Incipient Fault Detection Based on Exergy Efficiency and Support Vector Data Description

摘要

Automatic fault detection techniques for chemical processes are critical to process safety and reliability. Support vector data description (SVDD) has been widely used in the fault detection areas because of its fast calculation speed and low classification error. However, for incipient faults with slight changes of characteristics, the SVDD model has high complexity, and in addition, the feature sample selection of SVDD has a great impact on the effectiveness of fault detection. In the paper, the complexity of the SVDD model is not only reduced based on process exergy-data abstraction using the mutual information method, but also the proposed method presents great fault detectability and isolability. Meanwhile, the proposed method can detect incipient faults with different severity and indicate the evolution direction of faults. Therefore, the main contribution of this paper is to provide a novel fault detection method based on the EESVDD for incipient fault, in which the advantages of exergy data is combined with the SVDD method. Finally, the effectiveness of the proposed method is illustrated by a numerical simulation case and an industry distillation column, respectively.