FIGURE 2: BREAKING THE PATH
The batch switch closes the reset-feedback loop when output is below 100%.
Figure 3 shows how an unprotected (no-batch) PID controller can cause temperature overshoot during startup of an inert batch when the integral term is saturated at 100%. Derivative acting on the controlled variable causes the output to leave its limit before set point is crossed, but not soon enough to avoid overshoot. In an exothermic reaction, the overshoot would be even more severe, and could result in product loss. The controller protected by a batch switch is preloaded near the anticipated load of 50%, where split-range heating and cooling valves are both closed, thereby avoiding the overshoot. The controller’s mode settings can be tuned for optimum load rejection, and the preload adjusted to produce the desired approach to set point on startup. Too low a setting results in undershoot, and too high a setting causes overshoot. A preload setting of 100% would produce the same overshoot as the no-batch controller in Figure 3.
FIGURE 3: AVOIDING OVERSHOOT
Proper preloading of the batch switch optimizes the approach to set point.
While the loop is open, if the deviation happens to lie within the proportional band of the controller, its output may fall below the switch setting after transfer to preload. Then, the switch will reverse position, and integration will resume, driving the output back to the switch setting. The switch will then cycle between its two positions. In a digital controller, this reversal could happen at every scan interval. The ultimate effect is essentially to control the controller’s output at the switch setting, with the integral term “b” settling at an intermediate position representative of the current process load. Any upset reducing the deviation will then be countered by an immediate change in controller output away from the limit. This behavior is common in anti-surge compressor control, where suction flow is normally well above the surge set point and the recycle valve is held closed by the flow controller. A sudden loss in load will reduce flow, and the controller must start opening the recycle valve before the set point is reached.
The batch switch applies equally well when the manipulated variable encounters a lower limit. The switch then transfers to preload when the controller output falls below that lower limit.
Override control was recently described recently in H. L. Wade’s “Under the Hood of Override Control, Part 1 and Part 2,” CONTROL, Dec. ’05 and Jan. ’06. In these systems, two or more controllers compete for the same manipulated variable, which is selected based on whichever has the lower or higher output. Since only one controller can be selected at any given time, the remaining loops are open, and their unselected controllers will wind up unless protected.
Figure 4 below shows the most effective method for avoiding windup in the unselected controllers: the selected output is the common feedback signal to all of the controllers in the system. Simply stopping integration in the unselected controllers isn’t enough because it leaves their integral term loaded with a constant that loses its relationship to process load over time.
FIGURE 4: PREVENTING WINDUP
The selected output is fed back to the integral term of all controllers.
The selected output, however, represents current process load, and therefore keeps the unselected controllers current while their loops are open. Only the selected controller sees its own output fed back, and so it alone integrates. Because all controllers are biased at the same level, transferring control from one to another will tend to take place when both are at zero deviation, and it therefore becomes a smooth event.
The integral lag plays an important role in this transfer. Other methods of windup protection tend to be too abrupt because they include forcing all controller output limits to track the selected output or periodically initializing the unselected controllers. Any noise on the selected output is easily rectified into a bias with these methods, potentially shifting the transfer point on the unselected controllers, or causing offset in the selected controller. The integral lag avoids this problem by effectively filtering noise.
Access to the reset-feedback signal allows insertion of deadtime compensation as shown in Figure 5. Delaying integration in this way improves the controller’s performance, giving it the capability of a Smith predictor. The same behavior can’t be attained using other means of integration. However, the controller isn’t as limited in robustness as the Smith, and can be applied to all processes. The PI controller in Figure 5 has a performance close to a PID, without the derivative’s sensitivity to noise. However, with the addition of derivative action, the PID controller has the highest available performance of all, both in load rejection and set-point response, and doesn’t even require a batch switch for process startup.