Controlling head-pressure

[desA]

I'd like to investigate & debate the various methods used to control head-pressure in rhvac systems.

The simple ones first:
1. Fan speed control for air-cooled condensers;
2. Water flow control for water-cooled condensers.

There are number of others. I'd be interested in exploring the merits of the various methods, in terms of effectiveness, ease-of-control, stability, reliability, energy usage, cost.

A few questions to kick off:
a. What happens if the ambient temperature is very cold & cools the condenser too much, resulting in a low discharge pressure at startup?
b. Can head-pressure control result in rapid swings in condensing temperature/pressure?
c. What happens if a temp/pressure sensor controlling water pump speed, is overly sensitive?
d. What happens if the water pump cavitates - what effect does this have on the condenser?

[desA]

An example:

http://i40.tinypic.com/2je0pe.png

Plot of pressure ratio across compressor, versus time on lower axis - from a cold start-up.

What is happening in the first part of the plot?

[cadillackid]

im always a fan of soft start, variable speed compressors (granted im only on the domestic HVAC side of things).. but soft start allows for boil off of stagnant liquid to eliminate slugging on start up in cold weather.. (even crank case heaters dont always do their job)..

and then variable speed fans on the condensors...

you can also use a partially Liquid flooded coil approach too.. but a lot of techs will have a hard time getting the charge correct on such a system...

-Christopher

[Magoo]

Hi DesA.
Anything is possible so long as the cheque book is bottomless.
For a marketable product you need to consider localized conditions and probable applications conditions and design accordingly.
To design a blanket coverage product means pricing product out of reach of "Joe Average".
magoo

[desA]

Thanks Magoo.

I tend to agree with you & far prefer simple, reliable systems & system design.

What I suspect is happening in the example above is either of the following:
1. Head-pressure control circuit is unstable; or
2. Internal 4-way valve, of HGB, is bypassing.

I've seen this twice on the same machine (made by a.n. other) & it does look out-of-place.

[Tesla]

Hi desA
For a start in my mind there are two systems - old and new. New systems don't have oil (where oiled systems rely on elevated pressures/temperatures to pump it around a system). As a contols tech for the last six years I have only come across two systems that required derivative control. 1 with two stage condenser cooling fans and a short circuit chilled water loop. With the oilless centrif turbocor, it uses low ambiant cooling medium for efficiency where as oiled compressors do not. But there is a limit to the pressure differential across the throttling valve which produces the refrigerating effect. Hence there is a ballance of design with new technologies to produce a more efficient system.
Because we are dealing with finely ballanced and tuned systems where the control is or has the biggest effect. When we get into the bigger systems we are interested in efficiency and require more advanced contol like utilising P, I, D (with critical dampening) and anticipitory control with hysterisis. A good knowledge of pressure enthalpy of refrigerants and new technologies helps but personally I think the way forward is in plasma technologies (still in its infancy for us) So
A; limmited or no refrigeration effect
B; yes
C; = question B
D; sluggish condensing and slightly elevated cond temps, unless B
It becomes very viable to write a program for an integrated circuit when it works and is mass produced to serve a wide variety of systems from large to small.At the end of the day we have been in the computer age for more than three decades.

[desA]

Hi Tesla,
Can I ask you to please explain again - slowly - you've lost me.

[Tesla]

Hi desA
To sum up what I wrote above. The majority of systems out there use little or no control (pressure switch) to control head pressure. On bigger systems Most of these still use on/off control - the most basic form of control.
With new technologies emerging like VSD variable speed drive, VRF variable refrigerant flow, and oilless compressors such as the turbocor (centrif) and some others (scroll). We can control the refrigeration/AC systems so much better, with much tighter parameters using PID proportional intergral derivative control modes. It's just a mater of time before the smaller systems use these control methods. The cost of an intergrated chip on a controller only costs a couple of dollars and we can do a lot more.
eg on a previous post another member wrote that in practice it is impossible to build an instantanious hot water heatpump without a water tank. I know the technology is available to build one and I will (one day). To me a hot water tank is money up in smoke and hot/warm water down the drain is dollars down the drain. One of the more complex issues with heatpumps is controlling the head pressure.
If I digress a little and we go back in time the refrigeration systems were the phase change of ice. Now we are familliar with phase change of liquid/gas. As we progress the next logical method should be the phase chnge of plasma. Considering when I went to school they taught me there were only three states of matter, i know there are at least five with a possible sixth being disputed at the moment.
High temp hi COP heat pumps - which would be the best refrigerants - perhaps these;
Water, CO2, Amonia, R22.
1 VSD best method to control fans (Zeal Fans have it built into the motor for smaller ones)
2 VSD best method to control water flow and pressure in conjunction with a few valves.
Cycling fans off and on is a waste of energy - too much bypass air.

