Artificial Intelligence-Based Structural Health Monitoring Using CNN and MATLAB : A Case Study on Bridge Safety

Abstract

Structural health monitoring (SHM) is critical to maintaining the safety and lifespan of important infrastructure, such as bridges, that endure constant environmental and mechanical loads. In this paper, an improved SHM system utilizing convolutional neural networks (CNN) developed with MATLAB is demonstrated to identify and classify structural irregularities in bridge structures. A large database of 12,500 labeled images of bridge surfaces, comprising 7,500 images of sound structures and 5,000 images with various defects such as cracks, corrosion, and deformation, was used to train and test the model. The CNN model, comprising 12 convolutional layers, achieved a test dataset accuracy of 98.72%, outperforming conventional vibration-based approaches with an average accuracy of 84.35%. The model also demonstrated a precision of 97.86%, a recall of 98.43%, and an F1-score of 98.14%, ensuring the accurate detection of small structural defects. The system can handle 50 images per second, making it suitable for real-time monitoring. Experimental verification was conducted on a 30-meter steel bridge prototype instrumented with accelerometers and thermal imaging cameras, and tested under various loads ranging from 5 kN to 50 kN. The system, under standard conditions, registered a 1.28% false positive and a 1.57% false negative, indicating its robustness against environmental noise. Moreover, computational expenses were slashed by 22% in MATLAB-based systems when compared to those using Python-based frameworks, ensuring compatibility with established industry standards SHM protocols. This work provides evidence that SHM systems facilitated by AI will significantly enhance the early detection of bridge defects, thereby increasing the efficiency of maintenance and improving public safety.