庆祝工程可靠性与随机力学国际联合研究中心

成立 10 周年杰出讲座

(Distinguished Lectures for the 10thAnniversary 

Celebration of JCERSM) 第 4 期

工程可靠性与随机力学国际联合研究中心

2026 年第 7 期（总第 123 期）学术报告

工程力学研究中心第 78 期学术报告

文远讲坛 353 期

报告主题
TOPIC

面向全贝叶斯模型更新
Towards Fully Bayesian Model Updating

报告人
SPEAKER

Dr. Marcos Valdebenito
Senior Scientist,TU Dortmund University, Germany

报告时间
TIME

2026 年 6 月 29 日（周一）上午 10:00-11:00

报告地点
VENUE

同济大学土木大楼 A305

主持人
CHAIR

陈建兵 教授
联系人: 牛立志

报告内容
Abstract

Bayesian model updating is widely used in engineering to identify unknown parameters of computational models based on noisy experimental measurements. From a numerical point of view, however, the problem is demanding. Solving the associated inverse problem requires repeated evaluations of the computational model (also known as the forward model) in order to find parameter values that produce responses consistent with the observed data. This leads to a large number of model evaluations, which quickly becomes prohibitive when high-fidelity models are involved.

To mitigate this cost, surrogate models, in particular Gaussian process regression (GPR), are often employed. GPR provides both predictions of the system response as well as a probabilistic description of the approximation error. In many existing approaches, however, this additional source of uncertainty is not treated consistently within the Bayesian framework. In this contribution, a formulation for Bayesian updating is proposed in which the predictive uncertainty of the Gaussian process is explicitly propagated into both the likelihood function and the model evidence. The approach also allows for the identification of measurement noise characteristics.

An active learning strategy is further introduced to refine the surrogate model adaptively, with the aim of controlling the uncertainty in the estimation of the model evidence. The resulting formulation provides a consistent probabilistic framework in which surrogate uncertainty, measurement noise, and adaptive learning are treated jointly within Bayesian inference. The approach is illustrated by means of examples from structural dynamics.

报告人简介
Speaker Bio

Dr. Marcos Valdebenito is Senior Scientist at the Chair for Reliability Engineering at TU Dortmund University, Germany. He obtained his doctoral degree in Civil Engineering from University of Innsbruck, Austria, after completing his engineering and master’s studies at Santa Maria University, Chile. His research focuses on uncertainty quantification, structural reliability, stochastic dynamics, Bayesian updating, surrogate modelling, and reliability-based design optimization in computational mechanics. He has authored more than 100 journal publications. Dr. Valdebenito received the K.J. Bathe Award in 2016 and a research fellowship from the Alexander von Humboldt Foundation. He currently serves on the editorial boards of the international journals Computers & Structures, Structural Safety, and Machine Learning for Computational Science and Engineering.