2.38

Tuning by Computer R. M. BAKKE

(1970)

Tuning, Modeling and Diagnostics Software Package Suppliers:

P. D. SCHNELLE

© 2006 by Béla Lipták

B. G. LIPTÁK

(1995)

M. RUEL

(2005)

ABB Inc. (www.abb.com) Adersa (www.adersa.asso.fr) Aspen Technology Inc. (www.aspentech.com) CyboSoft (www.cybosoft.com) DOT Products (www.dot-products.com) Emerson Process Management (www.easydeltav.com) ExperTune (www.expertune.com) Foxboro Co. (www.foxboro.com) Honeywell (www.acs.honeywell.com and www.honeywell.com/imc) Hyprotech, Ltd. (www.hyprotech.com) Invensys (www.invensys.com) JMP Software (www.jmp.com) Lifestar (www.qimacros.com) Minitab Inc. (www.minitab.com) MYNAH Technologies division of Experitec, Inc. (www.mynah.com) Northwest Analytical (www.nwasoft.com) Oil Systems Inc. — PI (www.osisoft.com) Omega Simulation Co, Ltd. (www.omegasim.co.jp) Pavillion Technologies Inc. (www.pavtech.com) Process Systems Enterprise Ltd. (www.psenterprise.com) ProTuner, Techmation (www.protuner.com) Rosemount Inc./Emerson Process Management (www.rosemount.com and www.emersonprocess.com) Shell Global Solutions (www.shellglobalsolutions.com) Siemens (www.sea.siemens.com/ia) SIMSCI-ESSCO division of Invensys (www.simsci.com) Software Toolbox (www.softwaretoolbox.com) Stochos Inc. (www.stochos.com) Trident Computer Resources Inc. (www.tridentusa.com) Wonderware (www.wonderware.com) Xanalog Corporation (www.xanalog.com) Yokogawa (www.yokogawa.com/us) Zontec Inc. (www.zontec-spc.com)

The tuning steps that were described in Section 2.35 can all be performed automatically under computer control. The pattern recognition type self-tuning controllers do this by applying the “open-loop” method of tuning during their “pretune phase” (step change in manual mode) and by using the “closed-loop” method to evaluate the loop performance during upsets. Computers can also perform “model-based” tuning. In that case, the performance of the proposed tuning constants is first checked on a mathematical model of the process, and only if they perform well on the model are they used on the real process. Because of the overlap between some of the earlier sections in this chapter and this one, it is suggested that the 446

(1985)

reader also read those to gain a complete understanding of the tuning capabilities of computers and of digital systems.

INTRODUCTION Computer programs that were developed to aid in the design of process control systems are gaining general popularity and acceptance. Several good commercially available packages exist. These computer programs include nonlinear simulation tools, linearizing routines, predictor and estimation techniques, linear quadratic Gaussian methods, time series, and

2.38 Tuning by Computer

frequency response techniques. This section will discuss the approach to a design problem using such methods. The maintenance of control systems is equally important as is a good workable design. Many good control schemes fall into disuse because of a lack of proper fine-tuning, failure to adapt to changing process requirements, or poor maintenance. As time progresses and the operating processes change in the plant, some controllers might have to be detuned to prevent cycling or switched to a manual mode of operation. Good maintenance cannot be substituted by tuning. If equipment fails, the loop should not be re-tuned; the piece of equipment should be repaired. Computer-aided control analysis tools can be thought of in two categories. First there are the techniques that are useful in control system design. These are typically used in the design of new plants or new control systems. The nature of such tools is most often offline because that is the safest mode for investigating new control methods. Maintenance-related software is the second category of the computer-aided tools of control analysis required in maintenance of existing systems. The nature of these techniques is frequently online. They involve trying to identify the process dynamics while the plant is operating. They use step testing techniques and/or recursive estimation methods for adaptive tuning. After the process model is updated, these techniques can be used to calculate new control system parameters. In addition to keeping the controllers tuned, other methods are available to improve the quality and reliability of process measurements. Overall process balance calculations and the use of predictor/estimator filters (i.e., Kálmán filters) can help to improve the quality of measurements. These better-quality measurements are contributing to better control performance, which will be discussed in more detail in the following paragraphs.

