New Issue for Ged's email

[hardillb]

Main Issues

1. Fire Interlock is working the wrong way around. This should be a True or Boolean 1 to allow the loop to run. Fire is a normally closed input when it is OK.

2. The integral action does not work properly as far as I can test. The Gain, when set to 10, with a deviation of 1.5 (Setpoint 21, Temperature 19.5), results in an output of 15 with no integral action. The integral action needs to operate as per spec. I tried many values in this box but it certainly does not appear to represent the values shown in the above spec. I expect to be able to put in a value of say 300(s) and have the 15% output added over a 300s period to the output, resulting in an output of 30% after that 300s. It will continue to add that amount until the deviation goes to zero, where the output is held, and when the deviation goes negative, the same calculation is used to deduct from the output until the output gets to 0.

3. I do not understand why we need to rely on the polling rate of the temperature input to set the reschedule time of the loop, when this, in my view, as per spec, should be fixed inside the loop with an adjustment or even just simply fixed in code at 5s. The guy from Wales who has written his own loop has done this and this is as it is described in the spec.

/cc @davidoc85 @gedtyrrell

[hardillb]

Updates for 1 & 3 pushed, please test as soon as possible.

I will have to spend some more time looking at 2

[hardillb]

@gedtyrrell for 2
We need to clarify terms here when you say Gain, do you mean proportional value?

[gedtyrrell]

Hi Thanks for 1&3. Yes Gain is another term on PB. PB of 10 means 10 units of deviation between input and Setpoint gives 100% output; Degc in our case. Expressed as Gain it's just easier to understand since Gain of 10 is 10% output for every 1 unit of deviation so temp is 20, Setpoint is 21 with Gain 10 you get 10% on the output. What we see is no matter what integral action time is set the output is going straight to 100 if input is less than Setpoint. It's explains this in the spec. This is why you have to have a fixed reschedule or recalculation time, typically 5s so that you only increment the output slightly at each recalculation interval. I think your using the input polling time for this but it could be fixed in the node too. Hope that helps. Ged Tyrrell Managing Director Tyrrell Systems Ltd m: +447779 712538 w: +44845 0450171 e: ged.Tyrrell@tyrrellsystems.com<mailto:ged.Tyrrell@tyrrellsystems.com> i: www.tyrrellsystems.com<http://www.tyrrellsystems.com> On 24 Nov 2016, at 10:49, Ben Hardill <notifications@github.com<mailto:notifications@github.com>> wrote: @gedtyrrell<https://github.com/gedtyrrell> for 2 We need to clarify terms here when you say Gain, do you mean proportional value? - You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<#15 (comment)>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AQexbLewOk58-ei7rWhw0Gs5mkGT-AFcks5rBWuIgaJpZM4K7eI4>.

[hardillb]

Just checked in a new version of the PID loop that hopefully is a lot closer to the spec.

Please delete any current instances of the node in your flow and recreate them after upgrading.

I still need to add some deadband stuff, but it should be closer to what I think we want. Output is now 0-10 rounded to 3 decimal places rather than 0-100 previously.

[hardillb]

@davidoc85 new update with deadband in place. Please test as soon as you can.

Please delete any current instances of the node in your flow and recreate them after upgrading.

I still need to add some deadband stuff, but it should be closer to what I think we want. Output is now 0-10 rounded to 3 decimal places rather than 0-100 previously.

[hardillb]

@davidoc85 @gedtyrrell Any feedback on these updates?

[gedtyrrell]

Hello Ben I have been busy all day sorry and David is away. I have tested the loop again and it is a lot better but still needs some work. I have taken some screen shots so you should be able to recreate this. Shot1) I injected a setpoint of 21. The headband was set to 1. I injected a temperature of 21.5 but it started to add integral action straight away, yet there should be no action until the temperature goes above the setpoint + deadband. Shot 2) I injected setpoint 21 and temperature 21. Outputs correctly stay at 0. Then I injected 20. The integral action was added to the calculated proportional output. I then injected 21 again. It jumped very sharply on first injection to 20 from 30. See shot 3. I then injected 21.5 over and over again and the integral action wound down on the heating loop but when it got to zero, it then started to add it to the cooling loop, which should not happen when there is a deadband. I then injected 19.5 and the heating output shot up to 65. It is also hard to know what the Integral values mean. They are normally a period of seconds over which the output is added over, being added in increments of the same across the period each time it is recalculated. Finally, the fire input should be a 1 for normal, going off to disable the loop so it is fail-safe. I hope this input helps. Thanks [cid:10668183-E4BD-452D-AC47-B789D9C9F00E][cid:159B596E-F1D0-42B1-9ACE-97B38D1CC9EE][cid:4EE49DA6-1E93-4564-AECD-320E395819AE][cid:AE12CE6B-4AE4-4545-8F8F-5D52F6BF05C8][cid:D5CC2B6C-ABFF-49A0-9A3C-C66C3F288A2E][cid:6433E1EE-E897-4534-A119-86EA003F31C9] Ged Ged Tyrrell Managing Director Tyrrell Systems Ltd t: +441942 732819 m: +447779 712538 f: +44845 8056915 i: www.tyrrellsystems.com<http://www.tyrrellsystems.com> e: ged.tyrrell@tyrrellsystems.com<mailto:ged.tyrrell@tyrrellsystems.com> View our extended services on our new website. Call today to arrange a meeting to discuss any requirements you may have. On 28 Nov 2016, at 17:15, Ben Hardill <notifications@github.com<mailto:notifications@github.com>> wrote: @davidoc85<https://github.com/davidoc85> @gedtyrrell<https://github.com/gedtyrrell> Any feedback on these updates? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub<#15 (comment)>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AQexbCn8gjGip_A-YRtlIWsE4x9vrPF2ks5rCwxLgaJpZM4K7eI4>.