[mad fridgie]

Hi Tesla,
it was I who said it was not practical to build an instantinoues heat pump.
I think I need to clarify, the practical aspects.
I live in NZ and come from the UK, so the system has to work in cold weather, so at some point a defrost period is required, at this point no or limited heating is produced, so either a buffer storage tank or another source of energy is required.
you then need to look at the time constraints of mechanical refrigeration, time is required to get the momentum going, so there will always be a reasonable lag (can be counteracted by a buffer tank)
if we look a good shower uses 12L/M (0.2L/S) yes could be less, many prefer more. If we presume 45C out of the shower head, to allow for 40C on the body, with water entering at 5C we a unit that can provide around 33kw, if the ambient is -5C at the very best the COP would 3, then power supply needed would be 11Kw, thats quite draw on the electrical supply system.
If now then allow 40% reduction due to drain recovery, draw would still be 6.6kw inductive load, that a large starting current (DOL), so a soft starter is required,which again slows the heating process.
Then we have look at the number of starts (on to on periods) in a family house hold, this could be every couple minutes. Excess residual heat could build up in the compressor motor windings. (this could be over come with larger stator and rotor)
Either way some form of thermal buffer is required.
I hope this clear up any mis- understanding

[desA]

Thanks so much for clarifying you very interesting post. I'm fascinated by your forward thinking. It is a breath of fresh air. Many thanks. :)


Hi desA
To sum up what I wrote above. The majority of systems out there use little or no control (pressure switch) to control head pressure. On bigger systems Most of these still use on/off control - the most basic form of control.
With new technologies emerging like VSD variable speed drive, VRF variable refrigerant flow, and oilless compressors such as the turbocor (centrif) and some others (scroll). We can control the refrigeration/AC systems so much better, with much tighter parameters using PID proportional intergral derivative control modes. It's just a mater of time before the smaller systems use these control methods. The cost of an intergrated chip on a controller only costs a couple of dollars and we can do a lot more.

I understand fully where you are coming from. Yes, you are completely correct.

I would say this, though, I will bet that the dynamics of transient changes in the cycle induced by the current on/off control techniques, are probably not well understood - let alone bringing in PID to the equation.

For the case of a simple set of linear system blocks - one reverse acting, controlled by a PID, with miniscule reference signal perturbation, the system goes into oscillations & operational drift. This, ironically, models the system shown in the plot, in post #2, fairly well.

I know full well that the original system designer would not have expected this response & would, in all likelihood, not know how to control it effectively.


eg on a previous post another member wrote that in practice it is impossible to build an instantanious hot water heatpump without a water tank. I know the technology is available to build one and I will (one day). To me a hot water tank is money up in smoke and hot/warm water down the drain is dollars down the drain.

It is absolutely not an issue to build an instantaneous heater - in fact it is a cinch. The problem is that it generally ends up being huge, for the size of instantaneous heat delivery you are looking for. As MF rightly stated, some form of thermal storage would be useful. It does not have to be a water tank, though.


One of the more complex issues with heatpumps is controlling the head pressure.

I would tend to agree. :D


High temp hi COP heat pumps - which would be the best refrigerants - perhaps these;
Water, CO2, Amonia, R22.
1 VSD best method to control fans (Zeal Fans have it built into the motor for smaller ones)
2 VSD best method to control water flow and pressure in conjunction with a few valves.
Cycling fans off and on is a waste of energy - too much bypass air.

Some good points here. Thank you so much.

