Efficient Supervised Machine Learning Techniques for Structural Health Monitoring

Abstract

This thesis presents supervised machine learning techniques using acceleration responses recorded from a small number of sensors. Ensemble-based traditional machine learning models are developed as a multi output regression model for the damage identification of the civil engineering structures using acceleration responses and impulse response functions extracted from it. Further, to improve the damage identification performance, a LSTM auto-encoder based multi output regression model is proposed. Finally, for a large-scale bridge, a 1D-CNN based damage classifier is developed using less number of sensors than the existing study.