; Structural engineering; Periodicals; 624.1
• DOI: 10.1177/1369433220931217
• Languages: English
• ISSNs: 1369-4332

Abstract

A new engineering optimization approach using an adaptive metamodeling method is developed and studied. The adaptive metamodels are based on a high-dimensional model representation framework, and the high-dimensional model representation component functions are created using radial basis functions or augmented radial basis functions. The proposed optimization approach starts with an explicit first-order augmented radial basis function–high-dimensional model representation metamodel, before a numerical optimization algorithm is applied. In each subsequent iteration, an additional sample point is found, and a high-order high-dimensional model representation component function is created and added to the first-order augmented radial basis function–high-dimensional model representation metamodel. The accuracy of the augmented radial basis function–high-dimensional model representation metamodel is improved in an adaptive manner, especially in the neighborhood of the optimal design point. Several numerical examples are solved to demonstrate the method, including a practical three-dimensional reinforced concrete high-rise building structure. The proposed approach works well, and the convergence of the optimal solutions for each of the examples is obtained within a few adaptive iterations.