447

PROCESS MODELING A good control system can only be designed, if the process is well understood, so that it can be mathematically modeled (Sections 2.13 to 2.17). Some control systems are less sensitive to model errors than others. However, as the requirements for improved control performance increase, so does the need for model accuracy. Process control engineers use many types of process models. Each type of model is preferred for some group of applications. Table 2.38a provides a list of some of these computer models and of their typical applications. Time and Frequency Domains Software packages that are available to tune loops and to optimize a process can use either time series or frequency analysis techniques. To obtain the best PID tuning settings and the best process model, frequency response methods have several advantages over other methods. These are: • • • •

•

•

Frequency response methods require only one bump to identify the dynamics of the process. The bump can be made in manual mode or can be a set-point change in the automatic mode. The change can be a pulse, a step, or bumps of other characteristics. Frequency response methods do not require prior knowledge of the dead time or time constant of the process. In contrast, when using time response methods, one often needs an estimate of both the dead time and the time constant. In addition, when using time response methods, the type of the process model has to be known.

TABLE 2.38a Types of Computer Models Used in Various Control Applications Computer Model Type

Typical Application

Large-scale nonlinear dynamic simulation

Used for offline investigation of process, process dynamics, and control application; large effort required to generate models

Frequency response models, single-input single-output (SISO)

Used to investigate open-loop behavior and specify closed-loop feedback control

Frequency response models, multiple-input multiple-output (MIMO)

MIMO models are used to study interaction; model typically generated from state space models or plant testing

State space models, MIMO

Used for control system design with methods such as linear quadratic optimal and pole placement control; used for estimate/predictor models; models typically generated from linearization of simulation models or plant tests

Continuous and discrete linear models Time series models, SISO Difference equations

Used for adaptive control and state parameter estimation

Simple low order models SISO continuous time

Used for quick time domain model fits to plant tests for control-loop tuning

© 2006 by Béla Lipták

448

Control Theory

Currently the main tool employed for control system design is dynamic process simulation using basic physical laws such as time-dependent balance relationships of mass, heat, momentum, etc. This type of simulation effort is very time consuming and requires an in-depth understanding of the process and the availability of large dynamic modeling software packages, which are available commercially. Many of these packages have several common features. First, the simulation models require the input of many differential equations representing the process. The packages usually handle the numerical integration and output of the desired time variables with very little effort on the part of the user. Once the process has been adequately modeled, the control system is tuned by dynamic simulation. The time behavior of the closed-loop system is studied and is changed in order to try new intuitively devised schemes or control designs that are recommended by other computer design methods such as the linear quadratic optimal or the frequency response methods. Single-input single-output transfer function models allow control systems to be analyzed in the frequency domain. Computer-generated graphics are used by many control analysis packages to draw root locus, Bode plots, Nichols charts, and Nyquist plots. The individual transfer function models are typically generated from linearized differential equations, plant pulse or step testing, or transformed state space models (see Section 2.33). All of these procedures have been computerized, which makes the analysis more convenient, and are very helpful in control system design and specification in the offline mode. State Space Models State space models are nothing more than linearized matrix representations of multiple-input multiple-output or highorder differential or difference equations. The convenient matrix formulation of the state space models makes them great candidates for manipulation and study by computer. These linear models can be used for time simulation (over the range where the linearization is valid). They can be transformed to transfer function matrix representations. They can be used to formulate linear quadratic optimal controllers. They can also be used to build predictor/estimators (i.e., Kálmán filters). State space control models are very flexible and are a common starting point for many computer-aided control analysis packages. The state space model is useful for offline design and online analysis and recursive study. The above techniques are basically design tools used offline. One popular online use for state space models is the estimator/predictor configuration. Here the computer evaluates the state space model of the process alongside with the real process. The measurable information about the process is fed into the computer model. The model predicts or estimates those states of the process that cannot

