Structural Damage Detection Based On Principal Components Analysis

Structural health monitoring (SHM) has played a significant role in the field of civil engineering in the last few decades. The primary task of an SHM system is damage detection and identification. Damage identification is of importance for quantitatively assessing the structural state, estimating the remaining service life of structure, and determining the corresponding reinforcement techniques. Generally speaking, the initiation and development of damage will cause the stiffness degradation and change the natural frequency and mode shapes. Most of the currently available vibration-based model updating and identification methods for structural damage detection are based on the idea of extracting the eigenvalues, mode shapes. In this thesis, based on the principal component analysis and multivariate control chart a structural damage identification approach was proposed using measured dynamic response data. Moreover, the approach was realized by employing LabVIEW for remote damage identification. The major research work of this thesis is introduced in detail as follows,1. A structural damage detection approach with principal component analysis (PCA) on mode shapes was proposed and validated with numerical simulation. The mode shapes of the frame structure in healthy status and damaged status are determined by finite element analysis. By coordinating transformation to highlight the variation of mode shapes according to principal components analysis, the damage existence and location were identified. With the Hotelling T2control chart for the analysis of the principal components, the abnormal data were separated and the information of damage on the structure was quantitative described. The effect of the noise level was also discussed.2. A structures damage identification method based on frequency response function (FRF) and PCA was presented. FRF is easily obtained through the response measurements and can be dramatically reflected the structural dynamic characteristics. In this thesis, FRF were reconstructed using principal component analysis, and the influence of original data noise was reduced or eliminated.3. The vector angle between the reconstructed FRF and baseline FRF was used as index of damage index and these angles were subjected to multivariate statistic control chart analysis for damage detection. The distribution of angles was verified by normally distribution test chart, and the exponentially weighted moving average (EWMA) control chart was applied to identify damage when the angles distribute normally and standard deviation (SD) control chart was used to identify damage with the angles disobedient normally distributed. The severity of damage was confirmed by the dispersion of angles, which was reflected in a number of points who exceed the control limit of the control chart and the statistic values of the exceed points.4. In order to validate the reliability of the proposed method, dynamical test on a4-story frame structure with3damage cases were carried out. The result shows that the proposed method is capable of identifying structural damage without requirement of values of modal parameter and extraction of modal shapes.5. In this thesis, a brief introduction of virtual instrument technique and LabVIEW software was given, and remote data transmission function was provided by adopting remote panel technology of LabVIEW.