Towards Practical Cable Force Monitoring: A Label-Free Identification Strategy Using Deck Vibrations and Deep Feature Learning

Abstract

To address the critical challenge of scarce damage-state labels in cable force anomaly detection, this study proposes an unsupervised strategy integrating deck vibration responses with deep feature learning. The methodology integrates three technical phases: (1) Numerical simulation generates healthy-state deck vibration datasets to train an unsupervised convolutional autoencoder for extracting intrinsic operational features; (2) Three damage-sensitive indices are constructed by quantifying distributional shifts in latent feature space representations of real-time vibrations; (3) Cluster visualization techniques decode cable force degradation pathways. Validation on a Zhongshan No.1 Bridge scaled model demonstrates effective identification of vibration pattern alterations induced by progressive cable force reduction, with multi-index clustering trajectories explicitly revealing damage evolution. Field validation on an in-service cable-stayed bridge confirms technical viability under label-free monitoring contexts.