What are the relative merits of different PID structures, a setpoint (SP) filter, and analog output (AO) setpoint rate (velocity) limits? Should I seek a general solution I can use all the time and each knob fits a particular purpose, or a controller with fewer knobs that does exactly what I want for a particular application so there is less to "break"?
This is the 2nd in a continuing series of practical and useful questions on PID tuning raised by Brian Hrankowsky a knowledgeable process control specialist in the pharmaceutical industry after having been present at a panel session on controller tuning at ISA Automation Week 2010 developed by Michel Ruel. Brian is not representing his company in these posts. The questions and answers in all of these blogs will be addressed in much greater detail in the long overdue 4th edition of my book Tuning and Control Loop Performance being published by Momentum Press.
I am in favor of fewer knobs for the average user for most of the applications but want the flexibility to be creative if I have the motivation and the time. I have been frustrated over the years by parameters being hidden from the user. Reducing what the user can do can lead to downgrading of the need for user expertise and disappearance of specialists in core groups and at the plant. Parameter filters and various levels of view for various levels of expertise can be an effective approach to screen out excessive detail but does not prevent somebody changing a parameter or option that should be left alone. Maybe there should be a password for a specialist and expert view. Possibly the best solution is this restricted but not totally denied access and a generalized methodology offering a reduction in parameters needed for most applications.
Brian has found some advantage to reducing proportional action on fast loops and has looked at an Integral on error, Derivative on PV, and no Proportional (I-error D-PV) controller as a solution. On fast loops he has noticed that proportional action has caused excessive movement of the controller output. This movement can wear out valves and upset other loops. The rate time for these loops would be set to zero to provide integral only control. Most tuning methods will automatically reduce proportional gain and eliminate derivative for deadtime dominant loops. Some of these methods also decrease the reset time as the ratio of loop deadtime to time constant increases (dominance increases) to further accentuate integral action.
I have not investigated the I-error D-PV or the 2 degrees of freedom (2DOF) controllers enough to know if there can be some generalization as to settings or how these structures perform for a spectrum of objectives. I have seen where the setting of beta (setpoint weight for derivative) and gamma (setpoint weight for proportional) equal to 0.5 for a 2DOF provided a generally useful setpoint response. My initial reaction to the I-error D-PV controller is a concern that the lack of proportional action will increase the peak error for fast disturbances (liquid and gas pressure) and rise time for setpoint changes where batch cycle or continuous startup or transition time is important. Also, highly exothermic reactors that depend upon a high controller gain to prevent a runaway is a safety issue. An important question is whether new tuning rules can be developed for rate action to functionally replace the missing proportional action.
For biological processes, the disturbances are so slow and the batch cycle times so long, peak errors and rise times are not nearly as important in chemical processes. However, the product of the controller gain and integral time needs to be greater than the twice the inverse of the integrating process gain for integrating processes (e.g. batch temperature, gas pressure, and level) as shown on the bottom of slide 21 in the Emerson 2012 Exchange presentation "Effective Use of PID Features for Control and Optimization."
The presentation also describes some typical applications of various structures such as proportional plus derivative temperature control for a batch operation where temperature could only increase (only heating - no cooling). Integral action causes excessive and sustained overshoot for a one direction response common in batch operations unless the slope of the batch profile is the PV as described in the July 2008 Control article "Unlocking the Secret Profiles of Batch Reactors". The ease and effectiveness of various solutions certainly looks like fertile ground for personal investigation.
I suggest people use a virtual plant to investigate the relative merits of structures and tuning methods for various objectives. Note that process objectives affecting efficiency and production rate may be quite different than typical loop objectives for peak and integrated error. A virtual plant makes investigation easier and more productive. I will be making a Process Control Lab virtual plant available 11-30-2012 on the Emerson Exchange 365 site. I will also be posting ideas on how to use the labs on my ISA Interchange site. Tips #71 and #72 will be posted on this site describing a general tuning methodology for different objectives. The Process Control Labs have a simple generic operator interface including standard PID faceplate so that you do not need to get into a configuration or any special system specific interface. The labs allow you to change PID structures and options and add feedforward, bang-bang logic, and characterization on the fly (without a download), get metrics on load disturbances and setpoint changes, and set an incredibly wide spectrum of control valve, process, and measurement dynamics.
