Novel Mahalanobis Distance Based Fault Diagnosis Using Discrimination Neighborhood Preserving Embedding for Industrial Process

With the advancement of technology, the data collected by sensors have high-dimensional, non-linear characteristics. These data are difficult to be processed by traditional fault diagnosis methods. In this paper, an advanced fault diagnosis method based on discrimination neighborhood preserving embedding of Mahalanobis Distance (DNPE-M) was proposed. The proposed new method solves the problems of classification accuracy and data overlapping. Firstly, the high-dimensional and nonlinear data are dimensionally reduced by discrimination neighborhood preserving embedding based on the Mahalanobis Distance. Secondly, the fault data are classified using the integrated learning classifier AdaBoost. Finally, the Tennessee Eastman (TE) chemistry dataset is used to verify. The results of the experiments show that the proposed DNPE-M improves the performance of fault diagnosis accuracy.