Oscillation detection and causality analysis of control systemsReportar como inadecuado




Oscillation detection and causality analysis of control systems - Descarga este documento en PDF. Documentación en PDF para descargar gratis. Disponible también para leer online.

Causality analysis, fault diagnosis, oscillation detection, oscillation diagnosis, information transfer, impulse response analysis, connectivity map

Naghoosi, Elham

Supervisor and department: Huang, Biao Chemical and Materials Engineering

Examining committee member and department: Prasad, Vinay Chemical and Materials Engineering Shah, Sirish Chemical and Materials Engineering Tavakoli, Mahdi Electrical Engineering Kugi, Andreas Electrical Engineering

Department: Department of Chemical and Materials Engineering

Specialization: Process Control

Date accepted: 2016-07-07T10:49:46Z

Graduation date: 2016-06:Fall 2016

Degree: Doctor of Philosophy

Degree level: Doctoral

Abstract: Management of abnormal events in chemical processes requires detection and diagnosis of abnormal performance of individual elements of the system. Detection of abnormal performance is usually done by means of setting a control limit on measured variables. Abnormality due to any reason in one element, may propagate through feedback loops and interconnections through the process, downgrading the performance of the whole operation. Receiving several alarms due to one abnormality is quite common in industrial operations making it a challenging task to diagnose the root cause of the problem within a limited time. Causality graphs demonstrating how the measured variables relate to each other help the operators and engineers to short list the main faulty faulty variables which propagate the abnormality to the rest of the process. Causality graphs can be obtained based on deep process knowledge, process schematics or historical data. Qualitative causality graphs based on process knowledge or schematics, even though the most valuable ones, have some limitations and therefore, causality analysis based on historical data has gained a lot of attention. A part contribution of the thesis is to advance the causality analysis procedures proposing methodologies to extract more reliable information about cause and effect relations between recorded variables. This part of the study considers both causality analysis assuming linear relations between variables as well as nonlinear ones. Considering linear methodologies, a more appropriate model structure and parameter estimation methodology than the existing ones is proposed based on Bayesian framework. Estimating model parameters under Bayesian framework accompanied with a carefully designed prior probability for the parameters can solve some of the issues of traditional procedures. Also, a new procedure is proposed to decompose the energy transfer between variables in a way to obtain a complete picture on the different paths along which variables can influence each other in addition to providing an estimation of the energy transferred through each path independently. Regarding causality analysis with the assumption of nonlinear relations between variables, a more advanced methodology based on information transfer concepts such as mutual information and transfer entropy is proposed in order to more reliably detect the true relations between the variables. The work toward diagnosis of abnormalities leads this study toward developing more reliable algorithms to specifically detect and characterize oscillations in control loops. A methodology is developed to detect and estimate oscillation frequencies which could be otherwise hidden within noise and non-stationary trends in industrial variables. Oscillations may occur in control loops due to aggressive controller tuning, external disturbances or due to nonlinearity of the valve or the process itself. In order to help the root cause diagnosis of propagated oscillations, a novel method is developed to distinguish between these three types of oscillations. Oscillations caused by each one of these three sources have specific characteristics distinguishable from other types of oscillations. Examining the oscillation characteristics in the wavelet domain made it possible to develop a comprehensive methodology which is capable of detection and independent diagnosis of different oscillatory components of the variables. The proposed methodologies are verified through various simulations, laboratory experiments and industrial case studies.

Language: English

DOI: doi:10.7939-R3GB1XT2X

Rights: This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for the purpose of private, scholarly or scientific research. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.





Autor: Naghoosi, Elham

Fuente: https://era.library.ualberta.ca/


Introducción



Oscillation Detection and Causality Analysis of Control Systems by Elham Naghoosi A thesis submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Process Control Department of Chemical and Materials Engineering University of Alberta © Elham Naghoosi, 2016 Abstract Management of abnormal events in chemical processes requires detection and diagnosis of abnormal performance of individual elements of the system.
Detection of abnormal performance is usually done by means of setting a control limit on measured variables.
Abnormality due to any reason in one element, may propagate through feedback loops and interconnections through the process, downgrading the performance of the whole operation.
Receiving several alarms due to one abnormality is quite common in industrial operations making it a challenging task to diagnose the root cause of the problem within a limited time. Causality graphs demonstrating how the measured variables relate to each other help the operators and engineers to short list the main faulty faulty variables which propagate the abnormality to the rest of the process.
Causality graphs can be obtained based on deep process knowledge, process schematics or historical data.
Qualitative causality graphs based on process knowledge or schematics, even though the most valuable ones, have some limitations and therefore, causality analysis based on historical data has gained a lot of attention. A part contribution of the thesis is to advance the causality analysis procedures proposing methodologies to extract more reliable information about cause and effect relations between recorded variables.
This part of the study considers both causality analysis assuming linear relations between variables as well as nonlinear ones. Considering linear methodologies, a more appropriate model structure and parameter estimation methodology than the existing ones is proposed based on Bayesian framework.
Estimating model parameters...





Documentos relacionados