Κτηριακό κέλυφος : μελέτη επίδρασης της θερμομόνωσης και της υάλωσης στην ενεργειακή απόδοση κτηριών κατοικίας

This Diploma Thesis develops a generalized computational framework for the classification and prediction of the impact response of laminated composite plates. The methodology is based on the Impact Characterization Diagram and combines machine learning, multifidelity modelling and active learning. First, a Random Forest classifier is employed to automatically identify the response region as Local, Transition or Global. For the Local and Transition regions, a multifidelity model is developed in which the FSDT-TDSFE formulation is used as the low-fidelity model, while the more detailed CLW-TDSFE formulation provides the high-fidelity response. The discrepancy between the two fidelity levels is approximated through Gaussian Process regression, while Active Learning is employed to adaptively select the most informative high-fidelity simulations, thereby reducing the required computational effort. The final methodology is assessed through independent validation cases in order to examine its ability to predict the response of previously unseen impact conditions. For the Global region, a separate analytical damage dependent formulation is employed, in which delamination is incorporated through the reduction of the effective structural stiffness and represented within the Generalized Impact Characterization Diagram. Overall, the proposed framework combines the physical interpretation of the impact problem with data-driven prediction methods and enables the selection of an appropriate analysis strategy according to the identified impact-response region.