© 2006 by Béla Lipták

be measured. This information is then used for the control of the process. The methods used for estimating unmeasured states vary. One popular method is Kálmán filtering. In this method the estimated states are updated with information from the process model and measure their inputs on the basis of the statistical properties of the measurement noise. Details about Kálmán filtering and other estimation/prediction techniques can be found in References 1 and 2. These filtering techniques improve measurement quality and utility, thus improving the entire control system. Certain modeling applications can be used to signal instrument problems by monitoring the deviation between the output that is predicted by the model and the output of the actual process. If this deviation becomes too large, it is possible that sensor maintenance is required. Time Series Models Time series models are also useful in adaptive control applications. Discrete time models are used to represent singleinput single-output systems processes. The parameters of the time series models are estimated using recursive techniques. The updated model resets the online controller to keep the loop adequately tuned. This again is an online computer application that is primarily aimed at improving control quality. The low-order single-input single-output continuous model consists of a series of time lags associated with loop elements (valve, process, transmitter, etc.). Usually, the controlled processes also have transport delay (pure dead time) resulting, for example, from transport lag in pipes. Unlike nuclear reactors and modern missiles, these processes are usually stable without feedback controllers, just operating in the open loop. A general mathematical structure for these simple processes is composed of a first-order lag and dead time, as described in Equation 2.38(1). This model has three parameters: gain, time constant, and dead time.

dc(t ) 1 = [ Km(t − t0 ) − c(t )] τ dt

2.38(1)

where t = variable time; dc(t ) = derivative of the controlled dt variable with respect to time; c(t) = controlled process variable, the output of the process; m = manipulated process variable; K = process gain; τ = process time constant; and t0 = td = Lr apparent process dead time. This mathematical structure has the advantage of constrained dimensionality. Dead time in the model conveniently lumps the higher-order process time delays and the actual processes’ dead time into a single parameter. This model is used to tune a simple feedback loop by fitting the model to the process. This fitting is done by control loop tunings to achieve the desired response.

2.38 Tuning by Computer

After these four values have been obtained, the following parameters can be determined:

Controlled variable (e (t))

SKp=Y

Lr Rr =X Manipulated variable (m(t))

449

Lr =to

T# Kp =

Rr =

X SLr

2.38(2)

Kp =

Y S

2.38(3)

Y S

τ=

Kp

2.38(4)

Rr

Modified Step Response S Time Processes that drift during the conventional step response test will invalidate the measured parameters X, Y and Lr.

FIG. 2.38b First-order response to a step change in manipulated variable, while the controller is in manual, can determine the process gain (Kp) and dead time (Lr).

The conventional step response test is not practical in loops with large time constants. Furthermore, it does not provide a check on the accuracy in performing the test itself unless repeated. Even if the test is repeated, errors can still occur in unit conversion. A combination of a modified step response and an ultimate sensitivity test overcomes the above limitations (Figure 2.38c). When performing the modified step response, the operator retains the step just long enough to be certain that the process Controlled variable (c (t))

FITTING SIMPLE MODELS Step Response The conventional step response test (Figure 2.38b), which is performed while the controller is in manual, is probably the most popular method for fitting the three-term process model. This method works well for process loops that, after an upset, will achieve a new steady state. In addition, the common characteristics of these processes is that their time constant τ is small and that they do not appreciably drift during the test. However, if the process gain (Kp = Y/S) is unknown, which is often the case, this technique has the disadvantage that the step size (S) must be selected by trial and error. This is because too small a step might provide a reaction curve that is hard to read, and too large a step might result is an unacceptable upset. To fit this three-parameter model to the process while using this test, the following measurements have to be made manually or automatically: 1. Step size of the manipulated variable (S) 2. The apparent dead time (Lr = t0) 3. The total change in the controlled variable (X = Lr Rr) over the period of the apparent dead time (Lr = t 0 ), if the controlled variable is changing at its maximum rate of change (Rr) 4. The final value controlled variable’s response (Y )

© 2006 by Béla Lipták

X=Lr Rr Lr = to

Manipulated variable (m (t))

Time

Modified step response

S

Time ts Controlled variable (c (t))

Pu Time ts

Ultimate sensitivity

FIG. 2.38c The combination of the modified step response test and the ultimate sensitivity test is better suited for processes with dead time.

