Disadvantages of unnecessary increase in PID gain

[Abdul Wali]

hi, I read somewhere that if we increase the PID gains unnecessarily, it will cause more energy consumption of the controller and it will lead to noise and distortion in the practical. May someone please give me more explanation on this?? that how it lead to noise or distortion in practical??

 

[anorlunda]

Derivatives amplify noise, Integral action filters noise.

 

[Averagesupernova]

@jim hardy
I would think your power plant experience would be good input to this thread.

 

[jim hardy]

hi, I read somewhere

Where ?

if we increase the PID gains unnecessarily, it will cause more energy consumption of the controller

In the power plant energy consumption of the controller is of absolutely zero concern. It is miniscule compared to what it's controlling

and it will lead to noise and distortion in the practical.

That sounds to me like evasion of something the author didn't want to go into..

What is of concern is stability - do you understand "loop gain" and "Phase margin" ? Too much gain in any of the three terms will cause instability..

If it leads to "hunting", ie steady state oscillation , the constant motion wears mechanical linkages causing looseness. It wears valve stems.
Worn valve stems leak past the packing. Loose linkages cause hysteresis which itself causes hunting
Severe "Hunting " will likely cause thermal cycling which is hard on big mechanical parts.

To 'tune' a loop one typically selects proportional only , applies a small step change and observes system response. You start at low gain so it's sluggish and watch effect of adding more gain.
You can tell from overshoot when gain is too high so you back off.
Then you apply some integral to drive steady state error to zero.Too much integral will encourage oscillation at around 2/3 integral time setting so you back off..
Derivative you add carefully (if at all) to sharpen response , watching all the time for underdamping which shows you're approaching instability.
Here is an article by an old line company , Fisher, one of the pioneers of automatic control systems.
http://www2.emersonprocess.com/siteadmincenter/PM Articles/LoopTuning_CEP_Jan2013_LowRes.pdf

Fisher's 1950's instruction bulletins on tuning control systems by frequency response analysis were brilliant - i hope to stumble across them online one day.

hope above helps

old jim

 

[Asymptotic]

@jim hardy laid it out well.

more energy consumption of the controller

Not the controller itself, but this may be (and often is) true of the controlled process.

Consider a TCU (Temperature Control Unit) consisting of a centrifugal pump, 12 kW resistance heater, and a cooling solenoid connected to a 50°F chilled water source for 'direct injection' cooling. In temperature trending below, 'As-Is' loop tuning had the output slamming 'rail-to-rail' between 100% heating, and 100% cooling with temperature oscillating between 90°F to 104°F (14°F peak-to-peak variation) for a 70°F setpoint. After 'final' tuning, output remained between 30% cooling to 40% cooling providing a process temperature ranging between 69.6°F to 70.4°F (0.8°F peak-to-peak variation).

This controller has somewhat peculiar parameters including a separate term for cooling proportional gain (typically, commercial temperature controllers have a heating prop gain, and a multiplier term derived from it for cooling prop gain) and a 'Cooling Ratio' term that, as it turned out, was a range matching scheme so anything from chilled water to hot oil cooling sources could be used. 'Reset' and 'Rate' are scaled in seconds, and are an older terminology used to denote integrating and derivative gains. At the time I didn't have a copy of the manual, and was conducting multiple trials using different tuning parameters to better understand how they worked. Data collection for this trend continued to the end of my work shift. The 'final' reset (integrating) time of 120 seconds shown in the table is too high (note how long 'trial 12' temperature took to converge to setpoint), and was touched up the next day to 40 seconds.

My interest then was to minimize temperature swing, and I was collecting data by scrawling it into a notebook every 5 seconds so didn't bother (nor had time) to collect % output information. The 'humpedness' of 'As-Is' temperature variation suggests cooling on-time % wasn't symmetrical with heating % (it was probably more like 30% cooling/70% heating rather than 50%/50%), but for the purpose of cost estimation let's pretend it was a 50:50 split.

