Subspace Approaches to Dynamic Model Identification and Fault Diagnosis

Prof. Joe Qin, Ph.D.

   Zeit: Dienstag, 20. 11. 2001, 16:00
   Ort: Hörsaal V 9.31 Pfaffenwaldring 9, Universitätsbereich Stuttgart-Vaihingen

Abstract:

The number of industrial applications of multivariable model predictive control is reaching 4,600 and is rapidly growing. A key step in these applications is to build accurate dynamic models. In this talk we present some recent development in subspace approaches for building general dynamic models from process data. Principal component analysis will be used to achieve consistent models under input and output errors. The effectiveness of this approach is demonstrated using simulation and industrial examples. The second part of the seminar is concerned with detecting and identifying sensor faults using subspace models under dynamic operations. A unique way to identify faults with maximum sensitivity is presented. We then show how this method can be used for sensor validation for an industrial process.

Biographical Sketch:

Dr. S. Joe Qin is currently an Associate Professor in Chemical Engineering at University of Texas at Austin. He obtained his BS and MS degrees in Automatic Control from Tsinghua University in Beijing, China, in 1984 and 1987, respectively. His Ph.D. degree is in Chemical Engineering from University of Maryland. His current research interests include process monitoring and fault identification, model predictive control, run-to-run control, system identification, microelectronics process monitoring and control, chemical process monitoring and control, and control performance monitoring. He is a recipient of an NSF CAREER Award and is currently an Editor for Control Engineering Practice.