[hardillb]

@gedtyrrell I can't see any of the images you attached, github seams to have stripped them. Please log into github and add them to the issue directly as I think they will always be stripped when emailed in.

But also I don't think you have the latest version of the code running for several reasons.

1. I've just checked and the current version of the code had the deadband configuration commented out so there was no way to set it as it's not in the config dialogue. I have just updated the repo, with the latest.

2. the Fire control works exactly as you've described. when you send true/1 to the system it runs, when you send false/0 it pushes the outputs to zero. It defaults to true at the start

Please make sure you remove the old version from the flow and re-add it. You should also make sure you refresh the Node-RED editor in the web browser as well as restarting the node-red instance.

[davidoc85]

Hi Ben,

Back for testing now. Did a few tests and I think it just needs polishing:
start
Injected a set point of 21, temperature of 21 with the following values:
deadband=0.5
Gain=10
Ti=250 (seconds)
Recalculation time=5 (secs)
The cooling and heating output went to 0.

1. With the same settings, injected a temp value of 16.
   Output went straight to 5 for heating and left it there for a while until the output was maxed out at 10. Ok so far, then I injected a temperature slowly (on a 0.5 basis) to decrease the output and it went down to 7.5 when injecting 18.5, then it stayed at 7.5 all the time i.e. it would not increment any more.

• -->the integral stopped working, has this been implemented in the code that after a certain amount of time the integral can lock out?

  Kept increasing the temperature and when I reached set point, the output remained at 5
  When I increased the temperature step by step over a ~40s period to 24 the heating output went down to 1.3 and kept decreasing until it reached 0 and let the cooling output to increase up to 3.1. Regularly decreased the temperature and eventually set point was reached and the heating output was 0.49 at the stage while this occurred and remained at 0.49.

• -->Then I disabled the loop by injecting a false (fixed is set to 0) and when I re-enabled the green fixed 0 was still showing in the node instead of the set point. When will this be cleared?
  Tested with fire and this worked OK.

2 )Tested with Gain only in a similar way. When increasing the value and reached set point the heating is showing 0 and cooling is showing NAN. Injected up to 24 and cooling is still showing NAN.

• Overall the pid-node looks miles better. We can now

• fix and clarify why integral stopped working while increasing to setpoint

• fix and clarify issue with cooling giving out "NAN" while increasing and reaching setpoint

• fix "fixed" value tag staying at "green fixed"

• add the Derivative

Please find the export code attached:

pid test.txt

[hardillb]

• The integral was capped at half the max output so when the difference between the input and setpoint was small enough it would stop the system going to max output. I've put it back so the integral can be up to the full max output.

• fixed status message will now revert back to showing the setpoint when it's disabled.

• I need to work out what's going on to give NaN output. Can you clarify a little more about how you got into that state please. What config values did you use and what input values did you inject?

• The derivative is not in the spec document.

[davidoc85]

Injected a set point of 21, temperature of 21 with the following values:
deadband=0.5
Gain=10
Ti=0 (seconds)
Recalculation time=5 (secs)
The cooling and heating output went to 0.
Injected a temp value of 16.
Output went straight to 5 for heating and left it there for a while, then I injected a temperature slowly (on a 0.5 basis) to decrease the output and it went down to 7.5 when injecting 18.5, then it stayed at 7.5.
Kept increasing the temperature and when and when got to setpoint, the cooling started to show NaN and the heating was 0.

[hardillb]

OK, that makes sense, Ti can not be zero, as it ends up on the as the bottom of a fraction so you end up with infinity out the other end.

[davidoc85]

How can we use then a gain only value? In order to tune the pid it is important to have this functionality to see the oscillation level. Can a check be added to see if value is zero, then use gain only part of the algorithm. David

…

On 1 Dec 2016, at 15:04, Ben Hardill ***@***.***> wrote: OK, that makes sense, Ti can not be zero, as it ends up on the as the bottom of a fraction so you end up with infinity out the other end. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or mute the thread.