ISA ADV REACT MEAS CNTRL-2015 Advances in Reactor Measurement and Control.pdf

1、Advances in ReactorMeasurement andControlMcMillan-ARCM2015.book Page i Thursday, October 16, 2014 12:12 PMMcMillan-ARCM2015.book Page ii Thursday, October 16, 2014 12:12 PMAdvances in ReactorMeasurement andControlby Gregory K. McMillanMcMillan-ARCM2015.book Page iii Thursday, October 16, 2014 12:12

2、PMNoticeThe information presented in this publication is for the general education of the reader. Because nei-ther the author nor the publisher has any control over the use of the information by the reader, both theauthor and the publisher disclaim any and all liability of any kind arising out of su

3、ch use. The reader isexpected to exercise sound professional judgment in using any of the information presented in a particu-lar application.Additionally, neither the author nor the publisher has investigated or considered the effect of anypatents on the ability of the reader to use any of the infor

4、mation in a particular application. The reader isresponsible for reviewing any possible patents that may affect any particular use of the information pre-sented.Any references to commercial products in the work are cited as examples only. Neither the authornor the publisher endorses any referenced c

5、ommercial product. Any trademarks or tradenames refer-enced belong to the respective owner of the mark or name. Neither the author nor the publisher makesany representation regarding the availability of any referenced commercial product at any time. Themanufacturers instructions on use of any commer

6、cial product must be followed at all times, even if inconflict with the information in this publication.Copyright 2015 International Society of Automation (ISA)All rights reserved. Printed in the United States of America. 10 9 8 7 6 5 4 3 2ISBN 978-0-87664-089-0No part of this work may be reproduced

7、, stored in a retrieval system, or transmitted in any form or byany means, electronic, mechanical, photocopying, recording or otherwise, without the prior written per-mission of the publisher.ISA67 Alexander DriveP.O. Box 12277Research Triangle Park, NC 27709Library of Congress Cataloging-in-Publica

8、tion Data in process.McMillan-ARCM2015.book Page iv Thursday, October 16, 2014 12:12 PMvAcknowledgmentsPublications by Bill Luyben and Greg Shinskey are the source of most of mydeep knowledge in reactor control, particularly in dealing with recycleseffects and open loop unstable processes. Most of t

9、he rest of my knowledgehas been gained by working with and simulating continuous fluidized bedreactors for monomers and batch reactors for polymers, specialty chemicals,ethanol, and biologics.There were many Monsanto and Solutia Directors and Senior Fellows whoplayed a key role, technically and pers

10、onally in my journey as an automationprofessional. Foremost among those who shaped my career was the late Dr.Jim Fair, Honorary Professor Emeritus at the University of Texas, the creatorand first director of the Engineering Technology (ET) Department. Being partof the ET Department allowed access to

11、 the best technical minds in the world,along with the freedom to explore and the encouragement to publish. My sec-ond ET director, Dr. Stan Proctor, shared his extensive knowledge in model-ing and control and provided direction and support. The appreciation of myindependence and humor by my last ET

12、director, Dr. Adrian Hutton, anotherTexan, enabled me to publish many funny and creative articles and books.The knowledge I inherited from the late Vernon Trevathan and gained fromconversations with Senior Fellows at Solutia and Monsanto, including Dr.Henry Chien, Dr. Bob Otto, Dr. Terry Tolliver, a

13、nd Stan Weiner, has been ofinestimable value. When I retired from Solutia, I was fortunate to have the opportunity tobecome a consultant to Emerson Process Management at their headquarters inAustin in 2003. Here I have had access to some of the best minds on the sup-plier side, including James Beall

14、, Terry Blevins, Mark Coughran, LouHeavner, Mike Ott, and Willy Wojsznis. Jim Cahill has helped me to effec-tively use social media. The support by Mark Nixon, manager of DeltaV R there is essentially no inverse response orprocess time constant in these gas reactors. The process dead time, the resi-

15、dence time (volume/flow), is a few seconds for the high flow rate in thesehigh capacity processes. The largest time constant in the loop is the tempera-ture sensor, which leads to possible attenuation and deception. Other gas flowreactors have recycle streams and heat integration that introduce comp

