Wednesday, April 25, 2018

IAM Search

Process Characterization - Page 7

Article Index
Process Characterization
Self-Regulating Processes
Integrating Processes
Runaway Processes
Steady-State Process Gain
Lag Time
Multiple Lag (Orders)
Dead Time

Multiple lags (orders)

Simple, self-regulating processes tend to be first-order: that is, they have only one mechanism of lag. More complicated processes often consist of multiple sub-processes, each one with its own lag time. Take for example a convection oven, heating a potato. Being instrumentation specialists in addition to cooks, we decide to monitor both the oven temperature and the potato temperature using thermocouples and remote temperature indicators:



The oven itself is a first-order process. Given enough time and sufficiently thorough air circulation, the oven’s air temperature will eventually self-stabilize at the heating element’s temperature. If we graph its temperature over time with the heater power fixed in “manual” mode (no thermostat to control it), we will see a classic first-order function:



The potato forms another first-order process, absorbing heat from the air within the oven (heat transfer by convection), gradually warming up until its temperature (eventually) reaches that of the oven11. From the perspective of the heating element to the oven air temperature, we have a first-order process. From the perspective of the heating element to the potato, however, we have a second-order process.

Intuition might lead you to believe that a second-order process is just like a first-order process – except slower – but that intuition would be wrong. Cascading two first-order lags creates a fundamentally different time dynamic. In other words, two first-order lags do not simply result in a longer first-order lag, but rather a second-order lag with its own unique characteristics.

If we superimpose a graph of the potato temperature with a graph of the oven temperature (once again assuming constant power output from the heating element, with no thermostatic control), we will see that the shape of this second-order lag is different. The curve now has an “S” shape, rather than a consistent downward concavity:



This, in fact, is the tell-tale signature of multiple lags in a process: an “S”-shaped curve rather than the characteristically abrupt initial rise of a first-order curve.

If we were able to ramp the heater power at a constant rate and graph the heater element, air, and potato temperatures, we would clearly see the separate lag times of the oven and the potato as offsets in time at any given temperature:



As another example, let us consider the control of level in three cascaded, gravity-drained vessels:



From the perspective of the level transmitter on the last vessel, the control valve is driving a third-order process, with three distinct lags cascaded in series. This would be a challenging process to control, and not just because of the possibility of the intermediate vessels overflowing (since their levels are not being measured)!

When we consider the dynamic response of a process, we are usually concerned primarily with the physical process itself. However, the instruments attached to that process also influence lag orders and lag times. As discussed in the previous subsection, almost every physical function exhibits some form of lag. Even the instruments we use to measure process variables have their own (usually very short) lag times. Control valves may have substantial lag times, measured in the tens of seconds for some large valves. Thus, a “slow” control valve exerting control over a first-order process effectively creates a second-order loop response. Thermowells used with temperature sensors such as thermocouples and RTDs can also introduce lag times into a loop (especially if the sensing element is not fully contacting the bottom of the well!).

This means it is nearly impossible to have a control loop with a purely first-order response. Many real loops come close to being first-order, but only because the lag time of the physical process swamps (dominates) the relatively tiny lag times of the instruments. For inherently fast processes such as liquid flow and liquid pressure control, however, the process response is so fast that even short time lags in valve positioners, transmitters, and other loop instruments significantly alter the loop’s dynamic character.

Multiple-order lags are relevant to the issue of PID loop tuning because they make the process harder to control with proportional and integral actions. The more lags there are in a system, the more delayed and “detached” the process variable becomes from the influence of the controller’s output signal.

A mathematically convenient way to model the lags in a system is in terms of phase shift when driven by a sinusoidal-shaped stimulus. A function exhibiting lag will cause the outgoing waveform to lag behind the input waveform by a certain number of degrees at one frequency. The exact amount of phase shift usually depends on frequency – the higher the frequency, the more phase shift (to a maximum of -90o for a first-order lag):



The phase shifts of multiple, cascaded lag functions (or processes, or physical effects) add up. This means each lag in a system contributes an additional negative phase shift to the loop. This may be a bad thing for negative feedback, which by definition is a 180o phase shift. If sufficient lags exist in a system, the total loop phase shift may approach 360o, in which case the feedback becomes positive (regenerative): a necessary condition for oscillation.