450

Control Theory

response reaches the maximum rate of change and then returns the manipulated variable to its initial value. Because in this test the controlled variable does not have to achieve a new steady state, the selection of step size S is not critical. The maximum rate of rise in this case too is obtained by Equation 2.38(2). This technique should not be confused with pulse testing, which requires analysis of the total pulse response curve. Ultimate Gain Test The ultimate gain (Ku ) test calls for increasing the gain of a proportional controller until the process cycles at constant amplitude, without convergence or divergence. The following equations describe the behavior of the system: Ku K p 2

 2 πτ   P  +1  u 

=1

 2 πτ  2 π Lr tan − 2   + P =π  Pu  u

2.38(5)

Process gain can be computed from this test by Kp =

1 et0 /τ K ud et0 /τ − 1

2.38(10)

The amplitude of cycling obtained with the discrete test will be different from that of the continuous test, which provides an additional means of checking whether the gain is linear. The degree to which this test can be automated is strictly a function of available hardware and software. If a valid model cannot be obtained, the instrument engineer should tune a controller by trial and error, or, if a more scientific approach is desired, then more general model structures and identification approaches should be tried. If the process gain is nonlinear, frequent adjustment or online adaptation should be considered. The reader should also refer to the tables provided in Section 2.35, which provide equations to calculate the tuning parameters for a variety of tuning objectives. The tests described can be automated by the use of software packages, which not only perform these tests but also can analyze some of the results.

2.38(6) TUNING CRITERIA AND FORMULAS

where Ku is the ultimate gain of the controller and Pu is the ultimate cycle period. These equations hold true if proportional-only control is continuously applied or if sampling is used at such a rate that it is apparently continuous with respect to either the dead time (Lr) or the time constant (τ ). Equations 2.38(5) and 2.38(6) can be reduced to determine the process time constant and the gain:

τ=

Pu 2π

  2 π Lr     tan  π − Pu    

1 Kp = Ku

2.38(7)

2

 2 πτ   P  + 1

2.38(8)

u

With τ calculated from Equation 2.38(7) and the gain from Equation 2.38(8), Equation 2.38(4) can be used to give a cross-check on the accuracy of the test and on the validity of the model structure. Discrete Cycling The process gain can be determined by another equation by discrete cycling of the loop while using a sample rate, which equals the dead time (Lr = t0). If the gain required to reach ultimate sensitivity is Kud , then the conditions of stability are defined by: K ud K p [1 − et0 /τ ] = 1

© 2006 by Béla Lipták

2.38(9)

For those processes that fit the first-order plus dead time model, a variety of approaches are described in the literature for computing the tuning settings. With conventional control applications, the techniques that have been developed were found to work well. A set of equations are provided below to give guidance in the tuning of controllers. These transfer function equations allow the calculation of the manipulated variables M(s) (where s is the Laplace operator), for the control modes P, PI, and PID, as a function of the process error E(s): Control Mode: P

M(s) = Kc E(s)

2.38(11)

Control Modes: PI

M (s) = Kc (1 + 1/Ti s) E (s)

2.38(12)

Control Modes: PID

M (s) = Kc (1 + 1/Ti s + Td S ) E (s) 2.38(13)

In the following, a number of tables are provided that contain some traditional tuning criteria and the corresponding tuning equations for P, PI, and PID controllers, when they are tuned to correct for load disturbances in the process. Table 2.38d lists a variety of tuning criteria that have been discussed in earlier sections. Tables 2.38e, 2.38f, and 2.38g list the equations to be used in determining the proportional, integral, and derivative settings for the various criteria listed in Table 2.38d. Depending on which performance criterion is selected from Table 2.38d and on the number of control modes, the formulas can be selected manually or automatically (perprogramming) from the next three tables.

2.38 Tuning by Computer

TABLE 2.38d 1 Controller Tuning Criteria Selection Table ISE — 1 =

ISE — 2 =

ISE — 3 =

IAE — 1 =

IAE — 2 =

IAE — 3 =

∞

∫

0

∫

θ0

∫ ∫ ∫ ∫

∞ 0

∞

[c(t ) × c(∞)]2 dt

I A t  =  0 τ τ Ti

2

 c(t ) − c(∞)    dt  c( ∞ ) 

θ0 ∞ 0

∞ 0

ITAE — 2 =

∫

θ0

ITAE — 3 =

∫

∞

∞ 0

Criterion

c(t ) − c(∞) dt

B

c(t ) − c(∞) dt IAE

c(t ) − c(∞) dt c( ∞ )

ITAE

c(t ) − c(∞) t dt

Controller Mode

A

B

Proportional

1.305

–0.960

Reset

0.492

–0.739

Proportional

0.984

–0.986

Reset

0.608

–0.707

Proportional

0.859

–0.977

Reset

0.674

–0.680

c(t ) − c(∞) t dt c(t ) − c(∞) t dt c(t )

