Damage detection of a cable-stayed bridge specimen based on unsupervised subdomain adaptation transfer learning

Authored by

Naiwei Lu, Jian Cui, Weiming Zeng, Yang Liu, Michael Beer

Abstract

Structural damage detection (SDD) based on structural health monitoring (SHM) data is an essential task for maintaining the operational continuity and structural integrity for civil infrastructure. However, a major challenge arises from the insufficient historical labeled dataset, coupled with the disparity between the limited damage states in SHM data and the infinite potential states. This study presents a digital twin-driven feature-transferable method that predicts structural damage accurately using limited and unlabeled SHM data. In addition, an adaptive unsupervised transfer learning method is introduced to extract damage features in cross-domain dataset. The proposed approach generates damage-sensitive and domain-invariant features for unlabeled signals, in contrast to traditional methods that aim to reduce domain gaps. Experimental results from a scaled cable-stayed bridge specimen demonstrate the superiority of the proposed approach in enhancing learning characteristics and accurate damage prediction from limited unlabeled datasets.

Details

Organisation(s): Institute for Risk and Reliability
External Organisation(s): Changsha University of Science and Technology
University of Liverpool
Tongji University
Type: Article
Journal: Engineering structures
Volume: 346
ISSN: 0141-0296
Publication date: 01.01.2026
Publication status: Published
Peer reviewed: Yes
ASJC Scopus subject areas: Civil and Structural Engineering
Electronic version(s): https://doi.org/10.1016/j.engstruct.2025.121660 (Access: Closed )