Temperature Compensation in Vibration-Based Structural Health Monitoring Using Neural Network Regression

Vibration-based structural health monitoring (SHM) systems continuously estimate modal parameters to detect structural anomalies. The modal data corresponding to a healthy state are stored in a database during a training period, forming a baseline for comparison. However, variations in modal frequencies due to environmental and operational factors can lead to larger false positive rates and decrease the sensitivity of system to small damages, reducing the probability of damage detection. To mitigate these challenges, temperature compensation techniques are commonly employed to reduce variations in recorded modal data. In this paper, we propose a temperature compensation technique using neural network regression models. Unlike commonly used multivariate linear regression (MVLR), neural networks can capture the nonlinear relationship between temperature and modal frequencies effectively. The results of the numerical simulation in the present work demonstrate the superiority of the neural network-based compensation over the MVLR approach.