Dynamic fault diagnosis in chemical process based on SVM-HMM

Based on Hidden Markov Support Vector Machines (SVM-HMM) we present a novel dynamic fault diagnosis approach, in which the dynamic of chemical process is considered through augmenting each observation vector by using mean value and variance of the previous observations. Herein, SVM-HMM is a good method for dynamic continuous data which indentifies multiple kinds of faults with only one uniform discriminative model instead of multiple ones. A benchmark of Tennessee Eastman Process (TEP), a chemical engineering problem, is carried out to generate datasets to examine the performance of our new method. And the experiment results show the faults are identified more accurately applying the proposed method than that done by the state-of-the-art approaches.