TABLE 2.38g Tuning Equations for PID Controller Tuned for Correcting for a Load Disturbance Kc =

TABLE 2.38e Tuning Equations for Proportional-Only Controller Tuned for Correcting for a Load Disturbance Kc =

A  t0  Ku  τ 

B

A  tu  Ku  τ 

u

I A t  =  0 Ti τ τ

B

t  Td = τ A  0  τ

B

Constants

Constants Criterion

A

B

Ultimate

2.133

– 0.877

1

B

Constants

ISE

∞

A  t0  Ku  τ 

[c(t ) − c(∞)]2 dt

0

∫

ITAE — 1 =

TABLE 2.38f Tuning Equations for PI Controller Tuned for Correcting for a Load Disturbance Kc =

∞

451

Criterion ISE

Controller Mode

A

B

Proportional

1.495

–0.945

1.235

– 0.924

Reset

1.101

–0.771

ISE—1

1.411

– 0.917

Rate

0.560

1.006

ISE—2

0.9889

– 0.993

ISE—3

0.6659

– 1.027

Proportional

1.435

–0.921

IAE—1

0.9023

– 0.985

Reset

0.878

–0.749

IAE—2

0.6191

– 1.067

Rate

0.482

1.137

Proportional

1.357

–0.947

4

Decay

IAE

IAE—3

0.4373

– 1.098

ITAE—1

0.4897

– 1.085

ITAE—2

0.4420

– 1.108

Reset

0.842

–0.738

ITAE—3

0.3620

– 1.119

Rate

0.381

0.995

© 2006 by Béla Lipták

ITAE

452

Control Theory

Temperature control algorithm

c (t)

Feed

Controlled variable temperature, °F (°C) m (t)

Flow control algorithm

Reboiler

5 Minutes 223 (106) 221 (105)

Steam

219 (104) Manipulated variable (steam flow, %)

Bottom product

Modified step response

Drain 100

FIG. 2.38h The tuning example is based on the temperature cascade loop shown in this figure. Time

0 The operator performs the modified step test on the open loop process.

EXAMPLE In this example the previously outlined tuning method will be applied to a pilot plant distillation unit under the direct digital control of a time-shared computer. Figure 2.38h illustrates the unit process on which the process operator in cooperation with a computer tunes the cascade loop, which controls the temperature of this unit. Figure 2.38i shows the modified step response method of testing, where the upper two plots show how the operator applies a step in the open loop and obtains a response of the controlled variable, which allows for the determination of the ultimate sensitivity. This is then followed by the closed-loop portion of the test, which is shown in the lower two plots. Here the ultimate gain of the process is obtained. Once the data are collected, Equation 2.38(8) is used to calculate the process gain (Kp = 74%) This can than be verified by using Equation 2.38(4) to also calculate Kp = (time constant – T )(maximum rate of rise – Rr), which is considered to be adequate verification.

Tuning and Diagnostic Software Some software packages include a variety of automatic means for obtaining better process models and improved 2 tuning settings and also provide data for better maintenance. The information below is taken from Section 5.6 of Volume 3 of this handbook, where the topic is treated in more detail. Some of these software packages can perform a variety of analyses (auto-correlation, cross-correlation, power spectral density, statistical, linearity, etc.) They can also identify a variety of process models (integrator, dead time, first-order and higher orders, etc.); measure signal and valve noise; evaluate valve hysteresis, stiction, and nonlinearity; and evaluate the time and frequency response of the loop. In addition, they can evaluate and recommend control algorithm variations (error, reset, filtering, and sampling). These advanced software packages can also provide performance analysis (IAE, ISE, response time, robustness,

© 2006 by Béla Lipták

Controlled variable temperature, °F (°C) 5 Minutes 223 (106) 221 (105) 219 (104) Manipulated variable (steam flow, %)

Ultimate gain

100

0 Operator closes the loop with a simple proportional controller to find the ultimate gain.

Time

FIG. 2.38i The modified test response (top two plots) is obtained in the open loops as a step response, while the ultimate gain is determined (bottom two plots) in the closed-loop mode.

variability, statistics) and can suggest multivariable control strategies (cascade, feed forward, Smith predictor). EXAMPLE A flow loop in a refinery was analyzed and tuned using such a software package. Figure 2.38j illustrates the test for valve hysteresis, which was completed in the manual mode in less than 3 minutes. Software tools were used to analyze the data collected. It was found that the gain of this flow process varied from 1.3 to 2.4, which gives a linearity (maximum gain over

2.38 Tuning by Computer

453

50.

