Support Vector Machine for Regression of Ultimate Strength of Trusses: A Comparative Study

Abstract

Thanks to the rapid development of computer science, direct analyses have been increasingly used in the design of structures in lieu of member-based design methods using the effective length factor. In a direct analysis, the ultimate strength of a whole structure can be sufficiently estimated, so that the need for member capacity checks is eliminated. However, in complicated structural design problems where many structural analyses are required, the use of direct analyses requires an excessive computation cost. In such cases, Machine Learning (ML) algorithms are used to build metamodels that can predict the structural responses without performing costly structural analysis. In this paper, the support vector machine (SVM) algorithm is employed for the first time to develop a metamodel for predicting the ultimate strength of trusses using direct analysis. Several kernel functions for the SVM model, including linear, sigmoid, polynomial, radial basis function (RBF), are considered. A planar 39-bar nonlinear inelastic steel truss is taken to study the performance of the kernel functions. The results confirm the applicability of the SVM-based metamodel for predicting the ultimate strength of trusses. In particular, the RBF appears to be the best kernel among others. This investigation also provides a deeper understanding of the effect of the parameters on the efficiency of the kernel functions.