Parametric study on PCA-based algorithm for structural health monitoring

Using the measured frequency response function (FRF) data, a novel algorithm for structural health monitoring is proposed based on the principal component analysis (PCA) in this paper. First, the measured FRF data for a structure in both the healthy and the damaged states are used as the initial data. A PCA transformation is performed to obtain the features of an intact structure, in which an orthogonal transformation matrix packed by the first few eigenvectors of covariance matrix can be found. Further, the orthogonal transformation matrix is applied to the FRF data of damage structures to extract the features of structure in the damage state. By comparing the features of structure in the two states, a new damage feature index, namely the median values of the principal components (PCs), is proposed for structural health monitoring. Finally, the efficiency and robustness of the proposed method is illustrated and validated by using the real data provided by the Los Alamos National Laboratory, USA. This paper aims at investigating the effect of parameters, such as training data selection, sample points and so on. The illustrated results show that the PCA-based algorithm is correct and effective for structural health monitoring. The new damage feature index can reflect the structure state as well. The algorithm is purely based on the FRF data and independent of structure model, which makes it applicable for structural health monitoring in situ.