Locating and identifying damage in truss structures with modal analysis and artificial neural networks

This project is aimed at determining a method of investigation that can identify and evaluate damage in reticular structures, using vibration modes and artificial neural networks. In the field of civil engineering, the term Structural Health Monitoring (SHM) is used to identify the monitoring methods that allow for the identification of structural anomalies, allowing for the continuous and automated evaluation of the health of a structure. Deep learning methods applied to SHM are at the center of scientific research in recent years, thanks to technological progress and the introduction of computing tools with significant computational capabilities, capable of processing large amounts of data. In this context, the proposed supervised learning methodology is based on the use of a convolutional neural network (CNN), which is capable, based on a dataset of dynamic information related to the expected damage configurations, to recognize and classify the structural condition of a structure, identifying, localizing and quantifying any damage. The classification of the structural condition is based on the specific training of the network carried out on large amounts of information generated analytically from a structural model. The processing of the sample data allows the network to automatically identify the characteristics of interest of the problem, and to predict the structural condition for data in input not yet viewed. In this analysis, the Modal Assurance Criterion (MAC) parameter, a tool for comparing vibration modes, is used as an indicator of the structural health status. The problem of damage identification is evaluated, laying the foundations for further analysis, by applying the method to a case study, related to a flat model.