工程可靠性与随机力学国际联合研究中心

2025 年第 4 期（总第 96 期）学术报告

工程力学研究中心第 50 期学术报告

文远讲坛 319 期

报告主题
TOPIC

结构可靠性分析的自适应支持向量回归模型

Adaptive Support Vector Regression Models for Reliability Analyses of Structures

报告人
SPEAKER

Prof. Subrata Chakraborty

Professor (HAG) and former Head, Department of Civil Engineering,
Indian Institute of Engineering Science and Technology, Shibpur,
IndiaFellow of the Indian National Academy of Engineering

报告时间
TIME

2025年6月27日（周五）下午15:00-16:30

报告地点
VENUE

同济大学土木大楼 A305

主持人
CHAIR

彭勇波教授、陈建兵教授联系人：律梦泽博士

报告内容
Abstract

Support vector regression (SVR) based metamodel is a powerful means to alleviate the computational challenge of Monte Carlo simulation (MCS) based reliability analysis of structures. However, the sample size requirement is an important issue for accuracy in the estimated reliability. Two recently proposed adaptive SVR algorithms for estimating reliability efficiently will be presented. 

A two-stage iterative enrichment of the training dataset to construct an SVR model is proposed to estimate the reliability of a structure efficiently with reasonable accuracy. The two-stage algorithm is hinged on the prediction accuracy of a metamodel near the failure surface region. An initial design of experiment (DOE) is built with a space-filling design over the entire physical domain of the random variables. In the next stage, a subset of MCS samples is selected based on the prediction at MCS points using the previous SVR model. These are now used to enrich existing design by adding more data points sequentially so that the new points are closer to the limit state and as far as possible from the existing points.

The two-stage sequential updating of the SVR model presented above faces the difficulty of the scarcity of samples in the reduced space to estimate very small probabilities of failures. A three-stage adaptive SVR model is further proposed to estimate low probabilities successfully. Specifically, an importance sampling-based algorithm is proposed to ensure sufficient simulation points near the approximated failure plane. An initial DOE is first constructed by space-filling design over the entire domain. The optimum choices of the hyperparameters of the SVR model are determined by minimizing generalized root mean square error (GRMSE). A subset of MCS samples with an approximated limit state function value less than the noted GRMSE value is selected next. Subsequently, the data points are added sequentially from the subset based on the maximin criterion. The SVR model is iteratively updated until convergence. The efficacies of the algorithms are elucidated through three numerical examples.

报告人简介
Speaker Bio

Dr. Subrata Chakraborty is a Professor (HAG) and former Head of Civil Engineering at the Indian Institute of Engineering Science and Technology, Shibpur. He is a fellow of the Indian National Academy of Engineering, West Bengal Academy of Science and Technology, Council of Vibration Specialists and Institution of Engineers (India). Prof. Chakraborty completed his Bachelor's in Civil Engineering from Bengal Engineering College, Shibpur, M. Tech. and PhD from the Indian Institute of Technology Kharagpur. He was a postdoctoral researcher at the University of Cambridge, the University of Arizona, and the Technical University of Aachen, Germany.

Prof. Chakraborty’s research interests are uncertainty quantification in structural Engineering, structural health monitoring, structures under extreme events, vibration control, fibre-reinforced composites, etc.  Hey has completed several funded research projects and is also active in international collaborative research. He is also active in important industrial consultancy in safety assessments of real-life structures. Prof. Chakraborty has supervised fifteen doctoral and eighty master's theses and published over 135 journal articles, books, and book chapters. While his inspiring teaching, coupled with his innate urge for intensive research, has already established him as a distinguished academician at the national level, several awards and laurels have come his way. The Humboldt Fellowship for Experienced Researchers, the V. H. Joshi Award for significant contributions in structural dynamics, the INAE Young Engineer Award, the BOYSCAST Fellowship of the Government of India, and the Young Faculty Research Award by the US alumni association deserve special mention.