A Local Sensitive Discriminant Analysis Method Based on Mahalanobis Distance: Application of Industrial Process Fault Diagnosis

Industrial data suffers from high dimensionality and non-linearity in fault diagnosis. Therefore, it is important to extract effective features from measurement space to enhance fault diagnosis accuracy. In this paper, a Local Sensitivity Discriminant Analysis method based on Mahalanobis distance (LSDA-M) is proposed for fault diagnosis. LSDA-M not only focuses on the local geometric structure of high-dimensional data which is useful for subsequent classification, but also introduces discriminant information. Meanwhile, when measuring the distance between nearest neighbours, the Mahalanobis distance is used in LSDA to eliminate the correlation interference between variables. Finally, a group of Tennessee Eastman Process (TEP) data are utilized to validate the our proposed LSDA-M method and the positive results proved that our methodology is effective.