Energy cost for a 12 kW heater at $0.10 USD/hour is $1.20/hour at 100% heating. It is difficult to estimate an accurate value for cooling, what with cooling tower and process pump and chiller compressor energy, maintenance, makeup water, treatment chemical, and other costs, but a kWh of chilled water cooling ends up costing more than a kWh of electric heating. For now, let's guess $1.50 an hour for cooling.

An hour of 'As-Is' temperature control operation at 50% heating/50% cooling would be (0.5*$1.20)+(0.5*$1.50), or $1.35 an hour. After tuning, output no longer traversed into the heating side, and averaged 35% cooling (0.35*$1.50), or about $0.53 an hour.

In this case, improper loop tuning cost approximately $0.82 extra per operating hour.

 

[dlgoff]

What is of concern is stability

Stability in PID control can be augmented with dead-bands/limits.

 

[Abdul Wali]

@jim hardy @Asymptotic Thanks for your replies. From some research and simulation analysis, i found the following information. "Increasing a forward path gain will increase controller output voltage and current and will try to accelerate the plant at a higher rate; the response will be faster so viscous friction will increase, also amplifier saturation will cause nonlinear effects." So what is your opinion about this??

 

[Asymptotic]

@jim hardy @Asymptotic Thanks for your replies. From some research and simulation analysis, i found the following information. "Increasing a forward path gain will increase controller output voltage and current and will try to accelerate the plant at a higher rate; the response will be faster so viscous friction will increase, also amplifier saturation will cause nonlinear effects." So what is your opinion about this??

An answer depends upon the context. "Increasing a forward path gain will increase controller output voltage and current" could indicate a PID controller equipped with a 0-10 volt DC or 4-20 mA DC analog output signal scaled to represent 0% to 100% output. It could instead be referring to outputs from op amps forming an analog computer purpose-built to solve PID equations, although these days nearly all commercially available PID controllers are digital. Bottom line - increasing proportional gain increases output.

"... response will be faster so viscous friction will increase ...". Response speed increases as proportional gain increases, but whether 'viscous friction' is involved depends upon the application.

"... amplifier saturation will cause nonlinear effects." suggests the author is indeed referring to an analog PID controller built from op amps. However, output saturation doesn't depend upon whether the controller is analog or digital, and occurs whenever loop output is commanded to 100% (output is "saturated" because it can't go any farther) particularly if it is held there for an appreciable length of time. Controller output saturation is by definition non-linear, and causes another non-linearity known as integrator 'windup'.

PID controllers come in a wide variety of flavors. Off-the-shelf PID temperature controllers such as those available from Eurotherm, Watlow, Fuji, RKC, and other manufacturers usually handle only thermocouple and RTD inputs, and have parameter sets suitable for temperature control. AC inverters and DC motor controllers typically have at least two concurrently operating PID loops (a velocity loop for speed control, and a current loop controlling their respective IGBT or SCR bridges) embedded within them with I and D terms operating at vastly smaller time scales than those appropriate for temperature controllers. Also, as @anorlunda pointed out, the derivative term amplifies noise - certain process loops can't handle it, and the controller must be set up to operate as PI only (D is turned off).

It has been a while since I last read it (... don't know where it is) and it doesn't appear to have been updated since a 2nd edition was released in 1995, but I found the ISA (Instrument Society of America) publication, "PID Controllers: Theory, Design, and Tuning" quite useful. This book's ISA web page has a link to a set of ILMs (Interactive Learning Modules) regarding PID control that are available for download by non-members.

 

[Abdul Wali]

It could instead be referring to outputs from op amps forming an analog computer purpose-built to solve PID equations, although these days nearly all commercially available PID controllers are digital. Bottom line - increasing proportional gain increases output.@Asymptotic Actually, i am talking about the controller itself, i simulated and found that when the gains increase the noise in the output waveform of the controller also increase the waveform amplitude increases too.

