Nonlinear structural modeling using multivariate adaptive regression splines

• Main Authors: Zhang, Wengang, Goh, Anthony Teck Chee
• Other Authors: School of Civil and Environmental Engineering
• Format: Article
• Language: English
• Published: 2016
• Subjects: Nonlinearity; Basis function; Multivariate adaptive regression splines; Structural analysis; Neural networks
• Online Access: https://hdl.handle.net/10356/81543; http://hdl.handle.net/10220/39605
• Institution: Nanyang Technological University

Various computational tools are available for modeling highly nonlinear structural engineering problems that lack a precise analytical theory or understanding of the phenomena involved. This paper adopts a fairly simple nonparametric adaptive regression algorithm known as multivariate adaptive regression splines (MARS) to model the nonlinear interactions between variables. The MARS method makes no specific assumptions about the underlying functional relationship between the input variables and the response. Details of MARS methodology and its associated procedures are introduced first, followed by a number of examples including three practical structural engineering problems. These examples indicate that accuracy of the MARS prediction approach. Additionally, MARS is able to assess the relative importance of the designed variables. As MARS explicitly defines the intervals for the input variables, the model enables engineers to have an insight and understanding of where significant changes in the data may occur. An example is also presented to demonstrate how the MARS developed model can be used to carry out structural reliability analysis.