What is the key PID feature for Basic and Advanced Control?

April 26, 2012
The value of a key PID feature is increasingly becoming apparent. I got a preview of the importance when I found the 1980s vintage DCS required a fix for override control that was inherently addressed in the next generation of DCS by this feature. This wakeup call was followed by the realization this feature offered improved analyzer, blend, cascade, decoupling, feedforward, surge, valve position, and wireless control with a side benefit of simplified deadtime compensation.

The value of a key PID feature is increasingly becoming apparent. I got a preview of the importance when I found the 1980s vintage DCS required a fix for override control that was inherently addressed in the next generation of DCS by this feature. This wakeup call was followed by the realization this feature offered improved analyzer, blend, cascade, decoupling, feedforward, surge, valve position, and wireless control with a side benefit of simplified deadtime compensation.

Override controllers in a 1980s vintage DCS could walk off to an output limit when not selected despite the use of proper mode and configuration. The fix was to insert a filter in the path of each integral track signal where the filter time was equal to the reset time of the respective overridden PID. The key feature of the next generation of DCS provided this filter inherently in an external-reset signal.

The main concern with wireless control is that the PID output could continue to ramp from integral action during the time between updates or during communication interruptions. While wireless communication reliability has largely been addressed, the output from the integral mode can be ramping based on old information since the wireless transmitter default update rate and trigger level are set to increase battery life. The key PID feature available today allows the enhancement for wireless where the exponential response of the external-reset signal is computed only when there is a measurement update. The computation readily replaces the filter using the filter time setting. The external-reset signal can be the PV of what is being manipulated (e.g. secondary loop setpoint, speed, damper or valve position) so that a communication problem or a frozen reading or position will not cause the PID output to ramp. Sticking dampers and valves plugged impulse lines, and coated electrodes will simply suspend integral action.The first "aha" moment occurred when I realized this same feature could prevent integral action from ramping between updates from analyzers that can be quite long because of sample, cycle, and multiplex times. Furthermore, the extremely variable update times from lab analyzers would not be as much of a problem.  For more information and ramifications see April 7, 2010 demo-seminar "Deminar#1 - PID Enhancement for Sampled Measurements", the Control May 2005 article "Is Wireless Process Control Ready for Prime Time", and the InTech 2010 July-Aug article "Wireless - Overcoming the challenges of PID control and analyzer applications." This more intelligent limitation of integral action leads us to the principle benefit for cascade control.

Controllers can readily be tuned to provide a response faster than the secondary loop or final control element can respond. The problem is insidious in that the burst of oscillations only occurs for large setpoint changes or large disturbances when the change in the loop output is faster than the secondary or final control element (e.g. control valve, damper and VFD) can respond as described in the Control May 2005 article "The power of external-reset feedback." The key PID feature can inherently prevent the problem if the actual process variable of the secondary loop or final control element is used for the external-reset signal. The update must be faster than the execution time of the PID. Consequently wireless updates of valve position and VFD speed is presently not fast enough except when the trigger level is small enough and the response of huge pneumatic actuators or VFD rate limiting is slow compared to the default update rate. For more details see the May 12, 2010 demo-seminar "Deminar#3 - PID Control of Slow Valves and Secondary Loops"

The second "aha" moment occurred when I realized that suspension integral action when there is no update, would automatically kill the limit cycle from valve stiction in all loops and from deadband in integrating processes and in cascade loops as seen in my April 21, 2010 demo-seminar "Deminar#2 - PID Control of Valve Sticktion and Backlash" Then second "aha" part b was understanding that this could prevent valve position control (VPC) used for optimization from chasing limit cycles or insignificant changes in the valve position being optimized as discussed in the Control 2011 Nov article "Don't over look PID for APC" which led to the third "aha" moment.

In optimization, we want slow approaches to the optimum and a fast get way from any potential constraint such as a valve approaching an output limit. The PID feature allows the user to set directional velocity limits on the setpoint in an analog output (AO) or PID without regard to tuning of the controller driving this output or setpoint.