 

[Asymptotic]

It could instead be referring to outputs from op amps forming an analog computer purpose-built to solve PID equations, although these days nearly all commercially available PID controllers are digital. Bottom line - increasing proportional gain increases output.@Asymptotic Actually, i am talking about the controller itself, i simulated and found that when the gains increase the noise in the output waveform of the controller also increase the waveform amplitude increases too.

Can you capture and upload this waveform?

It sounds like you are describing steady oscillation caused by excessively high proportional gain (or, from the reciprocal perspective, a too narrow proportional band), and/or the loop has too much integration and/or derivative. The Ziegler-Nichols tuning heuristic can be useful. In a nutshell, turn off I and D, take note of the oscillation period, and adjust P gain to just under the value where oscillation stops. Adjust I and D parameters as suggested by the Ziegler-Nichols method, then (if necessary) tweak the PID parameters to achieve whatever response curve suits your needs, for instance, quarter damped response.

 

[Abdul Wali]

Can you capture and upload this waveform?

It sounds like you are describing steady oscillation caused by excessively high proportional gain (or, from the reciprocal perspective, a too narrow proportional band), and/or the loop has too much integration and/or derivative. The Ziegler-Nichols tuning heuristic can be useful. In a nutshell, turn off I and D, take note of the oscillation period, and adjust P gain to just under the value where oscillation stops. Adjust I and D parameters as suggested by the Ziegler-Nichols method, then (if necessary) tweak the PID parameters to achieve whatever response curve suits your needs, for instance, quarter damped response.

@Asymptotic https://drive.google.com/open?id=0B9NQhKDld_D4SzZuN2xUZ2RXRlU output of pid

 

[jim hardy]

"Increasing a forward path gain will increase controller output voltage and current and will try to accelerate the plant at a higher rate; the response will be faster so viscous friction will increase, also amplifier saturation will cause nonlinear effects."

Sounds like another catch-all phrase that doesn't really say much.

Of course there's non linearity in the process being controlled . Manufacturers go to greet lengths to give valves predictable % flow versus % opening characteristics , and the valve positioner that converts controller signal to valve position is characterizable. Ours came with three different cams to select linear, square or square root transfer function. Flow vs pressure characteristic of pumps and piping is present too. Fortunately it's the nature of closed loops that they are pretty forgiving of it.
When you set up your control loop you have to pay attention to expected range of signals and stay away from saturation.

Yes the controllers i used were op-amp based. Most were electronic

Here's a clever one that uses a single amplifier, devised back in the 40's when opamps were something exotic .

https://pdfs.semanticscholar.org/15ea/514e8e45966c99b0a3a0b7fb873c0ecf0400.pdf
Since it has integral(reset) a limiter on Vout would be prudent.

Some were pneumatic running on compressed air 3 to 15 or 3 to 27 psi. An orifice is a resistor and a volume chamber is a capacitor. They sum forces instead of currents.
Wow- here's an instruction manual for one . I didnt expect to find that.
The vane-nozzle is your null detector, and the booster relay is your opamp with gain of perhaps fifty.

http://www.infi90.com/Files/Bailey Infi90 Documentation/Mini-Line 500 Controller Type AD.pdf
They're impervious to EMP attack .
But we took out the last ones around 1988.

old jim

 

[Asymptotic]

@Asymptotic https://drive.google.com/open?id=0B9NQhKDld_D4SzZuN2xUZ2RXRlU output of pid

Is there any way to plot deviation separate from output (or is there an option for a second Y scale)? The Y axis is scaled +1*10⁵ to -1*10⁵, what appears to be the output signal swings from +80% to -80%, and this corresponds with a small perturbation in the blue line which I'm taking to be deviation from setpoint. In any event, if these output swings are what the PID loop is coming up with to maintain setpoint, it looks like proportional gain is too high by quite a bit.

 

[Asymptotic]

Some were pneumatic running on compressed air 3 to 15 or 3 to 27 psi. An orifice is a resistor and a volume chamber is a capacitor. They sum forces instead of currents.

