Decrease the PID Gain and Reset Time for Pipeline and Inline Control

The following tip is from the ISA book by Greg McMillan and Hunter Vegas titled 101 Tips for a Successful Automation Career, inspired by the ISA Mentor Program. This is Tip #78.

101 Tips for a Successful Automation CareerI quickly started to appreciate pipeline and inline control from working on inline pH control where a static mixer (a baffled piece of pipe) was the only source of mixing. Static mixer manufacturers touted radial mixing, but no back mixing, that is, there was only plug flow. No back mixing meant that the components at the static mixer inlet all arrived at the outlet at the same time. All of the components spent the same time residing in the static mixer. This tight residence time distribution is beneficial when static mixers are used as plug flow reactors to provide consistent conversion; reaction rate multiplied by residence time. However, for pH control that had an effectively instantaneous neutralization rate, I was looking for some back mixing so that a portion of the residence time (volume divided by volumetric flow rate) was a process time constant.

I appreciated one supplier proudly quantifying the lack of back mixing; it simplified my decision not to buy the advertised static mixer. The supplier’s sales engineers could not understand what I needed for pH control. This failure to have a meaningful dialog on the effects of equipment design on process control comes down to a lack of recognition that a process time constant is good and process deadtime is bad for tight process control.

Pipeline and inline control loops have little back mixing. The residence time is transportation delay and hence pure deadtime. The only time constants in the loop are from the sensor, transmitter damping, and signal filters. Flow, blend concentration, conveyor or feeder load, desuperheater temperature, extruder temperature, liquid and polymer pressure, jacket inlet temperature, sheet thickness, pulp stock consistency, and static mixer pH loops (among others) effectively have no process time constant. Heat exchangers, inline reactors, and fluidized bed reactors essentially have plug flow but have a series of process time constants in the temperature response from the thermal lags associated with the heat transfer surfaces.

The lack of process time constant means there is no filtering of process noise by the process and the process response is not gradual for a step change in a process input. Consequently, manual actions, sequential actions, on-off control, and oscillations in loops upstream are more problematic because there is no smoothing of response by a process time constant. If there are no appreciable measurement time constants, the proportional mode passes the abrupt change in the process on to the controller output. In contrast, the integral mode ramps the output, creating a gradual response that is missing in the processes.

Concept: Pipeline and inline control loops have no back mixing and hence no process time constant in the response of the outlet to changes in the inlet. What lags do exist are in the wrong place. For temperature control, thermal lags associated with heat transfer surfaces provide process time constants for changes in the cooling or heating fluid that unfortunately slow down corrections by the PID. Sensor lags and signal filters create the illusion of a process time constant. When the total loop deadtime is larger than the largest time constant in the loop, whether in the process or in the automation system, the process response is termed deadtime dominant. Such processes tend to be noisy and promote the transfer of oscillations downstream. Using less proportional action by applying a smaller controller gain and more integral action by applying a smaller reset time helps moderate the consequences of noise, interactions, nonlinearities, and short-term variability.

Details: For pipeline and inline loops, decrease the controller gain to be less than ¼ of the inverse of the open loop gain, and decrease the reset time toward a low limit of ½ of the total loop deadtime as the degree of deadtime dominance increases. See the July 2012 Control Talk blogs “The ABCs of Controller Tuning” and “Deadtime Dominance Does Not Have to Be Deadly” for more on these factors. Equation C-13 in Appendix C of Reference 1 shows how the integral time becomes ¼ of the ultimate period as the process time constant goes to zero. If you consider the ultimate period to be twice the deadtime for this case, the integral time ends up as ½ the deadtime. Add just enough signal filtering to ensure that the fluctuations in the controller output from noise are less than the final control element deadband.

Watch-outs: Fouled electrodes, sensors that fit loosely in thermowells, and ceramic protection tubes in furnaces will add a significant measurement time constant. If the measurement time constant or signal filter becomes much larger than the deadtime, you will only have the illusion of better control because you are seeing an attenuated version of the process changes. Too large of a measurement time constant shows up on a trend chart as a decrease in amplitude and an increase in the period of the oscillations. Oscillations are more prevalent and persistent. Slow final control elements (slow variable speed drives, dampers, or control valves) can give the appearance of a time constant in the manipulated flow.

Exceptions: Temperature control via manipulation of cooling and heating in pipeline and inline equipment will have process time constants from thermal lags that may approach or exceed the process deadtime.

Insight: Inline and pipeline control loops where inlet flows are manipulated have a deadtime dominant response unless the automation system introduces a time constant in the measurement or final control element.

Rule of Thumb: Decrease the PID controller gain and reset time for loops where the deadtime is much larger than the time constant.

Print Friendly

, , , ,