Novel Schur Decomposition Orthogonal Exponential DLPP With Mixture Distance for Fault Diagnosis

Currently, quickly identifying and handling fault types from massive, high dimensional, and nonlinear process data has become a major challenge in the industrial process. In order to solve this problem, this article proposes an improved discriminative locality preserving projection (DLPP) algorithm called Schur decomposition orthogonal exponential DLPP with mixture distance of Euclidean distance and Mahalanobis distance (EM-SOEDLPP) for industrial fault diagnosis. In EM-SOEDLPP, the mixture distance is used to enhance the differences and separability between different types of datasets. The problem of small sample size within DLPP is manifested in the singular matrix decomposition problem. EM-SOEDLPP not only addresses the small sample size problem by introducing matrix exponentiation, but also resolves the problem of redundant eigenvectors through the use of the Schur theorem for matrix exponentiation. Simulated experiments are executed on the Tennessee Eastman process. The results show that the EM-SOEDLPP method proposed in this article has higher accuracy in fault diagnosis compared with other fault diagnosis approaches.