Fault Monitoring of Batch Process Based on Moving Window SVDD

Traditional support vector data description (SVDD) methods fail to guarantee real-time monitoring online in relation to dynamical characteristics of batch process data. This study proposes a moving-window-based SVDD method to monitor faults online in real-time. By selecting a moving window with an appropriate size, and gradually updating the current sub-data space, sub-models of SVDD are constructed, thereby executing fault monitoring online. The proposed method not only solves the adverse effects of non-Gaussian and nonlinearity, but also considers dynamic characteristics of batch process data and improves the timeliness and accuracy of batch processes monitoring. Finally, a numerical example and industrial cases are presented to verify effectiveness of the proposed method.