[Tesla]

Hi mad fridgie
I was born and trained in NZ Windy Wellington. The ground temp from where the water flows in NZ rarely drops below 9degC and has an average supply temp of 11 to 15 summer or winter.
There is no mis-understanding I simply disagree. No disrespect intended mad fridgie and I welcome any further comments

[mad fridgie]

Hi mad fridgie
I was born and trained in NZ Windy Wellington. The ground temp from where the water flows in NZ rarely drops below 9degC and has an average supply temp of 11 to 15 summer or winter.
There is no mis-understanding I simply disagree. No disrespect intended mad fridgie and I welcome any further comments
No disrespect taken, this is a forum for descussion, It would be boring if we all thought the same.
I live in the country, on semi pumped system water can fall below 5C and in the UK ground water can be also as low as 5C (this why we need to insulate the cold water pipes)
If we are talking about systems in higher ambients, then becomes more feasable, no need for defrost, TD would be smaller, so load would reduce, and performance would increase, thus lowering power draw on both counts.
We are still left with mechanical reation times and possible short cycling (vapour compression cycles)
A small thermal buffer is required. (need not be water)
Moving on
Yes water is an excellent refrigerant, only problem at the moment is ther is no efficient compressor available, I also agree, that in the medium to long term vapour compression as we know it will change.

[Tesla]

Hi guys
Thanks this discussion has been most consrtuctive, interesting and... well motivating. I might add one other interesting point on the topic of most suitable refrigerant to use. I have played with the idea of these garden water atomisers. They use only 25% of the energy compared to heating for a change in state. I was wondering if they would be best utilised in the evaporator to help boil the refrigerant (water)? What are your thoughts on this and could it be incorperated in a centrif like the turbo cor - as I understand danfoss is working on or prototyping much smaller compressors.

[desA]

How would this affect the control of head-pressure (HP)?

[Tesla]

Looking outside the square it would reduce the heat of compression required to produce the refrigerating effect. By not having to do as much work to change itsself (water) as much as other refrigerants. Just a thought anyway desA

[mad fridgie]

One presumes that the graph is a multi pass system, then if that is the case why control head pressure, it is head pressure which primarily determines power draw, so its the averaged draw over time which is important, I can not see the point of an elevated head pressure just to keep a system stable, contol of the expansion device becomes the determining factor. At these lower ratios I do not seem any problems, even a level of benefit "variation in lubrication"

[desA]

How would pump flowrate affect the head-pressure of a water-cooled condenser?

[mad fridgie]

How would pump flowrate affect the head-pressure of a water-cooled condenser?
Yes this is the standard method of controlling head pressure "changing flow rate" either pump speed or reg valve, pressure activated valves sized correctly are very stable

[desA]

Thanks MF. The pressure-activated, or temperature-activated control valves seem reasonably stable, with decent time constants.

Now, for direct pump speed control - what controls the pump motor speed?

In other words, a sensor measures something, then a controller controls pump speed - via alternative means. What is typically sensed to effect this control mechanism?

(I'm labouring a point, on purpose here).

[Gary]

Thanks MF. The pressure-activated, or temperature-activated control valves seem reasonably stable, with decent time constants.

Now, for direct pump speed control - what controls the pump motor speed?

In other words, a sensor measures something, then a controller controls pump speed - via alternative means. What is typically sensed to effect this control mechanism?

(I'm labouring a point, on purpose here).

This would depend on what you are trying to accomplish. If the object is to control head pressure, then wouldn't you want to sense head pressure?

[desA]

This would depend on what you are trying to accomplish. If the object is to control head pressure, then wouldn't you want to sense head pressure?

Fair question.

Since pressure & temperature are inter-convertible in many instances, would it be better to sense off pressure, or temperature?

The aspect I'm trying to explore here is the time constants, or system response to a flow change, with its subsequent effect on the sensor.

For instance, would a pressure sensor act slower, or faster, than a temperature sensor?

[Gary]

Fair question.

Since pressure & temperature are inter-convertible in many instances, would it be better to sense off pressure, or temperature?

The aspect I'm trying to explore here is the time constants, or system response to a flow change, with its subsequent effect on the sensor.

For instance, would a pressure sensor act slower, or faster, than a temperature sensor?

Pressure would be sensed faster than temperature.

Also keep in mind that a pressure sensor (transducer) is much more expensive than a temperature sensor.

