Multiscale statistical process control with multiresolution data

Abstract

An approach is presented for conducting multiscale statistical process control that adequately integrates data at different resolutions (multiresolution data), called MR-MSSPC. Its general structure is based on Bakshi's MSSPC framework designed to handle data at a single resolution. Significant modifications were introduced in order to process multiresolution information. The main MR-MSSPC features are presented and illustrated through three examples. Issues related to real world implementations and with the interpretation of the multiscale covariance structure are addressed in a fourth example, where a CSTR system under feedback control is simulated. Our approach proved to be able to provide a clearer definition of the regions where significant events occur and a more sensitive response when the process is brought back to normal operation, when it is compared with previous approaches based on single resolution data. © 2006 American Institute of Chemical Engineers AIChE J, 2006