Damage identification based on AR coefficients and PCA

In this paper, a new method of damage identification is presented based on the time series of structure acceleration data. Firstly, the acceleration data of undamaged structure is extracted and partitioned into several stream, the auto regressive (AR) coefficients of all data streams are served as reference sample. Secondly, the AR coefficients of damaged structure acceleration data are added into the reference sample separately, and several original data matrixes are constructed. Then, the principal component analysis (PCA) is used to extract the first two principal components (PC) of these matrixes, and the corresponding control ellipse are constructed. It is observed that the first two PCs of damaged structure are all out of the corresponding control ellipse. At last, the simulation experiment of a steel-made frame structure is used to test the validity of the method.