I had an awakening as to the much greater than realized disconnect between what is said in the literature and courses and what we need to know as practitioners as I was giving guest lectures and labs to chemical engineering students on PID control. We are increasingly messed up. The disparity between theory and practice is exponentially growing because of leaders in process control leaving the stage and users today not given the time to explore and innovate and the freedom to publish. Much of what is out there is a distraction at best. I decided to make a decisive pitch not holding back for sake of diplomacy. Here is the start of a point blank decisive comprehensive list in a six part series.
Please read, think and take to heart the opportunities to increase the performance and recognized value of our profession. The list is necessarily concise in detail. If you want more information on these opportunities, please join the ISA Mentor Program and ask the questions whose answers can be shared via Mentor Q&A Posts.
-
Recognizing and addressing actual load disturbance location. Most of the literature unfortunately shows disturbances entering the process output when in reality disturbances enter mostly as process inputs (e.g., feed flow, composition and temperature changes) passing through the primary process time constant. Thinking of disturbances on the process output leads to many wrong conclusions and mistakes, such as large primary time constants are bad, tuning can be done primarily for setpoint changes, feedforward and ratio control is not important, and algorithms like Internal Model Control are good alternatives to PID control.
-
Tuning and tests to first achieve good load disturbance rejection and then good setpoint response. While most of the literature focuses on setpoint response tuning and testing, the first objective should be good load disturbance rejection particularly in chemical processes. Such tuning generally requires more aggressive proportional action. Testing is simply done by momentarily putting the PID in manual, changing the PID output and putting the PID back in auto. Tuning should minimize peak and integrated error from load disturbances taking into account needs to minimize resonance. To prevent overshoot in the setpoint response, a setpoint lead-lag can be used with lag time equal to reset time or a PID structure of proportional and derivative action on PV and integral action on error (PD on PV and I on E) can be used. If a faster setpoint response is needed, setpoint lead can be increased to ¼ lag time or a 2 Degrees of Freedom (2DOF) PID structure used with setpoint weight factors for the proportional and derivative modes equal to 0.5 and 0.25, respectively. Rapid changes in signals to valves or secondary loops upsetting other loops from higher PID gain setting can be smoothed by setpoint rate limits on analog output blocks and secondary PIDs and turning on external-reset feedback (ERF). We will note the many other advantages of ERF and its facilitation of directional move suppression to intelligently slow down changes of manipulated flows in a disruptive direction in subsequent months (hope you can wait). In Model Predictive Control move suppression plays a key role. Here we can enable it with additional intelligence of direction without retuning PID.
-
Minimum possible peak error is proportional to dead time and actual peak error is inversely proportional to PID gain. Peak error is important to prevent relief, alarm and SIS activation and environmental violation. The ultimate limit to what you can achieve in minimizing peak error is proportional to the total loop dead time. The practical limit as to what you actually achieve is inversely proportional to the product of the PID gain and open loop process gain. The maximum PID gain is inversely proportional to the total loop dead time. These relationships hold best for near-integrating, true integrating and runaway processes.
-
Minimum possible integrated error is proportional to dead time squared and actual peak error is proportional to reset time and inversely proportional to PID gain. The integrated absolute error is the most common criteria sited in literature. It does provide a measure of the amount of process material that is off-spec. The ultimate limit to what you can achieve in minimizing integrated error is proportional to the total loop dead time squared. The practical limit as to what you actually achieve is proportional to reset time and inversely proportional to the product of the PID gain and open loop process gain. The minimum reset time is proportional and the maximum PID gain is inversely proportional to the total loop dead time. These relationships hold best for near-integrating, true integrating and runaway processes.
-
Detuning a PID can be evaluated as an increase in implied dead time. The relationships cited in items 3 and 4 above can be understood by realizing that a larger than actual total loop dead time is the effect on loop performance of a smaller PID gain and larger reset time setting than needed to prevent oscillations. This implied dead time is basically ½ and ¼ the summation of Lambda plus the actual dead time, for self-regulating and integrating processes, respectively.
-
The effect of analyzer cycle time and wireless update rate depends on implied dead time and consequently tuning. You can prove almost any point you want to make about whether the effect of a discontinuous update is important or not by how you tune the PID. The dead time from an analyzer cycle time is 1½ times the cycle time. The dead time from a wireless device update or PID execution rate or sample rate is ½ the time interval between updates assuming no latency. How important this additional dead time is seen in how big it is relative to the implied dead time. The conventional rule of thumb is that the dead time from discontinuous updates should be less than 10% of the total loop dead time (wireless update rates and PID execution rates less than 20% of dead time). This is only really true if you are pursing aggressive control where the implied dead time is near the actual dead time. A better recommendation would be a wireless update rate or PID execution rate less than 20% of “original” implied dead time. I use the work “original” to remind us not to spiral into slowing down update and execution rates by increasing implied dead time and then further slowing down update and execution rates.
-
The product of the PID gain and reset time must be greater than the inverse of the integrating process gain. Violation of this rule cause very large and very slow oscillations that are slightly damped taking hours to days to die out for vessels and columns, respectively. This is a common problem because in control theory courses we learned that high controller gain causes oscillations and the actual PID gain permitted for near integrating, true integrating and runaway processes is quite large (e.g., > 100). Most don’t think such a high PID gain is possible and don’t like sudden large movements in valves. Furthermore, integral action provides the gradual action that will always be in a direction consistent with error sign and will seek to exactly match up PV and SP meeting common expectations. The result is a reset time frequently set that is orders of magnitude too small making the product of PID gain and reset time less than the inverse of the integrating process gain causing confusing slow oscillations.
-
The effective rate time should be less than ¼ the effective reset time. While PID controllers with a Series Form effectively prevented this due to interaction factors in the time domain, this is not the case for the other PID Forms. Not enforcing this limit is a common problem in migration projects since older controllers had the Series Form and most modern controllers use the ISA Standard Form. The result is erratic fast oscillations.
-
Automation system dynamics affect the performance of most loops. This should be good news for us since this is much more under the control of the automation engineer and easier and cheaper to fix than process or equipment dynamics. Flow, pressure, inline temperature and composition (e.g., static mixer), and fluidized bed reactors are affected by sensor response time and final control element (e.g., valve and VFD) response time. Pressure and surge control loops are also affected by PID execution rate.
-
Reserve feedforward multiplier and ratio controller ratio correction for sheet lines and plug flow systems. The conventional rule that on a plot of manipulated variable versus feedforward variable, a change in slope demands a feedforward multiplier and a change in intercept demands a feedforward summer is not really relevant. A feedforward multiplier introduces a change in controller gain that is counteracts the change in process gain. However, this is only useful for sheet lines and plug flow (e.g., static mixers and extruders) because for vessels and columns, the effect of back mixing from agitation and reflux or recirculation creates a process time constant that is proportional to the residence time. For decreases in feed flow the increase in process time constant from an increase in residence time negates the increase in process gain. Also, the most important error is often a bias error in the measurements. Span errors are smitten by a large span showing up mostly as a change in process gain much less than the other sources of changes in process gain. Also, the scaling and filtering of a feedforward summer signal and its correction is much easier.