[Tesla]

When talking about the pump and its water (condenser water loop) and the bigger scale like a building full of units. The pressure (water) is sometimes controlled via a pressure by-pass. Where the system (building) is by-passed and the water is diverted around the cooling tower. The above system is not very common. What is very common is temperature by-pass. Where the cooling tower is by-passed when the water temp gets too low. I have worked on a building that used the pressure by-pass which had about 30 water cooled package units. Each unit used a johnsons condenser water pressure regulator which sensed the head pressure. Over the years some of these units were refitted with stad valves which caused slight over condensing durring cooler months. The pressure by-pass was dissabled and this caused the pressure regulating valves to leak and cavitate creating loud noisy vibrations. An actuator (electronic) with a temp sensor is much cheeper than the water reg valve and is used on some other systems.

[desA]

Pressure would be sensed faster than temperature.

Ok, good. Now, when you say 'faster' - how much faster is a pressure sensor in sensing a pressure change, to a temperature sensor sensing the temperature equivalent of that change? 2x, 5x, 10x, 100x - ballpark ? I'm looking to gain an estimate of the time constant differences between the two control concepts.


Also keep in mind that a pressure sensor (transducer) is much more expensive than a temperature sensor.

Very true. It is also invasive in terms of requiring a line cut-in.

[Gary]

Ok, good. Now, when you say 'faster' - how much faster is a pressure sensor in sensing a pressure change, to a temperature sensor sensing the temperature equivalent of that change? 2x, 5x, 10x, 100x - ballpark ? I'm looking to gain an estimate of the time constant differences between the two control concepts.


I've never timed it and I would suppose it would depend on the specific setup.

The most accurate measure of head pressure is head pressure. You will need to find a measurement point whose temp consistently coincides with pressure.

[desA]

I've never timed it and I would suppose it would depend on the specific setup.

Fair-enough. I was looking for ballpark, rough estimates. Trigger a change, how long for system to sense change, then stabilise water flow - in seconds, or minutes.


The most accurate measure of head pressure is head pressure. You will need to find a measurement point whose temp consistently coincides with pressure.

The sensed variable should closely correlate to the controlled parameter, I'd imagine.

The problem may then come in when the sensitivity of the pressure sensor, via a short time constant, interacts with the pump controller in such a way as to throw the system into instability. It then becomes unstable, or could even limit cycle, I'd imagine.

For the system in post #2, it looks like the system goes into some sort of unstable mode - for whatever reason - not known at this stage. I suspect that the head pressure is sensed & control action taken on the pump, controlling water flow, to so control condenser head-pressure. Something is gravely wrong, however.

[Gary]

For the system in post #2, it looks like the system goes into some sort of unstable mode - for whatever reason - not known at this stage. I suspect that the head pressure is sensed & control action taken on the pump, controlling water flow, to so control condenser head-pressure. Something is gravely wrong, however.

Does the pump amperage jump around with the Pr?

[desA]

Does the pump amperage jump around with the Pr?

From the info I have to hand, the pump amperage was not measured.

However, back-calculating the pump water mass flowrate shows a slight oscillation about a gradually declining mean flow. I've understood the declining average flow to be a symptom of the system oscillation & perhaps some integral action is a controller of sorts.

The Pr oscillation seems far more pronounced than the pump response, but they do seem to correlate.

[mad fridgie]

Its a Rheem, if it is then non of the above, unable to say anymore sorry

[Gary]

I'm thinking the first step would be to narrow down to the cause.

I would bypass the control and put the pump at 100% to see if the oscillations continue. If they do, it has nothing to do with the pump.

[desA]

Its a Rheem, if it is then non of the above, unable to say anymore sorry

This system has not been rheemed. :D

[desA]

I'm thinking the first step would be to narrow down to the cause.

I would bypass the control and put the pump at 100% to see if the oscillations continue. If they do, it has nothing to do with the pump.

I don't have direct access to this machine, at present & so we'll have to continue the thought experiments. I do have experimental results & plots of the machine operation during the instability. :)

Progressing the gedanken a little further:
What is the relationship between a change in pump speed & a change in head-pressure? ... &... how much does the pressure change with say a 1% change in pump speed?

[Gary]

I don't have direct access to this machine, at present & so we'll have to continue the thought experiments. I do have experimental results & plots of the machine operation during the instability. :)

Progressing the gedanken a little further:
What is the relationship between a change in pump speed & a change in head-pressure? ... &... how much does the pressure change with say a 1% change in pump speed?

