EXPLAINABLE DEEP LEARNING WITH MULTIMODAL SENSOR FUSION FOR BRIDGE STRUCTURAL HEALTH MONITORING

Abstract

Structural health monitoring (SHM) of bridges is essential for ensuring public safety, reducing maintenance costs, and extending the service life of critical transportation infrastructure. Traditional bridge inspection methods depend heavily on periodic visual assessment, which can be time-consuming, subjective, costly, and unable to detect early-stage structural deterioration. This paper presents an AI-based SHM framework that integrates multimodal sensor fusion and explainable deep learning for reliable bridge damage detection and condition assessment. The proposed approach combines heterogeneous sensor data, including vibration signals, strain measurements, displacement readings, acoustic emission responses, and environmental variables, to generate a comprehensive representation of bridge behavior under operational conditions. Deep learning models are used to identify hidden damage patterns, classify structural states, and predict abnormal responses with improved accuracy compared with single-sensor monitoring methods. To enhance model transparency, explainable artificial intelligence techniques are incorporated to identify the most influential sensor channels, structural features, and decision regions contributing to damage predictions. The results show that multimodal fusion improves classification performance, reduces false alarms, and strengthens robustness under noise and environmental variability. Explainability outputs further support engineering interpretation by linking model decisions to physically meaningful indicators such as strain variation, modal frequency shift, and vibration intensity. Overall, the study demonstrates that AI-driven multimodal SHM can provide an effective, interpretable, and scalable solution for bridge monitoring, supporting early warning, preventive maintenance, and data-informed infrastructure management.