The fourth "aha" came when I recognized that these directional velocity limits were better than the move suppression in an MPC because the size of the change did not change with execution time and could be specified differently for up and down.  Now I could limit the transfer of variability from the controlled variable to the output variable and limit interaction without retuning by simply setting directional velocity limits, further extending the capability of VPC for simple optimizations. The intelligent suspension or limitation of integral action turns VPC from being a slow integral-only controller to a much more easily tuned and responsive PID with inherent decoupling.

The fifth "aha" occurred when I remember that some suppliers of compressor surge control systems put quick exhaust valves on surge valves with restrictors on supply to provide fast opening and slow closing. I objected to this crude solution because the discontinuous and extreme action of the quick exhaust valves and unknown rate of change from the restriction caused severe tuning problems and erratic control upsetting downstream users of the compressed gas. I favored boosters on the positioner output but setting speed was still problematic. With the key PID feature, directional velocity limits in the analog output block can exactly provide the speed of opening versus closing within the limits of the surge valve capability without retuning.

The sixth "aha" came while thinking about feedforward control. The key feature could ignore dynamic compensation errors as noted in the Jan 5, 2011 demo-seminar "Deminar#11 - Feedforward Control"  Additionally, directional velocity limits on the setpoints of flow loops being driven by ratio control (flow feedforward) could be set to provide simultaneous changes in addition rates without requiring the retuning of the primary process controller for temperature or composition. Consistent in-sync timing of feeds is particularly important for blending operations.

The icing on the cake is a much simpler implementation of deadtime compensation afforded by the key feature. All deadtime compensators can be reduced to the Smith Predictor either in form or effectiveness. Unfortunately with the Smith Predictor, there are 3 adjustments (process gain, process time constant, and process gain) that are difficult to get exactly due to nonlinearity and unknowns. Also, the PID PV is no longer the actual process PV but a predicted PV free of deadtime so special efforts are needed to restore a faceplate display for the operator. The key PID feature provides deadtime compensation by the simple insertion of a standard deadtime (DT) function block in the external-reset path between the AO BKCAL_OUT and the PID BKCAL_IN. The DT parameter is set equal to a slight underestimate of the total loop deadtime as discussed in Oct 13, 2010 demo-seminar "Deminar#10 - PID Deadtime Compensation" and in the March 25 2009 post of "Advanced Application Note 003- Compensation of Dead Time in PID Controllers"

The key feature is the positive feedback implementation of integral action where a filter output is added to the contribution of the proportional mode. The filter input is the PID output or external-reset feedback signal when the external-reset limiting action (dynamic reset limit) is enabled. The positive addition of the controller output to the output from the proportional mode is positive feedback. When the PID error is zero, the output of the proportional mode is zero and the output of the filter becomes equals the output of the PID suspending integral action. Of course this simple explanation does not account for structures of P on PV rather than on error. The external-reset signal can be the PV of anything being potentially driven by the PID output providing the inherent intelligent suspension of integral action when nothing is changing. Most text books show literally an integrator of error for the integral mode. I expect most users don't realize the existence and importance of the positive feedback implementation of the integral mode. A simplified depiction of the positive feedback implementation for the traditional PID and enhanced PID is shown on slide 122 of my April 18 2012 short course ISA-Edmonton-2012-Effective-Use-of-Measurements-Valves-PID-Controllers-Rev1.pdf

Intelligent integral action is particularly important because "reset is never satisfied, has no sense of direction, and is continually ramping" as exemplified in the case where the human answer is wrong on slide 100 of my ISA Edmonton short course.  Maybe we don't realize the consequences because traditional integral action is mimicking human limitations, such as impatience and lack of understanding of delay, interruption, and speed. I have a seventh "aha" where the key feature enables reset time adaptation to prevent overshoot and undershoot based on a PV projected about one deadtime in the future. Unlike a deadtime compensator, the consequences of an incorrect deadtime are minimal. The adaptation also prevents the reset time from being set too small for integrating processes and runaway processes by estimating integrating process gain and using the inequality relationship for non self-regulating processes that establishes the lower limit to the gain and reset time tuning settings.

About the Author

Greg McMillan | Columnist

Greg K. McMillan captures the wisdom of talented leaders in process control and adds his perspective based on more than 50 years of experience, cartoons by Ted Williams and Top 10 lists.