Optimal design of T~2 monitoring chart for chemical processes

摘要

In this paper, the optimal design of multivariate statistics-based monitoring charts for large scale systems is investigated. A cost function that explicitly accounts for the different quality monitoring costs is used to determine the parameters of the Hotelling's T~2 monitoring chart. The proposed approach is used to economically monitor the products in the Tennessee Eastman process (THP). The implementation of a multivariate economical criterion in the design of the Hotelling's T~2 control chart provides an improved basis for the evaluation and repairs of out of control states and results in monitoring's cost minimisation through optimal sampling schedules. Due to the nature of the used economic model, the decision of switching from traditional statistical design to economic design requires the balance between the gained cost saving and the expected statistical performance of the T~2 monitoring chart. Finally, in order to evaluate the reliability of the estimated optimal parameters, a detailed sensitivity analysis is presented.