Modeling and Control

Have you seen a loop more oscillatory in automatic than in manual? Is the problem loop interaction, aggressive controller tuning, or a sticky-sloppy valve? If you have a digital positioner with position readback on a throttle valve supplied by a control valve manufacturer, you have a chance of figuring out if the valve is the culprit. If you don’t have digital positioner and a real control valve, the valve probably is the problem.

For rotary valves, the position feedback measurement must be of stem rather than shaft position for the readback. Block or on-off valves provided by piping valve manufacturers often have position feedback on the actuator shaft. Even when the position of the stem is measured, the high friction of the rotary element (ball, plug, or disk) seal may cause shaft windup to the extent where even the stem position is not representative of the internal element position. I have seen this big time for a ball valve posing as a control valve in phosphorous service. Even for a new valve on the bench, the ball did not move for a 10% change in stem position. In these cases, the position feedback and readback are lying and the diagnostics from a smart positioner are misleading.

In the 1980s the slapping of piston actuators and positioners on piping valves became rampant. Since they were less expensive than a control valve, they were already in the piping spec, and the ISA standard for valve response testing had not been published, the process engineers were easy targets. After all, the piping valve worked well as a block and isolation valve, the leakage specs were impressive, and pluggage was often less of problem. Plus you could always blame process variability on mysterious causes since no one knew what the valve was really doing. Position readback was rare since it required a separate position transmitter and wiring. Valve tests by technicians in the field did not reveal a problem because 25% or 50% changes in signals were used. All but the biggest valves or dampers looked decent for such large changes in valve signals. The idea that enormous controller output changes (e.g. 25 to 50% per sec) did not occur except for special situations, such as surge control, didn't seem to cross most minds.

In the next two weeks we will get into the importance of readback in actual control valve scenarios. In the mean time if you want more information on the effect of control valves on loop performance, check out the following articles:

“Improve Process Loops”, Chemical Processing, October, 2007,

“A Fine Time to Break Away from Old Valve Problems”, Control, Nov, 2005

“What‘s Your Flow Control Valve Telling You?”, Control Design, May 2004

(1) A compressor shuts down
(a) The first out sequence indicates the compressor tripped on high speed
(b) A precipitous drop in suction flow followed by a rapid 1-2 second oscillation in suction flow preceded the speed deceleration from compressor shutdown
(c) The readback of actual surge butterfly position indicates the valve closed before the initial drop in flow.
(d) The surge set point flow controller set point was constant.
(e) The butterfly disc closed despite a controller output asking it to be open.
(f) The control valve is fail-open (inc-close) so a loss in signal or activation of the solenoid valve is not the cause
(g) Conclusion - the volume booster sensitivity and actuator size and type caused a butterfly disk instability at high flow

(2) A thermal oxidizer shuts down
(a) The first out sequence indicates the oxidizer tripped on high temperature
(b) A spike in natural gas flow occurred before the trip
(c) The natural gas set point was constant
(d) A readback of actual gas valve position indicates the gas valve position was relatively constant before the spike and started to closed after the spike
(e) The pressure transmitter upstream of the gas valve spiked high about the same time as the flow spike
(f) Conclusion - the natural gas pressure regulator upstream went open

(3) A pH tank has sustained nearly equal amplitude oscillations
(a) The pH oscillation amplitude stayed the same when the controller gain was increased or decreased*
(b) The pH amplitude changed for a different pH set point*
(c) A readback of actual valve position indicates the minimum change in valve position is 0.5% or alternately indicates a step in an actual valve position change always precedes the pH change.
(d) Conclusion - the oscillation is caused by the resolution limit (stick-lip) of the control valve which multiplied by the process gain is the amplitude of the pH limit cycle (a change in pH set point changes the process gain from the operating point nonlinearity associated with the titration curve)

(4) A column sump level has very slowly decaying oscillations
(a) The amplitude of the oscillation takes a day to decay
(b) Feed, steam, and reflux flows are relatively constant
(c) The oscillation is more persistent (decay and period slower) when the level controller gain is decreased
(d) Actual readback of level valve position matches the controller output within 0.05% a couple of seconds
(e) Conclusion - the controller gain is below the low gain limit (controller gain multiplied by controller reset time must be greater than 4 divided by the integrating process gain) A controller gain that is too high causes faster oscillations that would die out if the controller gain is decreased**

* - these controller tuning or set point changes provide affirmation but are not required to diagnose the problem

** - valve diagnostics confirm it is not a valve problem

(5) A column distillate receiver has a relatively constant amplitude level oscillation
(a) The amplitude of the oscillation is relatively constant for a given tuning
(b) The amplitude and period of the oscillation increases as the controller gain is decreased*
(c) Actual readback of level valve position indicates the minimum change in valve position routinely is 0.05% but the valve position stays at its previous position for a reversal of controller less than 0.5%
(d) Conclusion - the control valve has dead band which for an integrating loop causes a limit cycle

(6) A continuous evaporator has a constant amplitude temperature oscillation
(a) The amplitude of the oscillation stays the same when the controller gain is changed*
(b) Actual readback of steam valve position indicates the minimum change in valve position is 0.5%
(c) Conclusion - the steam valve has a resolution limit that for a self-regulating loop causes a limit cycle