I would expect a considerable lag between a change in flow and a change in pressure.

[desA]

I would expect a considerable lag between a change in flow and a change in pressure.

Good, so far.

How long a lag could be reasonably expected, in pressure, after a pump speed change? 1s, 5s, 10s, 30s, 60s, more?

What would the physical reason be for this action delay?

[Tesla]

The lag could be a long time like upto 60 seconds if the pump is a long way from the condenser. Last year I worked on one with a VSD that was 30 metres away - it had a control interferrance from the VSD being located too close and insufficient earth to motor and no screened cable. If you have the flow rate in litres/sec and the pipe diameter and length you can calculate how long it will take for the chnge of flow rate to the condenser. Then again through the condenser. This will be the time lag for a change in pump speed. A chnge in pump speed has a cube law less losses.

[desA]

This water pump would be located around 0.5-1 m from the condenser, as I remember.

[Tesla]

At a close guestimation 3 - 9 seconds. But on initial start up it could be a little longer due to slight compression and/or cavitation and a little for heat transfer time from water to refrigerant.

[desA]

Ok, so far we have a system with a response time from sensing, through change activation of between 3 & 9 seconds. I'd say, based on previous experience, that this is reasonable.

Continuing, the pressure sensor senses the pressure change & modifies the pump speed. This action would be much faster, I'd expect.

How much will the pressure signal be modified in response to a change in pump speed? Eg, pump speed changes 1%, what could the expected change in pressure be? 0.1%, 0.5%, 1%, 2%, 5%, 10%?

In other words, is the change in pressure amplified, or reduced relative to the change in pump speed?

[Tesla]

In short desA the answer is yes. In a controls algorithim as such above I would expect a time delay where your plot appears not to have one. I suspect the ommission of a time delay is the cause of the lack of control on start up. Time delays in control are important to allow the sensed parameter to stabilise before a change is made. What are your thoughts on this?

[Tesla]

In addition there appears to be too much proportional in the control response (on startup) - however with a time delay added this may take care of the problem. By the look of your plot I would estimate a time delay of 20 - 30 secs is reqiured but would start with 20 and re polt graph. As I wrote in a previous post I intend to write some software this year to calculate the best PID control for various system components with AC systems mainly for BMS control techs.

[desA]

In short desA the answer is yes. In a controls algorithim as such above I would expect a time delay where your plot appears not to have one.

The time scale for that plot is 260 minutes, with the unstable part being ~ 80 minutes.


I suspect the ommission of a time delay is the cause of the lack of control on start up. Time delays in control are important to allow the sensed parameter to stabilise before a change is made. What are your thoughts on this?

I agree totally about time lags (dead time) being a critical control issue. Very good observation.

Now, couple a highly sensitive pressure sensor, into a control system with long relative dead-time, & possible system gain > 1, & we have a very difficult scenario to control, in my view. This would even be the case with completely linear system element responses.

How to effectively stabilise such a system? ;)

[desA]

In addition there appears to be too much proportional in the control response (on startup) - however with a time delay added this may take care of the problem.

I wonder if we could simulate this system?

I have performed a preliminary system analysis & developed an initial Simulink model. If you have tools available, this could be an interesting exercise.

Thanks Tesla for your forward thinking - this could be fun.

[Tesla]

Ok bear with me here for a minute (from memory).
First taking a plot to record response time. plot the pump speed % and the head pressure or pressure ratio %. Start the system then induce a 10% step change (instantly) on pump speed.
Look at the steepest gradient on the plot from the head pressure or pressure ratio. Also take note of the delay between action and response - this will be roughly your response time. With the gradiant take dy dt - calculus, and I'm very rusty it's been 15 odd years. So cutting it short, so the gradiant will give you the intergral, the differance in amplitude of %s will give the proportional band and 1/pb = gain. Finally add the p + i to controller and replot, again take the gradiant dy dt and add again to controller this will be the derivative (if necessary) along with any bias. But please guys check this out from control strategies how to calculate pid before applying.
One last thing I thought your plot was in seconds so considering it ran unstable? for 80 odd minutes. What is the set point? and we need a more streatched out graph (in this period) to analyse please if possible.
We are aiming for quater amplitude decay for a steady system - assuming we are controling for a steady set point of head pressure for the fesult of a temperature output for a heatpump.

