Research On Robust Multivariate Process Control And Diagnosis Methods

Robust process control has been widely used in industrial fields.Traditional multivariate process control methods mostly map observation data from high-dimensional space to low-dimensional feature space for monitoring.Managers may find it difficult to diagnose the source and extent of anomalies.With the development of quality management technology,more and more industries conduct on quality monitoring and abnormal diagnosis of multivariable processes.As various outliers,correlations,and generalized abnormalities exist,traditional monitoring and diagnostic methods may fail,which brings challenges to multivariate process control research.This work uses the robust anomaly recognition method DDC as the core and expands it for generalized abnormalities,which is helpful to achieve higher anomaly diagnosis accuracy.Both traditional and robust time series models,such as BIP-ARMA,are introduced into the DDC framework to adapt to different abnormal data.Single-series and multi-series methods will be compared whereas the applicability of the robust estimators BIP-TAU,BIP-S,and BIP-MM in different situations will be discussed.Better fusion models are proposed,and a robust multivariate dynamic model is introduced to realize real-time model updates and forecasts.Simulation research and empirical analysis are conducted to explore the applicability of the improved DDC methods.Six sets of simulation experiments are performed on multivariate time series data with isolated additive anomalies and patchy anomalies.It is found that the improved DDC methods perform better in isolated additive anomalies,and the accuracy of the re-predicted interpolation data is higher than the original model.The results of four real datasets show that the DDC methods has more accurate anomaly identification results than traditional methods and can feedback the source and degree of anomalies.A series of improved DDC methods help the model to monitor different types and the starting points of the anomalous propagation along the sequence.In conclude,not only can this work catch anomalies more efficiently and timely,but it is capable of better interpretation.The improved robustness of the DDC methods can well resist the interference of extreme outliers and realize the robust diagnosis of outliers,which will be helpful for practical applications and future research.