The system discretizes the composite panel into regular Cartesian grids, connecting mesh nodes to represent spatial continuity. Sensor nodes are integrated using a K-nearest neighbor (KNN) strategy to the nearest mesh nodes.
A relative signal profile (r), calculated from the strain responses of the damaged structure compared to the healthy one (baseline), is generated through a finite element (FE) model and used as input to train the proposed MGN model.
The model was trained independently with two datasets: one for large damage regions and another for small damage regions. Results demonstrated reliable damage prediction in extensive damage scenarios, even with multiple regions, despite being trained exclusively on single-damage configurations.
Cross-material validation also showed the model's ability to generalize to other composite systems. However, the performance of the model retrained with data from a single small damage area substantially decreases when applied to scenarios with two small damage areas, although it can provide reasonable results for previously unseen single-small damage cases.
