A fault identification method based on asynchronous PCA

Traditional principal component analysis (PCA) only considers synchronous correlation between different variables. In this paper, a novel fault identification approach based on asynchronous PCA was proposed. Derivative dynamic time warping (DDTW) was used to obtain the best warping path between key variables and fault variables, and the elements in the best warping path were extended to two new series which reflected asynchronous correlation in-between. Eigenvalue decomposition was then executed based on the covariance of the new series to achieve a loading matrix. Finally, the fault pattern of snapshot dataset was identified by measuring similarity between the lower-dimensional representation vectors of the original key variables and the fault variables. Simulation results on a continuous stirred tank reactor (CSTR) process demonstrate that the proposed method can identify the pattern of fault snapshort dataset more effectively than the PCA similarity factor method. © 2016, Editorial Board of “Journal of Chemical Engineering of Chinese Universities”. All right reserved.