Loss factor analysis in real-time structural health monitoring using a convolutional neural network

Abstract

This study presents a novel approach to real-time structural health monitoring employing convolutional neural networks (CNN) to calculate a loss factor that measures energy dissipation in structures. As mechanical properties degrade over time due to service loads, timely detection of defects is crucial for ensuring safety. The loss factor, derived from the vibration energy spectrum, is used to identify structural changes, distinguishing between normal operation, the presence of defects, and noise interference. Using large data from real-time vibration signals, this method enables continuous and accurate monitoring of structural integrity. The proposed CNN model outperforms traditional models such as multilayer perceptron and long short-term memory, demonstrating superior accuracy in detecting early-stage defects and predicting structural changes. Applied to the Saigon Bridge, the method offers valuable insight into long-term structural behavior and provides a reliable tool for proactive maintenance and safety management. This research contributes to a machine learning-based solution for improving structural health monitoring systems in critical infrastructure.