Professional Documents
Culture Documents
TI 039-200
LOCATE 2 VERIFY 2
ERROR
TIME
EXACT tuning provides an innovative means of controlling a process more efficiently than conventional methods of control. It also provides a means of managing processes that are otherwise difficult to control.
Introduction
EXACT (EXpert Adaptive Controller Tuning) tuning with its pretune and selftune algorithms:
Allows faster process startup. Optimizes controller tuning in the presence of noise, variable dynamics, process nonlinearities, deadtime, setpoint changes, and load variations. Maintains tuning to current operating conditions. Gives skilled personnel the freedom to do other tasks. Reduces loop operating expenses through more efficient control.
E1
E1 Load Change
Error
Error
+ -
+ -
Figure 1. Pattern Recognition Characteristics The development of EXACT tuning is a distillation of the many knowledge-based rules based on the observations of tuning experts. The fundamental rules were selected and tested to assure general applicability. Each step in the process was broken down into its basic components, and the whole was reassembled and programmed into the memory of the controller. The results are EXACT tuning, which permits the controller to use current process data to continually calculate and implement optimum tuning parameters. With EXACT tuning, the process is checked five times a second for the need to either institute or change the corrective action. Thus, the process is always controlled for maximum efficiency.
Pattern Recognition
The pattern to be recognized is error versus time. The distinctive features are the presence or absence of peaks, the time between peaks, and the steady-state error. The information is represented using the terms overshoot, damping, and period (see Figure 1). The EXACT pattern recognition approach is unique. It uses direct-performance feedback (by monitoring the process variable) to determine the action required. Also, its algorithm does not need a mathematical model of the process.1, 2, 3 The selftune PID algorithm monitors the closed-loop recovery following a disturbance to set point or load. It automatically calculates P, I, and D to minimize process recovery time, subject to user-specified damping and overshoot constraints. For most processes, damping and overshoot are not independent. The period of the loop must be included to define the shape. The period is nondimensionalized by using the controller I and D to produce ratios similar to those proposed by Ziegler-Nichols 4 and Shinskey 5. The ratios I/period and D/period define the lag and lead angles of the controller. The result is that P, I, and D values are obtained based upon the period of oscillation, and constrained by the allowed damping and overshoot parameters (see Figure 5).
1. Rohrs, C. E., Valavani, L., Athans, M., and Stein, G., Robustness of Adaptive Control Algorithms in the Presence of Unmodeled Dynamics, MIT Industrial Liaison Program, Publication No. 01-016, 1983. 2. Fjeld, M. and Wilhelm, R. G., Jr., Self-Tuning Regulators - The Software Way, Control Engineering, October 1981, P. 99. 3. Clarke, D. W., The Application of Self-Tuning Control, Trans Inst MC Vol. 5. No. 2, April-June 1983, P. 59. 4. Ziegler, J. G. and Nichols, N. B., Optimum Settings for Automatic Controllers, Trans ASME, November 1942. 5. Shinskey, F. G., Process Control Systems, McGraw-Hill, NY, NY, 2nd Edition, 1979, pp. 96-99.
Locate 3
Verify 2
Locate 2
Verify 1
Locate 1
Quiet
Settle
Figure 2. Self-Tuning Algorithm State Diagram Up to this point, the controller is operating as a fixed parameter PID controller, but it has observed the loop response to a disturbance. The algorithm uses the response information to calculate new values of P, I, and D (during ADAPT). The first step in calculating new P, I, and D values is to use the period information to directly set I and D. The interaction between P, I, and D forces this algorithm to be slightly more sophisticated: P must be adjusted to compensate for the I and D changes. Ziegler-Nichols developed a tuning procedure that involved adjusting I and D until I/period equals 0.5 and D/period equals 0.12. It has been found that much better tuning and quicker convergence result when the desired ratios of I/period and D/period are changed by the algorithm. A dominant deadtime process needs smaller ratio values, and a dominant lag process uses larger ratio values. If the response is overdamped and distinct peaks are not found, I and D are adjusted using the damping and overshoot parameters. Before searching for the Peak 2, the algorithm verifies that the first is a true peak (during VERIFY 1). If a new extreme value occurs during this verification, it then becomes the first peak and the timer is reinitialized. Once Peak 1 occurs and is verified, the algorithm uses the same method to locate and verify Peaks 2 and 3. The peak information is then expressed in the previously defined terms of overshoot, damping, I/period, and D/period. Validity is determined for this information based upon both the peak heights relative to the nominal noise band, and the peak times as compared with the times of a nominal damped sinusoid.
4
The final step in computing P, I, and D compares the observed damping and overshoot to the maximum allowed values set by the user. If distinct peaks have occurred and both damping and overshoot are less than the maximum allowed values, P is decreased. The amount of decrease depends upon the difference between the maximum allowed and actual value of either the damping or the overshoot parameter. Since damping and overshoot are not independent, the algorithm uses the difference that has the smallest value. If distinct peaks are not detected, P, I, and D are all decreased by an amount that depends upon the maximum allowed damping or overshoot. Since it is very common to find the control loop response overdamped without three peaks, the algorithm must also deal with this situation. Overdamping is handled by first recognizing the response as overdamped, then aborting the search for Peak 2 or 3 and assigning pseudo peak values instead. The pseudo peaks allow new tuning parameters to be calculated using the same procedure as would be used if three peaks were located. When the loop is properly tuned, the values of P, I, and D are essentially unchanged from disturbance to disturbance. However, if the disturbance changes shape or the process changes, EXACT will determine new values. The self-tuning cycle is complete when the newly calculated P, I, and D values are implemented into the controller. The algorithm then goes through a settling state that allows a smooth transition into locating a new first peak, if necessary. The SETTLE state is only used to assure that the next peak found is a true peak. However, switching the controller from MANUAL to AUTO or initially activating the self-tuning feature forces the algorithm into the SETTLE state.
User-Adjustable Parameters
Any, or all, of the user-adjustable parameters can remain at the factory-set values, can be changed to new values by the user, or (in the case of the six key parameters) can have their values optimized by the pretune function.
ERROR
WMAX TIME
ERROR
TIME
ERROR
TIME
Process Sensitivity
Measurement
Figure 6. Typical Response to Step Change in Output The four main phases of pretune are shown in Figure 7. As they occur, they are identified in the display of the controller. The process upset occurs with a step change in controller output at point 1. The algorithm waits for steady state during 2 (the messages PTUN = SMALL 1 and PTUN = WAIT 2 will be displayed), calculates the control parameters, and returns the controller output to its starting value at point 3 (PTUN = PID 3). If the process is an integrating type or has high gain, point 3 is reached when the measurement changes by 10% of its span or the bump size, whichever is larger. Finally, the noise band and derivative factor are calculated during 4 (PTUN = NB 4). When the process is completed, the message PTUN = FINISH will appear.
Measurement
3 2 1 4
Time
10
11
Foxboro and EXACT are registered trademarks of The Foxboro Company. Copyright 1985 - 1995 by The Foxboro Company All rights reserved
MB 015
Printed in U.S.A.
0895