Lecture-Model-based methods in fault detection and diagnosis. By Janos Gertler

A significant class of fault detection and diagnosis methods relies on an explicit mathematical model of the monitored system. The fundamental idea is “analytical redundancy”: system outputs are compared to model responses, with discrepancies ideally indicating the presence of faults. The task of detecting such faults and determining their location (isolation) is made harder by noise and disturbances acting on the process, and by inaccuracies of the model. The discrepancies between system and model outputs are represented by “residuals”; it is then these residuals that are manipulated to facilitate detection and isolation. There are various approaches to residual generation, based on transfer function or state-space representations of the system, and using design techniques such as direct transformation, system inversion, geometric projection or observer design. We will review these methods and will show that, under identical circumstances, they all lead to the same residual generator.
The majority of model-based methods consider the model “given”, that is, determined outside the diagnostic framework, from first principles or by system identification. Another class, referred to as “data driven” methods, such as principal component analysis (PCA), incorporate model generation into the diagnostic procedure. We will show how PCA models are related to explicit models, and how PCA-based residual generators may be designed to have the same properties as analytical redundancy generators.

Friday, April 1, 2011  2:00 pm, Gray Conference Room, ISIS 1025 16th Avenue South

Janos Gertler was educated in Hungary. He graduated from the Technical University of Budapest, in Electrical Engineering, and then received the Candidate of Science (Ph.D.) and Doctor of Science degrees, in Control Engineering, from The Hungarian Academy of Sciences. For 10 years, he served as Vice Director of a large research institute in Hungary. He came to the US in 1981, and held visiting positions at Case Western Reserve University and the New York Polytechnic University, serving at the latter as Associate Dean of Engineering. He joined George Mason University in Fairfax, VA in 1985 as Professor of Electrical and Computer Engineering.

Dr. Gertler’s research interests have concerned various aspects of computer control and monitoring of engineering processes, including high-level programming, systems identification and, in the past 25 years, fault detection and diagnosis. He is the author of more than 170 papers, and a single-authored book on engineering diagnostics, and was a plenary speaker at nine international conferences. He led a six-year effort with GM’s Research Laboratory and Powertrain Division, aimed at developing system-level on-board diagnostic techniques for automotive engines; the algorithm is now running on several mass-produced GM models. Dr. Gertler is a Fellow of IEEE, Fellow and Advisor of IFAC (International Federation of Automatic Control), a Foreign Member of the Hungarian National Academy of Sciences, and a recipient of the Outstanding Research Faculty Award of the School of Engineering at George Mason. Most recently, he has developed an interest in the mathematical analysis of the effects of offshoring on the US economy.