PV FICI41 48.09

35.

900

920

940

960

980

1000

1020

1040

1060

1080

1040

1060

1080

60. C2 C3 C1 CO FICI41 55.05

50. 900

920

940

960

980

1000

1020

916.3 Time (seconds)

FIG. 2.38j Valve hysteresis test results, which were obtained in less than 3 minutes of manual testing.

minimum gain) of 1.8; the response time of the loop exceeded 1 minute; hysteresis was under 0.1%; and stiction (the sticking of the valve) was 1.1%. Based on the data collected, it was found that the process gain is 2.3, the dead time of the process is 3.6 seconds, and the time constant is 2.4 seconds.

PV FIC141 Statistics Sample (raw) 0.33 Num of Points 1819 0 Time Min Time Max 801 PV Min 46.47 49.56 PV Max 3.088 Range Mean (µ) 47.99 Standard 0.544 deviation (δ) µ±δ 47.446−48.534 µ±2δ 46.902−49.078 Variance 0.2959 Variability 2.27% 39 Variability CO FIC141 Statistics Travel 8.13 14 Reversals

= 2.176

878/day 1510/day

As a consequence of this evaluation, the sticking valve 3 was repaired, and then the loop was re-tuned. Figure 2.38k provides a tabulation of the performance statistics of this flow loop before (left) and after (right) of the tuning. Figure 2.38l provides a visual indication of the valve sticking (left) before maintenance was performed

PV FIC141 Statistics Sample (raw) 0.33 Num of Points 1817 0 Time Min 800 Time Max 46.82 PV Min 48.97 PV Max 2.156 Range Mean (µ) 47.65 Standard 0.258 deviation (δ) µ±δ 47.392 − 47.908 µ± 2δ 47.134 − 48.166 Variance 0.06656 Variability 1.08% Variability 0.475 CO FIC141 Statistics Travel 0 Reversals 1

= 1.032

0/day 108/day

FIG. 2.38k Performance data of the flow loop before (left) and after (right) the correcting of the sticking valve and re-tuning of the controller.

© 2006 by Béla Lipták

454

Control Theory

49.5 PV

49.5 46.5

100

200

300

400

500

600

46.5

700

100

200

300

100

200

300

400

500

600

700

400 500 Time (s)

600

700

CO

56.25

CO

56.25 54.25

100

200

300

400

500

600

700

54.25

Time (s)

FIG. 2.38l Visual description of the performance of the flow loop before (left) and after (right) the correcting of the sticking valve and re-tuning of the controller.

on it. As can be seen, the controller output (CO) had to change over 1% (from about 55 to 56%) before the valve would respond. The improvement in loop performance can be seen by comparing the fluctuation of flow between 46.5 and 49.5% (left) to about half as much (right), achieved by the fixing of the valve and the re-tuning of the loop.

CONCLUSIONS With the various techniques described in this section, the operator can check the controller tuning and the process model accuracy periodically or when operating conditions change. Up-to-date modeling and tuning should greatly improve process performance and safety because most of these techniques maintain process safety by allowing the operators to remain in the loop during the critical testing phase. In addition, cost reductions are also realized when reducing the amount of trial-and-error tuning because the operator need not wait to bring the process to a final value in the step response test. This saves time when dealing with slow processes. Test signal size is not critical, reducing trial-and-error selection. Time-shared systems with DCS or supervisory control have greater scaling and sensitivity capabilities than conventional industrial instruments. Also, it can perform unit conversion and can be easily programmed to abort the tests automatically and set off alarms if necessary. In general, the computer can be a very valuable tool for control system analysis and design in both online and offline testing. Offline techniques range from full-blown computer simulations of process and the associated control system to simple linear first-order plus dead time type models generated from step testing. Online techniques can involve state space optimal control (Section 2.33) and Kálmán filtering tech4 niques, or they can be as simple as some of the automated tuning procedures that were discussed in this section. The computer has become a very important tool, and software is widely available for the process control and instrument engineer to use to implement a variety of offline

© 2006 by Béla Lipták

techniques of testing. Online techniques are also becoming more popular as the process control computer is becoming part of most control system installations. Testing, modeling, and tuning improve process performance and result in cost reductions by reducing the amount of trial-and-error tuning. Time-shared systems with DCS, PLC, or supervisory control offer a convenient vehicle to assist operators in tuning, while data acquisition can be done automatically. In minutes a loop can be analyzed and re-tuned.

