Fault Diagnosis Method of Electromechanical Equipment in Expressway Toll Station based on Deep Learning

The reliable operation of electromechanical systems at expressway toll stations is essential to ensuring traffic safety. Nevertheless, the considerable diversity among devices in terms of structural complexity, operating principles, and functional hierarchy results in a wide range of diagnostic target scales. Conventional fault detection approaches struggle with such multi-scale characteristics, yielding suboptimal accuracy and generating excessive false alarms. These spurious alerts not only burden maintenance crews with unnecessary signal verification but also impair the responsiveness and reliability of on-site fault assessment. To address these challenges, this study proposes an intelligent diagnostic framework that incorporates a multi-scale feature interaction module. By enabling cross-level feature exchange and multi-resolution fusion, the model aligns fine-grained spatial details with high-level semantic cues, thereby enriching representational capacity and strengthening its adaptability to equipment of varying scales. Comparative experiments on real-world toll station data demonstrate that the proposed method outperforms both YOLOv5 and ResNet34 in fault identification accuracy. Moreover, it offers an interpretable end-to-end solution that facilitates real-time alarming and predictive maintenance, contributing to more robust toll station operations.