16、lex andfast dynamics. The tuning objective for these reactors is to minimize suddenchanges and overshoot in the PID output rather than in disturbance rejection.PID gains in the range of 1 to 4 are used, as larger sensor lags offer the abilityto increase the PID gain and reduce the amplitude of proce

17、ss oscillations, giv-ing an illusion of better control. This illusion can lead to product quality prob-lems and an increase in the lag by sensor selection, installation, andmaintenance practices. Liquid well-mixed continuous reactors have completely different dynamicsand tuning. The process time con

18、stant is the residence time. The process timeMcMillan-ARCM2015.book Page xi Thursday, October 16, 2014 12:12 PMxii Advances in Reactor Measurement and Controlconstant is large, approaching an hour for slow reactions. The process deadtime is of the turnover time and is generally just a few seconds. I

19、n thesereactors, the process time constant is 100 times or more the process dead time.The tuning rules minimize excursions of the process variable, maximizing dis-turbance rejection. The PID output is changed aggressively to minimize errorsin the PID input. Overshoot of the PID output is necessary t

20、o reach setpoint ina reasonable time frame with PID gains of 10 to 40 being common. The processresponse is approximated as near-integrating, and the integrating process tun-ing rules developed for batch processes are used. Batch processes have an extremely slow response and even more aggressivechang

21、es in the PID output are employed. Overshoot is necessary for the PIDinput to reach a new setpoint. PID gains of 20 to 80 are used for tight control.Runaway processes require high gains for stability. A key feature to enable tight control without upsetting utility systems fromaggressive changes in t

22、he PID output is external reset feedback, while ratelimits on the setpoint being manipulated by the PID output provide direc-tional move suppression. This PID feature will be shown to be helpful inmany other ways for control and optimization of all types of reactors. Despite the importance of reacto

23、r control and the wide variety of processresponses, which leads to a considerable spectrum of control strategies andtuning objectives, this is the first book dedicated to the subject of reactor mea-surement and control from a practitioners perspective. Concepts are devel-oped to help the reader unde

24、rstand the fundamental differences in reactorapplications and provide the ability to improve the performance of nearly alltypes of reactors. McMillan-ARCM2015.book Page xii Thursday, October 16, 2014 12:12 PM1System DynamicsMcMillan-ARCM2015.book Page 1 Thursday, October 16, 2014 12:12 PMMcMillan-AR

25、CM2015.book Page 2 Thursday, October 16, 2014 12:12 PM31System DynamicsThe types of process responses encountered in industry are defined based onan open loop test where the PID controller is in manual or remote output, sothere is no feedback correction by the PID to the process (i.e., there is nocl

26、osed loop response). This test removes the effect of the PID algorithm andtuning in defining the process response. Tests are done for step changes in thecontroller output both above and below the normal output. The processresponse is observed until the process can be identified. For processes thatsl

27、owly change or have no steady state, the identification of the dead time andthe subsequent maximum ramp rate in the right direction is sufficient. Forprocesses that can runaway (e.g., highly exothermic reactors), an open looptest must be terminated before the process starts to accelerate. For proces

28、ses where the ramp rate or acceleration is potentially too fast, thecontroller must never be in manual mode. For these and other systems wherethe feedback control must always be active, the identification of the open loopdynamics is done with the controller in automatic mode by step changes inthe se

29、tpoint or by pulses injected into the PID output. These pulses are stepchanges that are held long enough to identify the response. During the test,there should be no disturbances to the system, so that the process responseseen is entirely the result of the step changes in PID output or setpoint.The

30、response observed in these tests includes: response of the analog output,final control element (e.g., control valve or variable frequency drive), process,sensor, transmitter, analog input, and process variable (PV) input to the PID.The observed response includes the effect of velocity limits, dead t

31、imes, timeAn open loop response (response without feedback action) for a step change in PID output is used to define a control loops dynamics. For loops that must stay in automatic mode, the open loop response is computed from a step change in the PID setpoint or a pulse injected into the PID output

32、.McMillan-ARCM2015.book Page 3 Thursday, October 16, 2014 12:12 PM4 Advances in Reactor Measurement and Controlconstants, and gains in the automation system. Better terminology would be“open loop response” than “process response” because the observed responseincludes almost everything in the loop re

33、sponse. Also, a qualifier that desig-nates the source of the dynamics should precede the particular dynamic term(e.g., valve dead time, process dead time, and measurement dead time). 1-1. Sources of Loop DynamicsFigure 1.1 shows the types and locations of dynamics in a reactor control loop.In this f

34、igure and elsewhere in the book, “lag” is used as the shorter term for“time constant” and “delay” as the shorter term for “dead time”. Figures, text,and tables will use these terms interchangeably.All processes have multiple time constants in the response of process andautomation system components.

