A using of just-in-time learning based data driven method in continuous stirred tank heater

Abstract

As model-based methods have difficulty to solve more and more complex processes fault detection problems today, data-driven based techniques have been wildly used in industrial systems monitoring because of its ability to process unknown physical model. However, conventional static data-driven fault detection method have problems in processing nonlinear systems fault detection with deterministic disturbances in nonlinear system. In order to deal with this, a method called just-in-time learning based data-driven (JITL-DD) was invented. In this method, JITL is used for learning the nonlinear model and the disturbances to predict the output. The residuals of the predict and real one will be processed by static data-driven method to decide wether it has fault. In this article, A numerical example will be used to test the algorithm and a case study of CSTH are proposed to show the performance of JITL-DD method. As comparisons, JITL-PCA method is also employed to solve the same problem.