PSI - Issue 78

Ciro Canditone et al. / Procedia Structural Integrity 78 (2026) 1855–1862

1857

for the development of robust correlations between hazard parameters (e.g. intensity, position, distribution) and structural response (e.g. recurrent failure modes, variation in dynamic properties), which may help support the development and improvement of SHM systems. Within this work, a multi-model strategy for model-based SHM of heritage URM structures is proposed. The structural response of an Italian stone masonry archetype, dating from before 1919, is used as testbed. Hence, the building archetype, and its representativeness for a wider building stock, is first discussed. Then, the main features of two continuum-based and discontinuum-based numerical approaches for URM analysis, that is, Finite Element-based macro-modelling and Applied Element-based micro-modelling, are presented. Both numerical models are subjected to the same loading conditions, that is, the application of Gaussian soil settlement distributions at building’s base. A comparison is drawn between the numerically obtained initial modal shapes, frequencies, and stress distribution, as well as in terms of settlement-induced damage patterns. This advanced numerical output is then set to be used to train a robust surrogate model, which, based on the archetype’s representativeness of a wider building stock, will provide significant hindsight into the structural response of URM buildings to abnormal loading-induced damage and improve current design practices for SHM systems. 2. The proposed methodology To address model uncertainty in SHM, a multi-model framework integrating FEM and AEM models is proposed (Fig. 1). The methodology begins with the acquisition of structural and material data from the real structure to develop two independent numerical representations of the building: (i) FEM model – to generate settlement profiles and corresponding model features; (ii) AEM model – to generate damage due to settlement effects.

Figure 1. A schematic representation of the proposed methodology

The FEM model outputs settlement profiles and structural response features (such as displacements or modal features), which are used to train a surrogate model based on machine learning techniques. This surrogate model serves as a fast-executing predictive tool that can reproduce FEM-derived settlement behavior with minimal computational cost. To ensure robustness, the surrogate model is tested against outputs from the AEM damage simulations. This cross-model validation approach reduces bias associated with reliance on a single numerical model, thereby mitigating epistemic model uncertainties. Once validated, the surrogate model can be integrated into a model-based SHM scheme, enabling the reconstruction of settlement profiles from measured structural responses. The use of two complementary modeling approaches enhances reliability in SHM-based decision-making for cultural heritage structures.