New Adaptive Moving Window PCA for Process Monitoring

Abstract Slow and normal process changes often occur in real processes, which lead to false alarms for a fixed-model monitoring approach. In this paper, two recursive PCA algorithms for adaptive process monitoring are studied. the first algorithm is based on Moving Window Principal Component Analysis (MWPCA), and the second is based on Forgetting Factors Principal Component Analysis (Recursive Weighted PCA). Furthermore, by changing the size and the shift of the window, also for the forgetting factor, we will see the influence of these changes on the monitoring performances. Then, adaptive forgetting factors will be used, for increasing the robustness against outliers. Using the same concept of varying forgetting factors, a new recursive algorithm for adaptive process monitoring based on Moving Window is proposed. By using the current model and the updated mean and covariance structures and an Adaptive Moving Window, a new model is derived recursively (AMWPCA). Based on the updated PCA representation the Q-statistic (SPE) (monitoring metric) is calculated and their control limits are updated. The feasibility and advantages of each algorithms is illustrated by application to Tennessee Eastman process.