A progressive fault detection and diagnosis method based on dissimilarity of process data

In modern industrial process control, most traditional fault detection and diagnosis methods have been researched and applied widely. Recently, a novel MSPC method known as DISSIM has been developed focusing on continuous processes and batch processes, the result is significant. Firstly, this paper describes a progressive multiple variables fault detection and diagnosis method based on dissimilarity analysis, which is applied to continuous processes. Meanwhile, for the diagnosis of multiple variables fault, a fault can be detected when the dissimilarity index D is out of the control limit, the contributions of process variables can be computed and compared, then we can determine the first fault variable. If several variables are out of order simultaneously in the system, after reconstructing the first faulty variable, we can repeat the procedure until it is normal. Secondly, a process fault recognition method with faulty historical data is validated. Finaly, the performance of the proposed method in multiple variables fault diagnosis and the process fault identification method are validated through a numerical example and the Tennessee Eastman (TE) benchmark process respectively.