A Framework for Structural Monitoring: Deep Learning and Digital Twin for Anomaly Detection

Structural monitoring plays an important role in ensuring the safety and longevity of infrastructure; nevertheless, conventional methods frequently lack the capacity for continuous and predictive analysis. This study presents an advanced Structural Health Monitoring (SHM) system that integrates IoT, Digital Twin (DT), and Deep Learning (DL) to automate the recognition of issues with structure and enhance preventive maintenance. The methodology entails obtaining vibrational data from MEMS accelerometer sensors, analyzing it with a Convolutional Autoencoder (CAE), and visualizing the anomalies immediately within the digital twin of the monitored structure. The deep learning model, trained on temporal sequences of vibration signals acquired during normal operating conditions, acquires a concise description of standard structural behaviour. During the inference phase, the reconstruction error is a metric to detect substantial deviations and produce real-time alerts. The paper delineates the issue, current advancements, methodology employed, outcomes, and a comparison with conventional procedures. The findings underscore the potential of merging IoT, DT, and DL to enhance structural monitoring, facilitating more efficient infrastructure management and mitigating the risk of unforeseen failures.