A cable force anomaly identification method based on bridge deck vibration signals and an unsupervised convolutional autoencoder

Supervised learning-based methods for cable force anomaly detection rely on extensive vibration signals from cables and necessitate prior knowledge of cable damage states, which is often unavailable for in-service bridges. To address this challenge, this paper proposes a cable force anomaly identification method based on bridge deck vibration signals and an unsupervised convolutional autoencoder (CAE). The methodology is structured as follows: First, numerical simulation techniques are employed to generate a large volume of bridge deck vibration signals, thereby alleviating the scarcity of real-world bridge data samples. Second, a CAE is trained using data from normal operating conditions to minimize reliance on labeled datasets, a key limitation of supervised approaches. The trained CAE is then used to reconstruct real-time bridge deck vibration signals. Additionally, three damage assessment indices are proposed to evaluate reconstruction performance, and clustering techniques are applied to identify anomalies in cable forces. Validation experiments were conducted using a scaled model of the First Zhongshan Bridge. Results indicate that the CAE effectively identifies anomalies in bridge deck vibration signals associated with reduced cable forces. Furthermore, clustering and trend analysis, derived from integrating the three assessment indices, clearly delineate progressive changes during cable force loss, demonstrating the method’s feasibility and effectiveness in practical engineering contexts. The proposed approach was also validated on a real cable-stayed bridge, confirming its broad applicability for cable force anomaly identification in bridge structures.