Practical implementation of temperature compensation strategies for real-time structural health monitoring of a pre-stressed concrete bridge

This study assesses the effectiveness of various temperature compensation strategies aimed at reducing temperature-induced variations in modal data within a real-time structural health monitoring (SHM) system. The case study focuses on a pre-stressed concrete railway bridge, equipped with a comprehensive sensor array, including thermocouples for temperature measurements and accelerometers for modal identification through operational modal analysis. The collected data are then processed in near real-time for unsupervised anomaly detection. A critical challenge in SHM systems is addressing temperature-related effects, which, if left unmitigated, can lead to an increase in false alarms and a decrease in the system’s ability to detect actual damage. In this work, field data are analyzed to compare various compensation methods, including subtraction-based approaches (using multivariate linear regression and deep neural networks) and projection-based techniques (such as linear and nonlinear principal component analysis). A damage index is defined using the compensated features, and each approach is evaluated based on its ability to minimize false alarms and enhance damage detection. The results underscore the importance of feature averaging for noise reduction prior to compensation and demonstrate that neural network-based compensation outperforms other methods in terms of damage detection accuracy.