Never played with pneumatic PID loop controllers, but have fondly frightening memories of pneumatic logic controllers before electronic PLCs took over. Remember how timing relay setpoints were tuned by cutting nylon tubing coils to various lengths? Once a newbie decided to 'clean up' the resulting rat's nest ... it took days to get the machine back into operation ;).

 

[jim hardy]

At 500% i can't tell which moves first

 

[Abdul Wali]

At 500% i can't tell which moves first

View attachment 210632

@jim hardy and @Asymptotic One more question, can i call this output voltage of the controller as voltage consumption / Energy consumption of the controller?

 

[Asymptotic]

@jim hardy and @Asymptotic One more question, can i call this output voltage of the controller as voltage consumption / Energy consumption of the controller?

I don't think you can consider loop output to be voltage. 0.8 *10⁵ is 80,000, and a controller output of 80 kV wouldn't make sense.

 

[jim hardy]

can i call this output voltage of the controller as voltage consumption / Energy consumption of the controller?

I agree with @Asymptotic

Two points:
1. A digital computer can calculate and report that an ordinary little analog controller is producing 80 kilovolts
but outside of particle accelerator or something similar, ie in normal world of process control, that's laughable .
ever heard the phrase "Garbage In Gospel Out" ?
Are you working on CERN or something ? Otherwise,
see my signature

2. Energy is power X time. Power is volts X amps and you haven't defined amps.

old jim

 

[Abdul Wali]

I don't think you can consider loop output to be voltage. 0.8 *10⁵ is 80,000, and a controller output of 80 kV wouldn't make sense.

@Asymptotic on page 277 of the attached article look at the equation 7 and then on page 279 look at the Simulink model, from the equation it can be clearly seen that the E(s) is the input to the motor or output of the controller. but yeah it is E(s) not E(t). but still voltage. https://drive.google.com/open?id=0B9NQhKDld_D4WG1KSVBJWnA3eW8 Now what do you think? @jim hardy

 

[jim hardy]

https://drive.google.com/open?id=0B9NQhKDld_D4WG1KSVBJWnA3eW8 Now what do you think? @jim hardy

I think with 230 volt supply it's hard to make 80 kilovolts
and even if you did you'd wreck your servomotors

from that article, page 278

 

[Abdul Wali]

I think with 230 volt supply it's hard to make 80 kilovolts
and even if you did you'd wreck your servomotors

from that article, page 278
View attachment 210661

@jim hardy as in the article he is comparing PID and IMC, so normally if u see the waveforms IMC response is better than PID but when i tune pid to the level where its response is even better than PID than the 80kv happens. else in normal condition, when the pid response is same as shown in the paper, this does not happen.

 

[Asymptotic]

@Asymptotic on page 277 of the attached article look at the equation 7 and then on page 279 look at the Simulink model, from the equation it can be clearly seen that the E(s) is the input to the motor or output of the controller. but yeah it is E(s) not E(t). but still voltage. https://drive.google.com/open?id=0B9NQhKDld_D4WG1KSVBJWnA3eW8 Now what do you think? @jim hardy

A 230V motor winding insulation system would punch through within microseconds with 80,000 volts applied. I'm not conversant with MATLAB and Simulink, but for it to allow 80kV as a legitimate value suggests the model has no means to flag physically unrealizable answers.

Actual servo positioning controllers have a 'following error' parameter used to trip the servo drive amplifier (to save it from an untimely death) if error between commanded and actual position rises beyond a certain point. One of the several common causes of following errors is too high a ramp rate, and attempting to accelerate the driven system faster than the drive/motor can provide power for.

 

[jim hardy]

@jim hardy as in the article he is comparing PID and IMC, so normally if u see the waveforms IMC response is better than PID but when i tune pid to the level where its response is even better than PID than the 80kv happens. else in normal condition, when the pid response is same as shown in the paper, this does not happen.

Doesn't that suggest your computer program is not bounded by reality when it comes to calculating voltages ?

One can calculate things that are physically impossible .
For example
"How much tension would be required in a 1/4 inch rope to lift Titanic?"

What is the power calculated by your program when it asks for 80 kv ?