Research on the application of deep learning in bridge health monitoring data analysis and anomaly diagnosis

Abstract

Bridge SHM systems generate large amounts of multi-sensor time-series data for structural condition assessment. However, sensor faults, transmission errors, missing data, noise, drift, and environmental disturbances often reduce the reliability of anomaly diagnosis and may lead to unnecessary maintenance alarms. This study addresses abnormal monitoring data identification in bridge SHM by using a conditional diffusion model. A reconstruction-based anomaly diagnosis framework is developed to characterize condition-consistent normal structural states and improve abnormal state discrimination. In the proposed method, temporal multi-sensor features are extracted to capture time-series dependencies and cross-sensor correlations, while temperature and operational conditions are introduced as conditional information to guide the reverse denoising process. An adaptive anomaly scoring strategy based on signal-level and feature-level reconstruction discrepancies is further developed to reduce false alarms under complex operating conditions. Experimental results show that the proposed method achieves a Precision of 0.971, Recall of 0.980, F1-score of 0.975, and Accuracy of 0.978. The results provide technical support for bridge SHM data quality assessment, operation alarm screening, and anomaly diagnosis, and they also indicate the potential of diffusion models for intelligent infrastructure monitoring.