Visualization and Dimension Reduction of High Dimension Data for Structural Damage Detection

This study presents a structural health monitoring method for damage identification and localization of a structure, which incorporated with the principal component analysis (PCA) based data compression and pattern recognition. Through the distribution of sensing node in the structure, an experimental twin-tower steel structure on shaking table test, the dynamic response at different point in the structure is collected. Two different damage scenarios are created in the structure: (1) buckling of the first story bracing member of tower A, (2) buckling of the 2nd floor bracing member of tower A. To extract the features from the vibration measurement, first, stochastic subspace identification and recursive subspace identification methods were applied to extract the modal parameters of their structural system. Then the FRF-based damage assessment using PCA data compression and the scalogram-based novelty detection are used to obtained patterns in some significant respect on damage assessment. Finally, based on the change of flexibility matrix of the system, damage location can be identified..