Dual-monitoring scheme for multivariate autocorrelated cascade processes with EWMA and MEWMA charts

This paper presents a dual monitoring scheme for multivariate autocorrelated cascade process control using principal components regressions. The autoregressive time series model is imposed on the time-correlated output variable which depends on many multicorrelated process input variables. A generalized least squares principal component regression is used to describe the relationship between product and its process input variables under the autoregressive regression error model. A dual monitoring scheme consisting of residual-based EWMA control chart, applied to product characteristics, and the MEWMA chart, applied to the multivariate cascade process characteristics, is proposed. EWMA control chart is applied to increase the detection performance, especially to small mean shifts. The MEWMA is applied to a selected set of input variables from the first principal component to increase sensitivity to detecting process failures. The proposed dual scheme for product and process characteristics enhances both the detection and prediction performance of the monitoring system of multivariate autocorrelated cascade processes. The proposed dual monitoring scheme outperforms the conventional residual type control chart applied to the residuals of the principal component regression alone. Implementation of the proposed methodology is demonstrated through an example from a sugar beet pulp drying process.