[desA]

Ok bear with me here for a minute (from memory).
First taking a plot to record response time. plot the pump speed % and the head pressure or pressure ratio %. Start the system then induce a 10% step change (instantly) on pump speed.
Look at the steepest gradient on the plot from the head pressure or pressure ratio. Also take note of the delay between action and response - this will be roughly your response time. With the gradiant take dy dt - calculus, and I'm very rusty it's been 15 odd years. So cutting it short, so the gradiant will give you the intergral, the differance in amplitude of %s will give the proportional band and 1/pb = gain. Finally add the p + i to controller and replot, again take the gradiant dy dt and add again to controller this will be the derivative (if necessary) along with any bias. But please guys check this out from control strategies how to calculate pid before applying.

Ok, excellent. I have included a few links to the Ziegler-Nichols PID tuning method, for those following this thread:

http://en.wikipedia.org/wiki/Ziegler%E2%80%93Nichols_method

http://www.chem.mtu.edu/~tbco/cm416/zn.html

I will also state, for the record, that I'm not quite sure what control philosophy the manufacturer of this machine is actually using. I do suspect though that it is some form of PID controller, operating off a pressure sensor.



One last thing I thought your plot was in seconds so considering it ran unstable? for 80 odd minutes. What is the set point? and we need a more streatched out graph (in this period) to analyse please if possible.
We are aiming for quater amplitude decay for a steady system - assuming we are controling for a steady set point of head pressure for the fesult of a temperature output for a heatpump.

What is the set-point? This is the million $ question, in my mind.

Will a single set-point & set of tuning parameters, be sufficient for all operating temperatures, & temperature ranges?

What about auto-tuning algorithms?

[desA]

http://i39.tinypic.com/f9m4i1.png

A simulation using linear response elements & PID control. Dead-time & saturation not yet included in the model. Realistic physical gains to still be inserted & PID tuned accordingly.

Please observe that for very small pressure reference perturbation (see scale in top left scope), the response is both amplified & drifts. Proportional ~ 0.98, I ~ 0.002, D = 0.

Also observe that water massflow change & wall temp change are inversely proportional - linear. Change in refrigerant pressure proportional to change in wall temp.

[Tesla]

Ok desA
No. A single set point is not sufficient for all operating conditions. eg for chillers we use chilled water temp reset to match the load - but we don't always adjust the pid's.
Auto tuning algorithms to take into account all the important parameters takes quite a lot of programing, engineering and analys for the real world practical situations. What is required here is a lot of historical plots of key parameters under differing values, along with algorithms for each. Considering the scan time for most BMS type contolers is 1sec with very limited ram and rom. We may be waiting until the next generation of computing tech such as photons instead of electrons. I know the tech is there but is generally only used for govt/millatary apps. Hence the reason for me to intending to write a programe in basic code for us in HVACR to use as a tool for calculating pid's. Realy when we look at the big picture we are the dumb ones with very limmited tech compared to ? but we have the innovations.
Thanks mf you have motivated me to finnish designing and build an instantanious domestic heatpump - considering you have a heatpump with a COP of 9 (wow, very impressive).

[desA]

Ok desA
No. A single set point is not sufficient for all operating conditions. eg for chillers we use chilled water temp reset to match the load - but we don't always adjust the pid's.

Very good point. So, you are linking the PID reset to a moving parameter - eg. chilled water temp, to provide for the drift in process conditions. Good point - yes.


Auto tuning algorithms to take into account all the important parameters takes quite a lot of programing, engineering and analys for the real world practical situations. What is required here is a lot of historical plots of key parameters under differing values, along with algorithms for each. Considering the scan time for most BMS type contolers is 1sec with very limited ram and rom. We may be waiting until the next generation of computing tech such as photons instead of electrons. I know the tech is there but is generally only used for govt/millatary apps.

I agree. That's why I asked the question. Too much reliance on auto-tuning algorithms in many, many process applications.


Hence the reason for me to intending to write a programe in basic code for us in HVACR to use as a tool for calculating pid's. Realy when we look at the big picture we are the dumb ones with very limmited tech compared to ? but we have the innovations.

