Acoustic Emission Monitoring for Mechanism Interpretation and Early Warning of Pigment-Layer Degradation in Mural Heritage under Dynamic Environmental Loading

The early mechanical damage caused by salt crystallization cycles in museum mural pigments generates weak and intermittent acoustic emission (AE) signals that are easily masked by noise. This study presents an AE-based ultra-early warning framework. Under controlled temperature--humidity cycling, AE and environmental data were synchronously collected. Asynchronous AE events and environmental series were aligned and aggregated into overlapping time windows. Multi-scale descriptors combining environmental trends, event statistics, burst features, and parameter histograms were then extracted to amplify faint precursors. The warning task was formulated as a sequence-to-future risk prediction problem, using 20~minutes of historical data to forecast high-risk AE activity within a subsequent 2~minute horizon. Warning thresholds were optimized via segment-level F$_1$-score under an explicit false-positive constraint, followed by temporal post-processing to generate coherent alarm episodes and a two-level alert scheme. A gated recurrent unit (GRU) model achieved the optimal balance of accuracy, episode recall, and alarm continuity, providing a mechanism-oriented tool for preventive protection of cultural relics.