Integrated physics-informed and unsupervised learning assessment of hemp-reinforced masonry barrel vaults under shaking table testing

The seismic assessment of strengthened masonry vaults through full-scale shaking-table testing remains limited, particularly for sustainable, heritage-compatible systems. This study investigates the dynamic response and damage evolution of full-scale masonry barrel vaults strengthened with a hemp-mesh biocomposite embedded in a cocciopesto matrix. A two-phase experimental campaign was carried out on two geometrically identical vaults subjected to progressively scaled three-component earthquake records, enabling the analysis of three representative structural configurations: unreinforced, newly reinforced, and damaged-then-reinforced. Reinforcement effectiveness and damage progression were assessed through a multi-indicator framework combining modal parameters with two proposed indicators: an energy-based indicator derived from Arias intensity and an autoencoder-based indicator. Each is formulated in a baseline-relative (excitation-dependent) form and in an input-normalized form where the energy-based indicator is normalized by the Arias intensity of the input and the autoencoder-based indicator by the squared RMS of the input. It follows that the input-normalized indicators isolate structural regime transitions from the excitation amplitude growth. Results show that the reinforcement, in both newly reinforced and damaged-then-reinforced configurations, enhances seismic capacity and damage tolerance, with delayed damage onset and reduced stiffness degradation. The proposed indicators consistently captured structural regime transitions and damage evolution across all configurations, showing strong agreement with modal trends. In particular, energy-based indicators provide a straightforward signal-level description of regime changes, while the feature-based autoencoder approach proves effective in detecting subtle statistical deviations that may precede visible damage, confirming its effectiveness in tracking stiffness degradation and nonlinear structural response.