Fault detection and diagnosis based on transfer learning for multimode chemical processes

Abstract Fault detection and diagnosis (FDD) has been an active research field during the past several decades. Methods based on deep neural networks have made some important breakthroughs recently. However, networks require a large number of fault data for training. A chemical process may have several modes during production. Since fault is a low possibility event, some modes may have few fault data in history. Furthermore, collecting and annotating industrial data are extremely expensive and time-consuming. With scarce or unlabeled fault data, networks cannot be effectively used for modeling. In this paper, we present a FDD method based on transfer learning for multimode chemical processes. To overcome the fault data rareness and no label issues in some modes, transfer learning transfers the knowledge from a source mode to a target mode for FDD. Tennessee Eastman (TE) process with five modes is utilized to verify the performance of our proposed method.