(7) A process dead time varies
(a) The production rate is relatively constant
(b) The dead time increases as the controller gain is decreased *
(c) The dead time increases as the change in set point is decreased*
(d) Actual readback of level valve position indicates the minimum change in valve position routinely is 0.05% but the valve position stays at its previous position for a reversal of controller of less than 0.5%
(e) Conclusion - the changes in process dead time are caused by valve dead band (dead time is dead band divided by the rate of change of controller output for signal reversal)

(8) A process dead time varies
(a) The production rate is relatively constant
(b) The dead time increases as the controller gain is decreased *
(c) The dead time increases as the change in set point is decreased*
(d) Actual readback of level valve position indicates the minimum change in valve position routinely is 0.05% but the control valve takes from 5 to 50 seconds to catch up to the controller output for a reversal of controller (time to catch up increases as the change in controller output is decreased).
(e) Conclusion - the changes in process dead time are caused by a positioner with poor sensitivity that causes a slower exhaust or fill rate to actuator for smaller changes in controller output

* - these controller tuning or set point changes provide affirmation but are not required to diagnose the problem

** - valve diagnostics confirm it is not a valve problem

Process control is rich with mythology probably because what happens in the field is pretty remote from what was described in control text books. It is best summed up by a button given to me by my daughter 25 years ago that says “Reality Reeks.” Here are some myths that come to mind this Friday evening after updating a simulation library whose main threat is reality.

(1) Decreasing the scan time will improve control – for slow processes and older DCS with 12 bit A/D for I/O, the faster scan reduces the signal to noise ratio. This was particularly a problem for temperature loops that used thermocouple input cards with large spans. Often the noise from A/D chatter precluded the use of rate action even though these loops had significant second order time constants. The more prevalent reason a reduction in scan time may have no impact on control is the implied dead time from the use of current tuning practices as seen in the next myth. You can estimate how much dead time you can add before you see an increase in integrated absolute error for a load disturbance. Next week I will show the development of the equations that predict the implied dead time and the impact on peak and integrated error when the dead time added causes the total actual dead time to exceed the implied dead time. The dead time for a load upset from an unsynchronized scan time can be estimated to be on the average the latency plus one half of the scan time.

(2) Controllers are tuned for rapid set point response – controllers are tuned slower than what is shown in nearly every academic paper and book. This slower tuning creates an implied dead time that is greater than the actual dead time. Intuitively you can visualize this effect by considering as the tuning is slowed down more and more, the loop approaches manual control where the dead time for automatic corrective action is infinite. Whenever articles show the improvement from reduced dead time, the controller is retuned for best response to take advantage of the better dynamics.

(3) Unmeasured disturbances are a side issue – if there were no unmeasured disturbances, control would be a non issue because you could home in on the controller output that corresponds to the desired set point for a process variable. You would just need to run some data fitting algorithm one time and the loop would be set for the life of the process. In reality, there are always unmeasured disturbances.

(4) Disturbances enter directly into the measurement – in almost every process I have worked on the disturbance gets into the loop via a process input. For example, changes in raw materials to a reactor are feed inputs that go through the mixing and reaction process before they appear in the reactor temperature. If the upset enters downstream of the process, it is noise to me.

(5) Disturbances are step inputs – this is the case for almost every published analysis of control loops but in the real world, except for on-off control, there is nearly always a load disturbance time constant whether it is due to reset action in the culprit controller or the mixing time of a volume (even unagitated vessels have some degree of dispersion even if it just from temperature or concentration gradients).

Myths are a fertile topic maybe because of all the fertilizer in process control. You can make almost any point you want, by changing what are often obscure details on process and automation system dynamics. For example, you can show a variable speed drive can do better or worse than a control valve. The results can easily be swayed by VSD settings (e.g. deadband and rate limiting), VSD options (e.g tachometer feedback and vector control), and valve type\accessories (e.g. throttling sliding stem or rotary isolation valve and digital dual relay positioner versus pneumatic spool positioner). For insights into the relative merits of the VSD versus control valve in terms of control loop performance, check out the February Control Talk column in Control magazine titled "Deal or No Deal.”

I promised to post this week the development of equations that are a myth buster. The equations show there is an implied dead time greater than the actual dead time in most loops because the controller is tuned to be slower than what is shown in academic articles and papers. Control loop performance does not appreciably deteriorate until the actual dead time exceeds the implied dead time. The equations go on to provide an estimation of the peak and integrated absolute errors for the implied and actual dead times for step disturbances. The effect of the slowness of real life load disturbances can be roughly included by adding the load disturbance time constant to the process time constant in the equations for the peak and integrated errors. The first page appeared in a blog and Control Talk column in 2006. This updated document has better explanations/nomenclature and adds a second page for the estimation of the peak and integrated errors. When I bounce out of negative free time, I will do an application note to study the accuracy and implications of the equations. Next week we will continue on with mythology.

ScanTimeEffectonPeakandIntegratedErrors