References 1. 2.

3. 4.

Lopez, A. M., Miller, J. A., Smith, C. I., and Murrill, P. W., Instrument Technology, Research Triangle Park, NC: ISA, November 1967, p. 57. Lipták, B. G. (Ed.), Instrument Engineers’ Handbook: Process Software and Digital Networks, Boca Raton, FL: CRC Press, 2002 (Section 5.6). Ruel, M., “Stiction: The Hidden Menace,” Control Magazine, Vol. 13, No 11, November 2000, pp. 69–76. Blevins, T. L., Brown, M. McMillan, G., and Wojsznis, W. K., Advanced Control Unleashed, Research Triangle Park, NC: ISA, 2003.

Bibliography Aplevich, J. D., The Essentials of Linear State-Space Systems, New York: John Wiley & Sons, 2000. Astrom, K. J., and Wittenmark, B., Computer-Controlled Systems: Theory and Design, Englewood Cliffs, NJ: Prentice Hall, 1997. Badgwell, T. A., and Qin, S. J., “Review of Nonlinear Model Predictive Control Applications,” in B. Kouvaritakis and M. Cannon, Eds., Nonlinear Predictive Control: Theory and Practice, IEE Publishing, 2001, pp. 3–32. Bertsekas, D. P., Dynamic Programming and Optimal Control, Vols. I and II, Athena Scientific, 2001. Blevins, T. L., McMillan, G. K., Wojsznis, W. K., and Brown, M. W., Advanced Control Unleashed, Research Triangle Park, NC: ISA, 2002. Blevins, T. L., Brown, M., McMillan, G., and Wojsznis, W. K., Advanced Control Unleashed, Research Triangle Park, NC: ISA, 2003. Bryson, A. E., Dynamic Optimization, Reading, MA: Addison-Wesley, 1999. Chan, P. C., Process Control, A First Course with Matlab, New York: Cambridge University Press, 2002, pp. 21–24.

2.38 Tuning by Computer

Cheng, G. S., and Huo, L. Q., “Leading Edge Plant Control System — Model-Free Adaptive Control Puts Ling-Yuan Iron and Steel in League of Its Own,” InTech, October 2003. Corripio, A. B., Tuning of Industrial Control Systems, Research Triangle Park, NC: Instrument Society of America, 1990. Dahlin, E. B., “Designing and Tuning Digital Controllers,” Instruments and Control Systems, June 1968. Goodwin, G. C., Graebe, S. F., and Salgado, M. E., Control System Design, Englewood Cliffs, NJ: Prentice Hall, 2001. Haykin, S. S., Neural Networks: A Comprehensive Foundation, 2nd ed., Englewood Cliffs, NJ: Prentice Hall, 1998. Higham, J. D., “Single-Term Control of First- and Second-Order Processes with Dead Time,” Control, February 1968. Howell, J., and Rost, D., “PC Software Tunes Plant Startup,” InTech, June 1988. Kraus, T. W., and Myron, T. J., “Self-Tuning PID Controller Uses Pattern Recognition Approach,” Control Engineering, June 1984. Lelic, M. A., and Zarrop, M. B., “Generalized Pole-Placement Self-Tuning Controller,” International Journal of Control, 46, 1987, pp. 569–607. Luyben, W. L., Process Modeling, Simulation and Control for Chemical Engineers, 2nd ed., New York: McGraw-Hill, 1990. Lyashko, S. I., Generalized Optimal Control of Linear Systems with Distributed Parameters, Dordrecht: Kluwer, 2002. Maciejowski, J. M., Predictive Control with Constraints, London: Pearson Education Limited, 2002. Mansy, M. M., McMillan, G, K., and Sowell, M. S., “Step into the Virtual Plant,” Chemical Engineering Progress, February 2002, pp. 56–61. Marlin, T. E., Process Control, Designing Process and Control Systems for Dynamic Performance, 2nd ed., New York: McGraw-Hill Higher Education, 2000. McMillan, G. K., Tuning and Control Loop Performance, 3rd ed., Research Triangle Park, NC: ISA, 1994 McMillan, G. K., Process Industrial Instruments and Control Handbook, New York: McGraw-Hill, 1999. McMillan, G., “A New Era for Model Predictive Control,” Control, Vol. 13, No. 5, pp. 45–50, 2000. McMillan, G. K., and Cameron, R. A., Models Unleashed — Virtual Plant and Model Predictive Control Applications, Research Triangle Park, NC: ISA, 2003. Naidu, D. S., Optimal Control Systems, Boca Raton, FL: CRC Press, 2003. Nelles, O., Nonlinear System Identification: From Classical Approaches to Neural Networks and Fuzzy Models, London: Springer-Verlag, 2000. Omatu, S., Khalid, M., and Yusof, R., Neuro-Control and its Applications, London: Springer-Verlag, 1996. Ou, J., and Rhinehart, R. R., “Grouped Neural Network Model Predictive Control,” Control Engineering Practice, Vol. 11, No. 7, 2003, pp. 723–732.

