Incipient Fault Detection Method Based on Stream Data Projection Transformation Analysis

Abstract

Early detection of incipient faults is a challenging task in the field of chemical process monitoring. For this problem, this paper proposes a new data-driven process monitoring method called stream data projection transformation analysis (SDPTA). First, we determine a set of projection transformation vectors, orthogonal basis vectors, based on original data to solve the problem that the data space original basis vector has relevance. Then, we use a sliding window to project data onto the basis vectors to obtain the basis vector components which is defined as projection transform components (PTCs). In this way, the stream data local sequence information can be utilized effectively. Furthermore, each PTC represents the coverage of local data on the corresponding basis vector. The length of PTCs can reveal some important process features, implying that condition changes can be detected by monitoring the length of PTCs. Finally, the potential of the window-based SDPTA method in monitoring continuous processes is explored using two case studies (a numerical example and the challenging Tennessee Eastman process). The performance of the proposed method is compared with the existing MSPM methods, such as PCA, DPCA, and RTCSA. The monitoring results clearly demonstrate the superiority of our method.