CRE Researchers Visit Tongji University and Present at ISRERM 2026

Research Visit to Tongji University

Prof. Matthias Faes, Miriam Dodt, Mauricio Misraji and Weiming Zheng from the Chair for Reliability Engineering (CRE) visited Tongji University in Shanghai from 22nd to 26th June 2026. Hosted by Prof. Jianbing Chen and his research group, the visit brought together researchers from both institutions for presentations, discussions and scientific exchange in the fields of reliability engineering and stochastic mechanics.

During the visit, they presented their current research alongside three researchers from Tongji University. Their presentations were entitled Hierarchical Bayesian Updating with Explicit Noise Consideration (Miriam Dodt), Importance Line Sampling for First Excursion and Sensitivity Analysis of Linear Dynamic Systems (Mauricio Misraji), and Reliability Analysis of Sequential Systems by Bayesian Active Learning (Weiming Zheng).

The programme also included visits to laboratory facilities at Tongji University’s Siping Road and Jiading campuses. These activities provided insights into the university’s experimental research infrastructure and created opportunities to discuss current research and potential areas for future collaboration.

Participation in ISRERM 2026

Following the visit to Shanghai, Miriam Dodt, Mauricio Misraji and Weiming Zheng participated in the 10th International Symposium on Reliability Engineering and Risk Management (ISRERM 2026) from 28th June to 1st July 2026 at Hokkaido University Conference Hall, Sapporo, Japan. The symposium brought together international researchers to exchange recent advances in reliability engineering, risk management and uncertainty-informed engineering.

The CRE researchers contributed the following presentations:

Miriam Dodt Research:

Hierarchical Approximate Bayesian Updating considering measurement uncertainties

Bayesian updating provides a well-established framework for parameter inference, but evaluating the likelihood function is often infeasible. Approximate Bayesian computation (ABC) addresses this challenge by updating based on simulated quantities of interest and a discrepancy measure. Hierarchical Bayesian updating further allows to update not only deterministic quantities of interest but also stochastic ones by inferring their distribution parameters. In this case especially, measurement noise needs to be considered appropriately, else the posterior distribution will be affected negatively.

Two ABC approaches that explicitly incorporate measurement noise are discussed and compared on two analytical case studies to an implicit ABC scheme. The study examines ease of implementation, accuracy and robustness and highlights the strengths and challenges of the different approaches as well as guidance for applying noisy ABC in engineering contexts.

Mauricio Misraji’s Research

Importance Line Sampling for Dynamic Reliability of Linear Stochastic Systems

Earthquake and wind loads can be represented as stochastic processes, allowing their uncertainty to be propagated to structural responses. The probability that a response exceeds a threshold within a fixed period is known as the first-excursion probability. Its estimation remains challenging because discretized stochastic loads lead to high-dimensional problems, while Monte Carlo simulation can require many dynamic response evaluations, particularly for small failure probabilities.

This contribution explores Importance Line Sampling (ILS) for linear structural systems subjected to Gaussian stochastic loading. The method combines an importance sampling density, which favours regions contributing most to failure, with Line Sampling and uses multiple exploration lines to exploit system linearity. It is demonstrated on a large-scale finite element model with failure probabilities of 10⁻³ or smaller. The results show that ILS estimates small failure probabilities with high precision and low computational effort, highlighting its potential for dynamic reliability analysis.

Weiming Zheng’s Research

Reliability Analysis of Sequential Series Systems by Bayesian Active Learning with Multiple Gaussian Process Models

Many engineering systems have a sequential series structure in which the output of one subsystem becomes an input to the next, while the entire system fails if any limit state is violated. Reliability analysis of such systems is challenging because uncertainty must be tracked both within each subsystem and throughout the complete dependency chain.

This contribution presents a Hierarchical Active Learning Probabilistic Integration (HALPI) framework for efficient reliability analysis of first-order sequential series systems. Each subsystem is represented by an individual Gaussian Process surrogate. A new acquisition function combines local classification uncertainty with uncertainty propagated from upstream subsystems, allowing the method to select samples that significantly influence the failure-probability estimate. In a finite element case study, the framework accurately estimated stage-wise failure probabilities while requiring substantially fewer model evaluations than the compared all-in-one surrogate approaches. The results demonstrate its accuracy, robustness and computational efficiency for complex multilevel systems.

Further information: ISRERM 2026 – Official Website