Drexel University Home Pagewww.drexel.edu DREXEL UNIVERSITY LIBRARIES HOMEPAGE >>
iDEA DREXEL ARCHIVES >>

iDEA: Drexel E-repository and Archives > Drexel Theses and Dissertations > Drexel Theses and Dissertations > Probabilistic approach for sensor fault detection and identification (A)

Please use this identifier to cite or link to this item: http://hdl.handle.net/1860/68

Title: Probabilistic approach for sensor fault detection and identification (A)
Authors: Mehranbod, Nasir
Keywords: Fault location (Engineering);Chemical process control -- Mathematical models;Chemical engineering
Issue Date: 27-Feb-2003
Publisher: Drexel University
Abstract: Operation performance of chemical, petrochemical and biochemical processes can be enhanced considerably by an integrated advisory system that is capable of performing simultaneous tasks of (1) instrument fault detection and identification and (2) process fault detection and diagnosis. Lack of such advisory system, at present, is due to the fact that one of the assumptions of normal process operation and functional sensors is always required in existing methods that perform tasks 1 and 2. This dissertation presents a Bayesian-belief-network-based method that is capable of performing task 1 for both steady-state and transient conditions. The method does not need the assumption of normal process operation, thus it should allow one to perform simultaneous instrument fault detection and identification, and process fault detection and diagnosis. A new Bayesian belief network (BBN) model with discretized-adaptable nodes is proposed for fault detection and identification in a single sensor under steady-state and transient conditions. The single-sensor model can be used as a building block to develop a multi-sensor or stage BBN model for all sensors in a process under steady-state and transient conditions, respectively. A new fault detection index, a fault identification index, and a threshold setting procedure for the multi-sensor model are introduced. Single-sensor model design parameters (prior and conditional probability data) are optimized to achieve maximum effectiveness in detection and identification of sensor faults. In the context of BBN, conditional probability data represents correlation between process measurable variables. For a multi-stage BBN representing a transient process, the conditional probability data is required for each time instant during batch operation. This requires processing a massive data bank that reduces computational efficiency. A method is presented to minimize the required conditional probability data to one set. It allows for improving the computational efficiency without sacrificing detection and identification effectiveness. The method is applicable to model- and data-driven techniques of generating conditional probability data. Therefore, there is no limitation on the source of process information. The capabilities of the BBN-based model to detect and identify bias, drift and precision degradation in sensor readings are illustrated for single and simultaneous multiple faults by several case studies.
URI: http://dspace.library.drexel.edu/handle/1860/68
Appears in Collections:Drexel Theses and Dissertations

Files in This Item:

File Description SizeFormat
mehranbod_thesis.pdf637.73 kBAdobe PDFView/Open
View Statistics

Items in iDEA are protected by copyright, with all rights reserved, unless otherwise indicated.

 

Valid XHTML 1.0! iDEA Software Copyright © 2002-2010  Duraspace - Feedback