Dynamic-Inner Canonical Correlation Analysis based Process Monitoring

Recently, a dynamic-inner canonical correlation analysis (DiCCA) algorithm has been developed for high dimensional dynamic data modeling, which provides separate and explicit modeling of dynamic components and static components in the data. In this paper, a DiCCA based process monitoring algorithm is proposed. Compared to existing statistical process monitoring algorithms, the proposed algorithm has three advantages. First, the proposed algorithm is able to monitor dynamic variations and static variations separately. Second, the predictions of the dynamic components from the past values are utilized to shrink the uncertainty in process monitoring, leading to reduced type II errors without increasing type I errors. Third, once a faulty sample is detected, the contribution of this sample is not further utilized for fault detection in future samples, mitigating fault propagation issues. Case studies on simulation datasets and the Tennessee Eastman dataset demonstrate the effectiveness of the proposed method.