It is worthy to note that multiple-order lags are constructively applied in electronics when the express goal is to create oscillations. If a series of RC “lag” networks are used to feed the output

of an inverting amplifier circuit back to its input with sufficient signal strength intact12, and those networks introduce another 180 degrees of phase shift, the total loop phase shift will be 360o (i.e. positive feedback) and the circuit will self-oscillate. This is called an RC phase-shift oscillator circuit:



The amplifier works just like a proportional-only process controller, with action set for negative feedback. The resistor-capacitor networks act like the lags inherent to the process being controlled. Given enough controller (amplifier) gain, the cascaded lags in the process (RC networks) create the perfect conditions for self-oscillation. The amplifier creates the first 180o of phase shift (being inverting in nature), while the RC networks collectively create the other 180o of phase shift to give a total phase shift of 360o (positive, or regenerative feedback).

In theory, the most phase shift a single RC network can create is -90o, but even that is not practical13. This is why more than two RC phase-shifting networks are required for successful operation of an RC phase-shift oscillator circuit.

As an illustration of this point, the following circuit is not capable14 of self-oscillation. Its lone RC phase-shifting network cannot create the -180o phase shift necessary for the overall loop to have positive feedback and oscillate:



The RC phase-shift oscillator circuit design thus holds a very important lesson for us in PID loop tuning. It clearly illustrates how multiple orders of lag are a more significant obstacle to robust control than a single lag time of any magnitude. A purely first-order process will tolerate enormous amounts of controller gain without ever breaking into oscillations, because it lacks the phase shift necessary to self-oscillate. This means – barring any other condition limiting our use of high gain, such as process noise – we may use very aggressive proportional-only action (e.g. gain values of 20 or more) to achieve robust control on a first-order process15. Multiple-order processes are less forgiving of high controller gains, because they are capable of generating enough phase shift to self-oscillate.


11Given the presence of water in the potato which turns to steam at 212 oF, things are just a bit more complicated than this, but let’s assume a completely dry potato for now!

12The conditions necessary for self-sustaining oscillations to occur is a total phase shift of 360o and a total loop gain of 1. Merely having positive feedback or having a total gain of 1 or more will not guarantee self-sustaining oscillations; both conditions must simultaneously exist. As a measure of how close any feedback system is to this critical confluence of conditions, we may quantity a system’s phase margin (how many degrees of phase shift the system is away from 360o while at a loop gain of 1) and/or a system’s gain margin (how many decibels of gain the system is away from 0 dB while at a phase shift of 360o). The less phase or gain margin a feedback system has, the closer it is to a condition of instability.

13At maximum phase shift, the gain of any first-order RC network is zero. Both phase shift and attenuation in an RC lag network are frequency-dependent: as frequency increases, phase shift grows larger (from 0o to a maximum of -90o) and the output signal grows weaker. At its theoretical maximum phase shift of exactly -90o, the output signal would be reduced to nothing!

14In its pure, theoretical form at least. In practice, even a single-lag circuit may oscillate given enough gain due to the unavoidable presence of parasitic capacitances and inductances in the wiring and components causing multiple orders of lag (and even some dead time). By the same token, even a “pure” first-order process will oscillate given enough controller gain due to unavoidable lags and dead times in the field instrumentation (especially the control valve). The point I am trying to make here is that there is more to the question of stability (or instability) than loop gain.

15Truth be told, the same principle holds for purely integrating processes as well. A purely integrating process always exhibits a phase shift of -90o at any frequency, because that is the nature of integration in calculus. A purely first-order lag process will exhibit a phase shift anywhere from 0o to -90o depending on frequency, but never more lagging than -90o, which is not enough to turn negative feedback into positive feedback. In either case, so long as we don’t have process noise to deal with, we can increase the controller’s gain all the way to eleven. If that last sentence (a joke) does not make sense to you, be sure to watch the 1984 movie This is Spinal Tap as soon as possible. Seriously, I have used controller gains as high as 50 on low-noise, first-order processes such as furnace temperature control. With such high gain in the controller, response to setpoint and load changes is quite swift, and integral action is almost unnecessary because the offset is naturally so small.

Related Articles


  • ...more


Important: All images are copyrighted to their respective owners. All content cited is derived from their respective sources.

Contact us for information and your inquiries. IAMechatronics is open to link exchanges.

IAMechatronics Login