Prediction of long-term dynamic responses of a heritage masonry building under thermal effects by automated kernel-based regression modeling

Field monitoring via ambient vibration tests is a reliable approach to assessing the health and integrity of masonry structures. However, the implementation of long-term field monitoring is often expensive, time-consuming, and labor-intensive. The best solution to addressing this challenge is to leverage the capacity of machine learning for predicting dynamic responses. This chapter intends to propose an automated kernel-based regression method for predicting the modal frequencies of a heritage masonry building under seasonal thermal effects. The crux of this method is to select an optimum kernel regressor between Gaussian process regression and support vector regression by Bayesian hyperparameter optimization. Due to the importance of thermal effects on long-term monitoring of masonry structures, temperature records are used as the main predictors, while structural modal frequencies are the main responses for regression modeling. Results show that the proposed method is successful in predicting the dynamic responses of the masonry building with high prediction accuracy.