© 2006 by Béla Lipták

455

Qin, S. J., and Badgwell, T. A., “An Overview of Industrial Model Predictive Control Technology,” in Chemical Process Control – V, pages 232–256, Fifth International Conference on Chemical Process Control, AIChE and CACHE, 1997. Rossiter, J. A., Model-Based Predictive Control, Boca Raton, FL: CRC Press, 2003. Rothstein, M. B., “Tuning a Nearly Optimal Control by Digital Simulation,” Instrumentation Technology, February 1972. Rovira, A. A., Murrill, P. W., and Smith, C. L., “Tuning Controllers for SetPoint Changes,” Instruments and Control Systems, December 1969. Ruel, M., “Loop Optimization: How to Tune a Loop,” Control Magazine, Vol. 12, No. 5, May 1999, pp. 83–86. Sastry, S., and Bodson, M., Adaptive Control: Stability, Convergence and Robustness, Englewood Cliffs, NJ: Prentice Hall, 1989. Seborg, D. E., Edgar, T. F., and Shah, S. L., “Adaptive Control Strategies for Process Control,” Chemical Engineering Journal, 32, 1986, pp. 881–913. Seborg, D. E., Edgar, T. F., and Mellichamp, D. A., Process Dynamics and Control, 2nd ed., New York: John Wiley & Sons, 2003. Shinskey, F. G., Feedback Controllers for the Process Industries, New York: McGraw Hill, 1994. Shinskey, F. G., Process Control Systems, 4th ed., New York: McGraw Hill, 1996. Smith, C. A., and Corripio, A. B., Principles and Practice of Automatic Process Control, New York: John Wiley & Sons, 1985. Smith, O. J. M., “Close Control of Loops with Dead Time,” Chemical Engineering Progress, May 1957. Sood, M., “Tuning Proportional-Integral Controllers for Random Load Changes,” M.S. Thesis, Dept. of Chemical Engineering, University of Mississippi, 1972. Sood, M., and Huddleston, H. T., “Optimal Control Settings for Random Disturbances,” Instrumentation of Technology, March 1973. Suykens, J. A. K., Vandewalle, J. P. L., and De Moor, B. L. R., Artificial Neural Networks for Modelling and Control of Non-Linear Systems, New York: Kluwer Academic, 1996. Thayer, D., “Tuning Direct Digital Control,” Instruments and Control Systems, October 1967. VanDoren, V., Techniques of Adaptive Control, Burlington, MA: Butterworth-Heinemann, 2003. Wellstead, P. E., and Zarrop, M. B., Self-Tuning Systems, New York: Wiley, 1991. Wojsznis, W., Gudaz, J., Mehta, A., and Blevins, T., “Practical Approach to Tuning MPC,” ISA Transactions, Vol. 42, No. 1, January 2003, pp. 149–162. Wood, R. K., “Computerized Design and Analysis of Control Loops,” Instrumentation Technology, 1971.