PSI - Issue 78

Available online at www.sciencedirect.com

ScienceDirect

Procedia Structural Integrity 78 (2026) 1466–1473

XX ANIDIS Conference Towards the definition of fragility curves via machine learning Ebrahim Aminifar a , Mattia Zizi a, , Gianfranco De Matteis a, *

a Department of Architecture and Industrial Design, University of Campania “Luigi Vanvitelli,” Via San Lorenzo, Aversa (CE) 81031, Italy

© 2025 The Authors. Published by ELSEVIER B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of XX ANIDIS Conference organizers Keywords: Seismic vulnerability; Typological classification; Historical churches; Fragility curves; Machine learning 1. Introduction Historic masonry churches constitute a fundamental element of Italy’s architectural and cultural heritage, yet they are among the most vulnerable typologies in seismic-prone regions. Their structural fragility arises from a combination of geometrical features (e.g., slender walls without effective transversal connections, the presence of curved elements, etc.), aging materials, poor construction detailing, and the absence of seismic design principles. The repeated occurrence of earthquakes across Italy, such as those in Friuli (1976), Umbria-Marche (1997), Abruzzo (2009), and Abstract The seismic vulnerability assessment of historical buildings, particularly churches, is crucial for risk mitigation in seismic-prone regions like Italy. The study focuses on the typological characterization of existing Italian churches with the aim of deriving fragility curves through data-driven methods. Based on an available extensive dataset (Da.D.O. – Database Danno Osservato), a subset of single-nave churches was identified based on geometric and architectural criteria. The paper presents a detailed typological analysis of single-nave churches spread along Italian territory with the aim of evaluating the influence of parameters on seismic vulnerability. The results of these analyses will be used to generate a dataset suitable for future machine learning applications aimed at predicting fragility parameters based on the execution of a number of nonlinear static analyses. The research presents the typological framework and the methodology adopted for model selection, parameter variation, and the setup for developing a comprehensive fragility dataset. While the application of machine learning is beyond the scope of this study, the groundwork laid here is intended to support the integration of data-driven techniques in the seismic risk assessment of heritage buildings.

* Corresponding author. Tel.: +39 0815010823. E-mail address : gianfranco.dematteis@unicampania.it

2452-3216 © 2025 The Authors. Published by ELSEVIER B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) Peer-review under responsibility of XX ANIDIS Conference organizers 10.1016/j.prostr.2025.12.187