Dynamic process monitoring based on orthogonal locality preserving projections and exponentially weighted moving average

Following the intuition that the process data usually distribute on a low-dimensional structure and tend to be characterized by autocorrelation, we propose an approach integrating orthogonal locality preserving projections and exponentially weighted moving average (OLPP-EWMA) for process monitoring. The OLPP explicitly considers the low-dimensional manifold structure in the data and finds the orthogonal mapping from the input space to the reduced space. In order to capture the process dynamic behavior, the EWMA is combined with the traditional monitoring statistics to construct two new monitoring statistics. What is more, a novel contribution plots method is built based on the sensitivity analysis to identify the fault variables. The simulation results on the Tennessee Eastman benchmark process demonstrate that the proposed OLPP-EWMA method outperforms both the LPP and PCA in terms of the fault detection rate, and the built contribution plots method can effectively distinguish the fault variables from the normal variables.