Dynamic reliability analysis using the extended support vector regression (X-SVR)

For engineering applications, the dynamic system responses can be significantly affected by uncertainties in the system parameters including material and geometric properties as well as by uncertainties in the excitations. The reliability of dynamic systems is widely evaluated based on the first-passage theory. To improve the computational efficiency, surrogate models are widely used to approximate the relationship between the system inputs and outputs. In this paper, a new machine learning based metamodel, namely the extended support vector regression (X-SVR), is proposed for the reliability analysis of dynamic systems via utilizing the first-passage theory. Furthermore, the capability of X-SVR is enhanced by a new kernel function developed from the vectorized Gegenbauer polynomial, especially for solving complex engineering problems. Through the proposed approach, the relationship between the extremum of the dynamic responses and the input uncertain parameters is approximated by training the X-SVR model such that the probability of failure can be efficiently predicted without using other computational tools for numerical analysis, such as the finite element analysis (FEM). The feasibility and performance of the proposed surrogate model in dynamic reliability analysis is investigated by comparing it with the conventional c-insensitive support vector regression (epsilon-SVR) with Gaussian kernel and Monte Carlo simulation (MSC). Four numerical examples are adopted to evidently demonstrate the practicability and efficiency of the proposed X-SVR method. (C) 2019 Elsevier Ltd. All rights reserved.