In this seminar, different methods that can be used for a rapid evaluation of the structural condition of bridges and buildings are discussed and validated using measurements of the dynamic response, both from numerical simulations and from field tests. Falling into the category of “data based” methods, such methods only need as input the time histories of the structural response measured at different locations on the structure and, when possible, in the case of seismic ground motion, the time histories of the ground acceleration, without any information on the geometry or mechanical characteristics of the structure.
Special attention will be given to methodologies that can be framed within a statistical pattern recognition framework, ideal for machine learning applications. Such methodologies focus on the recognition of certain patterns in the behavior of damage sensitive features, features that can be easily extracted from the time histories of the structural response. Quite popular in the field of speaker recognition, cepstral coefficients extracted from the time histories of the structural acceleration through simple digital signal processing are used as damage sensitive features, through the use of a Time-Delay Neural Network and of Generalized Autoencoders. These methodologies will be validated using field data from a real bridge (the Z24 bridge) that, before being demolished, went through an extensive monitoring campaign with progressive imposed damage.
March 21, 2024 3:00pm-5:00pm
The seminar will be given in ENGLISH and will take place in blended mode: in presence at the Aula Caveau - DISG, Faculty of Engineering, via Eudossiana 18, Rome and online via Zoom.
For the courses supplied by the PhD Program of the DISG credits are not provided: A certificate of attendance will be produced upon request (only after a check of the effective presence)