Research on Intelligent Identification of Vortex-Induced Vibration of Stay Cables Based on Multi-Dimensional Feature Extraction and Deep Learning

Abstract

The multimodal vortex-induced vibration (VIV) of stay cables in long-span cable- stayed bridges poses a severe challenge to the real-time identification capabilities of bridge health monitoring systems due to its complex dynamic characteristics and multifactor coupling effects. Existing methods mostly rely on single-dimensional features (such as time-domain or frequency-domain indices) and fail to systematically integrate environmental and structural parameters, thereby limiting identification accuracy. To address this issue, this study proposes an intelligent identification framework based on multidimensional feature extraction and deep learning to enhance the accuracy and real-time performance of VIV identification. Based on the analysis of health monitoring data from a cable-stayed bridge with a main span of 620 meters, a statistical analysis of environmental parameters sensitive to VIV was conducted to determine the wind speed and direction ranges most likely to trigger VIV. On this basis, the time-domain and frequency-domain characteristics of VIV were analyzed, and a multidimensional feature vector was constructed. Subsequently, a BiLSTM-MHA model combining bidirectional long short-term memory networks (BiLSTM) and multi- head attention (MHA) mechanisms was developed to dynamically capture the time- frequency features of vibration signals. Finally, leveraging this model, accurate identification of various stages of VIV was successfully achieved, significantly enhancing the intelligence level of bridge health monitoring and providing an effective technical approach for real-time detection and control of VIV.