چكيده به لاتين
Structural health monitoring has been the subject of many researches in various engineering fields, including civil engineering, in recent years. The main objective of these studies is to detect the presence of damage in a structure and specify its location and severity. Data from the sensors installed on the structure are used for this purpose. The remarkable advances in sensors and measurement technologies in recent decades have paved the way for getting access to different kinds of data in large amounts and thus, have led to a progress in structural health monitoring field. However, how to fuse these data and reach more accurate results based on them, is still a big challenge.
In this thesis, the concepts of structural health monitoring and the popular data fusion approaches used in this field are reviewed and the main concepts of machine learning and deep learning, as the most popular data fusion approaches in structural health monitoring in recent years, are explained. A new deep learning-based method is also presented to monitor the health states of civil structures. The proposed method uses a deep convolutional autoencoder to fuse raw acceleration data from multiple sensors installed on a structure and estimates the extent of damage in the structure. Unlike most machine learning-based approaches proposed for structural health monitoring which need data from the healthy state and different damaged states of a structure, the method presented in this thesis needs only the healthy state data in the training phase. This is especially useful because data from damaged states of a real civil structure is not usually accessible. The efficiency of the proposed method was studied through its application on two numerical models and a full-scale concrete bridge. The results show that this method can successfully detect and quantify damage in various damage scenarios and assess the global health state of the structure.
