Training-Free Structural Damage Localization Using Spatial-Correlation Sensor Networks: Full-Scale Validation on a Seven-Story Reinforced-Concrete Building.

Damage identification in instrumented structures is often framed through modal-parameter changes, finite element updating, or supervised classifiers. These approaches are powerful, but they require an explicit structural model, identified modes, labeled damage cases, or expert choices about reference sensors. This paper introduces a new data-driven and training-free approach with limited physical priors, defining a sensor network where each sensor is a node and the edges are defined from the spatial correlation of sensor responses. The idea is to use each sensor time history as the measured structural dynamics feature while damage is localized from the edges, whose correlations change relative to a baseline. The method is demonstrated on a full-scale seven-story reinforced-concrete shear-wall building tested at UC San Diego, considering four progressive earthquake-induced damage states and one brace-modification state. The results are compared with those obtained from a previously published finite element model. The results reveal that this network-based approach localizes the damage states in agreement with previous studies with limited prior requirements and low computational cost. Beyond damage localization, this network representation provides sensor centrality, allowing informative sensors to be selected from data rather than chosen randomly or only from experimental intuitions. For the case study, using this sensor network, we find the most central sensors, those carrying the most information with reduced trial-and-error and reduced expert intervention, and use them to recover the first three natural frequencies as a secondary dynamic check. The results show that spatial correlation networks can screen for damage, localize affected regions, and guide modal parameter extraction without building an FE model. This study opens a research avenue in which network representations of multi-sensor structural dynamics complement traditional modal analysis for structural health monitoring, with dense or heterogeneous sensing systems.