A methodology for robust fault detection in dynamic systems

This paper is concerned with the robust detection of abnormalitlies occurring in complex operations such as those found in industrial processes. A methodology for detection of process abnormalities is presented which uses only uncertain dynamic process models derived from historical operational data and little a priori fault information. The emphasis of the methodology is to provide a robust, on-line abnormality detector. An illustration of the application of this methodology is given for the on-line detection of abnormal reactor temperature variations in an ethoxolate reactor. This case study exhibits many of the above issues arising in process monitoring.