A novel fault diagnostic strategy based on time-varying dynamic PCA model has been presented, which makes use of continuous updating time-lagged data matrix to construct a changeable PCA model to fulfill the task of fault detection and identification.