This paper proposes a process monitoring method based on dynamic local keeping principal component analysis to solve the problem of strong correlation between sampling points in a dynamic process by constructing extended matrixes. Locality preserving projections(LPP) and principal component analysis(PCA) are combined to maximize the overall variance while keeping the local structure unchanged. On this basis, a process monitoring and fault diagnosis method based on the layered and block DLPPCA-SVM(dynamic locality preserving principal component analysis-support vector machine) is proposed for complex industrial processes with different characteristics. DLPPCA and PCA are used to model the sub blocks with different characteristics, and the support vector machine is used for fault diagnosis. The method is applied to the on-line monitoring and fault diagnosis of the TE process and generator set. The simulation results verify the effectiveness of the proposed method.