A generalized methodology for me would be to use a structure of PI on error with D on PV, a setpoint filter set equal to the reset time, and the external-reset option (ERO) (e.g. dynamic reset limit) enabled if the user has the positive feedback implementation of integral action as shown on slide 15 in "Effective Use of PID ...". The ERO feedback signal (e.g. BKCAL_IN and BKCAL_OUT) must be properly configured to provide the PV of the AO going to the final control element and the PV of the secondary loop for a primary loop in cascade control. For signal characterizers and splitters on the PID output, the configuration to maintain the continuity of the ERO feedback signal requires some best practices that are important regardless of ERO. For a final control element that is slow or doesn't always respond (e.g., variable frequency drive deadband and resolution limit or control valve backlash and stiction), a fast readback and assignment to the AO PV of the actual response of the final control element (e.g. pump speed or control valve actual position) is needed to get the full benefit of ERO. The readback update must fast enough to show current PV of the final control element.
If a faster setpoint response is wanted you can replace the setpoint filter with a setpoint lead-lag where the lead is set to be about 20% of the lag time that is the controller reset time. If you want the fastest possible setpoint response you can use bang-bang logic as demonstrated in the Process Control Lab and described in the May 2006 Control article "Full throttle batch and startup response"
The addition of analog output (AO) setpoint rate limits provides move suppression to meet process objectives in terms of absorbing rather than amplifying variability and to provide coordination and decoupling of loops. The classic case of the need to absorb variability is surge tank level control. Most of these level loops are increasing variability by overreaction from too large of gain or slow oscillations from too low of a reset time causing violation of the rule on slide 21 of "Effective Use of PID ...". For surge tank level control a rate limit or controller gain set to use the entire level range can be computed. The rate limit provides the move suppression proven over decades to be so effective in model predictive control (MPC). Move suppression or "penalty on move" is the principal tuning MPC knob to set the transfer of variability from the process variable to the manipulated variable. The AO setpoint rate limits can be set differently for an increasing versus a decreasing setpoint providing more flexible directional move suppression. For cascade control secondary PID setpoint rate limits are used.
This generalized methodology of a structure of PI on error and D on PV with a setpoint lead-lag where the lag is set equal to the reset time and lead is set to reduce rise time (e.g. lead = 20% of lag), and ERO enabled with PV readback offers the following benefits (2DOF and ERO enabled with PV readback can accomplish the same results with more flexibility from setting beta and gamma):
- Minimization of rise time and overshoot for a setpoint change
- Minimization of peak error and integrated error for fast unmeasured disturbance
- Elimination of the need to set anti-reset windup (ARW) limits per last week's Control Talk Blog
- Prevention of overshoot and faltering for changes in dynamics or poor tuning by enabling an adaptive reset time per last week's Control Talk Blog
- Elimination of the oscillations from violation of the cascade rule (e.g. slow secondary loop, positioner, or VFD speed loop)
- Elimination of the oscillations from a limit cycle (e.g. deadband or resolution limit in VFD or backlash or stiction in control valve)
- Coordination of the timing of flows for blending or reaction without retuning the primary PID by the use of setpoint filters in the secondary loop to prevent a transient composition or stoichiometric unbalance for a change in production rate
- Prevent unnecessary crossing of the split range point by AO rate limit in the direction of crossing the split range point
- Prevent overaggressive optimization by a valve position controller via a AO or secondary PID setpoint rate limit in the direction of optimization
- Prevent overaggressive return to normal operation by a valve position controller via an AO or secondary PID setpoint rate limit in the direction of return (e.g. fast opening and slow closing surge control valve and fast recovery from RCRA pH limit and slow return to normal pH setpoint for inline pH)
- Suppression of interaction without retuning by adding AO or secondary PID setpoint rate limits in least important and most disruptive loop
- Absorption of variability (e.g. surge tank level control) by setting AO or secondary flow PID setpoint rate limits to use up the entire available range of tank level
The methodology also enables the elimination of oscillations from long and variable wireless and analyzer update times without retuning by extension of external-reset feedback to an enhanced PID that waits till there is a change in the PV. A traditional PID would develop severe oscillations due to integral action working during the excessive additional deadtime introduced by a large analysis time. Even offline analyzers with incredibly analysis update times of hours to days can be used for closed loop control without retuning with an enhanced PID. If the update time is much longer than the 63% process response time, the controller gain can be increased to be the inverse of the open loop gain (product of measurement, process, and final control element gains). See the InTech July/August 2010 article "Wireless - Overcoming challenges of PID control & analyzer applications" and the Sept 28, 2012 ISA Interchange post "Tip #100 - Use an Enhanced PID for At-line and Offline Analyzers"