Good on you. Put up a few links & we'll help to beta test.


Thanks mf you have motivated me to finnish designing and build an instantanious domestic heatpump - considering you have a heatpump with a COP of 9 (wow, very impressive).

That may very well be one of these eCOP thingiemajiggies. :D

[desA]

http://i41.tinypic.com/2hrjyhu.png

Some odd things happen when time delays, saturation & pressure sensor white noise are added.

Interesting that the system eventually sorts itself out, but with an offset.

[Gary]

http://i41.tinypic.com/2hrjyhu.png

Some odd things happen when time delays, saturation & pressure sensor white noise are added.

Interesting that the system eventually sorts itself out, but with an offset.

At any condition other than design condition, there must always be an offset.

[Gary]

What is the set-point? This is the million $ question, in my mind.


To my mind, the million $ question is... what happened at the 80 minute mark that changed everything?

[desA]

To my mind, the million $ question is... what happened at the 80 minute mark that changed everything?

Wish it were so easy. Actually, the simulations show that, given a random disturbance on the pressure signal, the system instability can stop on its own, given the 'correct' instantaneous value - strange as it may seem.

If the random number generator of the white noise (pressure disturbance) is changed, the correction point can move.

So, what could very well be happening is that the system is being disturbed (noise) on the pressure signal. This then, via perhaps a poorly-tuned PID controller, coupled with saturation & system local transport lags, ends up setting the system into a limit cycle for a period, whereafter it, then moves onto a stable trajectory. This is an uncontrolled system - in chaos.

This is why I ask the question as to whether, in this case, a temperature sensor may not be less noisy than a pressure sensor for such a system?

[Gary]

The object of PID is to minimize the offset. In the Pr chart I would expect to see a large offset decreasing in amplitude. Instead, the offset was limited compared to the later offset (after 80 minutes).

To my mind, it seems unlikely that the oscillations were caused by the water flow/pump control.

[desA]

Fair comments. So we are sure that pump control is not the cause of the above disturbance?

The interesting thing about this head-pressure problem is that it occurred on two tests - around 10 days apart - where the ambient air reached around 10'C, & was not seen on other tests with higher ambients.

There could be a few other issues - but, these did not show at other ambients, only at the 10'C ambient tests:
1. Reversing valve stiction, or malfuncton;
2. Cavitation in the water pipes.

[Gary]

Fair comments. So we are sure that pump control is not the cause of the above disturbance?

Possibly others will see the situation differently.

Lacking hands-on, I'm not sure of anything... lol



The interesting thing about this head-pressure problem is that it occurred on two tests - around 10 days apart - where the ambient air reached around 10'C, & was not seen on other tests with higher ambients.

There could be a few other issues - but, these did not show at other ambients, only at the 10'C ambient tests:
1. Reversing valve stiction, or malfuncton;
2. Cavitation in the water pipes.

Or could be just about anything.

[desA]

Or could be just about anything.

What 'anything' would cause the head pressure to bounce around like this? This is the question.

With a high side so responsive, it would pay to understand what the root cause/s could be.

[Gary]

Speculating about what might be wrong is not what I do, and in fact I am opposed to the practice. I don't like guessing. Its right up there with parts changing. I want to know what is wrong.

That said, if you don't have access to the machine, then all we can do is play guessing games.

Anything could include instrument error, pump cavitation, air bubbles... maybe even intermittant compressor flooding... and who knows what else.

[desA]

Thanks Gary. Let's see how the discussion progresses. Thanks for your help, so far.

With a PID controller, as I remember things, there is a reference junction. When this is either incorrectly configured, or even faulty, the controller may exhibit limit cycle phenomena, amongst other things e.g. drift.

This thought came off the back of Tesla's mention of referencing to chilled water temperature.

[Peter_1]

Also keep in mind that a pressure sensor (transducer) is much more expensive than a temperature sensor.
Or use a temperature sensor somewhere in the 2-phase region at LP and convert it to pressure. Mitsubishi Electric did his on their VRF's

[desA]

^ Thanks Peter. That is an elegant way to do it.

I would think that the temperature sensor may perhaps have a slower response than the pressure sensor & better internal damping - from the pipe wall and thermal mass. This could turn out to be a good thing.