A State Monitoring Method for Shear Hinges in Floating Slab Tracks Based on FMTF and PCNN-RWKV Fusion Network

Floating slab track (FST) is a core vibration and noise reduction technology in urban rail transit. Shear hinges, key connectors of floating slabs, greatly affect rail transit operational stability and safety. Limited by complex operating conditions and weak feature extraction of traditional signal processing, existing methods yield inadequate feature representation, low diagnosis accuracy, and poor anti-noise performance. This study develops a novel shear hinge state recognition network combining Frequency Domain Markov Transition Field (FMTF), Parallel Convolutional Neural Network (PCNN), and Receptance Weighted Key Value (RWKV). Based on shear hinge working principles, vibration differences at floating slab ends indirectly reflect hinge health conditions. One-dimensional vibration spectrum sequences are transformed into 2D FMTF images to capture spectral dynamic transition features. PCNN extracts multi-scale image features, while RWKV fuses cross-channel dependencies for precise state identification. The model achieves accuracies of 98.19%, 95.39%, and 99.37% for normal, abnormal-response, and fracture states, respectively, with an overall accuracy of 97.62%. Featuring a compact architecture, it extracts multi-scale features efficiently with faster inference, offering a noise-resistant approach for FST shear hinge monitoring.