35、A time constant is the time required for thechange in component output to reach 63% of its final value after the outputstarts to respond to a step change in the input to the component. In reactors,the process time constants are primarily associated with heat transfer andwell-mixed volumes. Temperatu

36、re measurement time constants originatefrom sensor, thermowell and transmitter lags. The controller time constant isthe result of filters on the analog input or PID process variable. Processes with a steady state are termed self-regulating processes becausethese processes have internal negative feed

37、back that will result in the processoutput variable settling out at a final value for a change in a process input. Adisturbance or a change in a PID output that becomes a change in a manipu-lated variable, which is in most cases is a change in flow, is the source of thechange in the process input. F

38、or self-regulating processes (e.g., flow and con-tinuous composition, pH, and temperature control of small volumes), a firstorder plus dead time or second order plus dead time approximation is used.In a first order plus time dead time model (FOPDT), the largest time constantis identified along with

39、the open loop steady state gain and total dead time. InThe open loop response includes the dynamics of the automation system and process. Qualifiers should be added to the terms that contribute to the open loop response to identify the source of the dynamics.McMillan-ARCM2015.book Page 4 Thursday, O

40、ctober 16, 2014 12:12 PM1 System Dynamics 5a second order plus dead time model (SOPDT), the second largest time con-stant is also identified. The largest time constant is called the primary timeconstant and the second largest time constant is called the secondary timeconstant. In this book, the larg

41、est time constant is also termed the open looptime constant, signifying that the response is for the PID in manual mode (theloop is open). For processes where a steady state does not exist or is not attainable in a rea-sonable time frame, an open loop integrating process gain, dead time, and aFigure

42、 1.1 Types and Locations of Dynamics in the PID Control LoopSelf-regulating processes have an observable steady state when the PID controller is in manual mode. The dynamic response of self-regulating processes is defined by a total loop dead time, primary and secondary time constants, and an open l

43、oop steady state gain.Y is the fraction of a small time that becomes dead time (Y approaches 1 as ratio of small to largest time constant approaches 0) McMillan-ARCM2015.book Page 5 Thursday, October 16, 2014 12:12 PM6 Advances in Reactor Measurement and Controltime constant is identified. In sectio

44、n 1.1.3 on open loop gains, Equation 1.3ashows that an equivalent open loop steady state gain and primary time con-stant can be computed from the open loop integrating process gain. The con-cept of an equivalent primary time constant enables faster identificationmethods and better tuning rules.Runaw

45、ay processes have a dynamic response that can accelerate. Runawayprocesses are modeled by a positive feedback primary process time constant,secondary time constant, dead time, and an open loop runaway process gainthat can be computed for an exothermic reactor using the equations in Appen-dix F. Whil

46、e this model is useful for control system analysis and design, inindustrial applications the online measurement of the positive feedback timeconstant and the open loop runaway process gain is not practical becauseacceleration is prevented for safety reasons. 1.1.1 Dead TimesAll processes have multip

47、le sources of dead times that result in a total deadtime, which is defined as the time interval between the step change in PIDoutput and the first recognizable change in the PID input. Noise can delay thesignal recognition until the excursion is beyond the noise band, thus creating alonger dead time

48、. The observed dead time is frequently called the processIntegrating processes continually ramp when the PID is in manual mode. The dynamic response of integrating processes is defined by a total loop dead time, secondary time constant, and an open loop integrating process gain.Runaway processes wil

49、l continually move and eventually accelerate when the PID is in manual mode. The dynamic response of runaway processes is defined by a total loop dead time, positive feedback primary time constant, secondary time constant, and an open loop runaway process gain. In industrial systems, runaway processes are treated as integrating processes.McMillan-ARCM2015.book Page 6 Thursday, October 16, 2014 12:12 PM1 System Dynamics 7dead time. The observed dead time is really a total loop dead time ( ) that isthe sum of all the pure dead times and the equivalent dead times from all timeconstan