21.07.2025, 22:40
Changer la langue
S’inscrire
Connexion
Vous n’êtes pas connecté.
RBMDragon
Junior Member
vendredi 4 mars 2022, 20:53
Changed the variable factor by 100.
The reponse is not as agressive or fast, but it seems to handle a Heaven Benchmark.
Step Response 90 is the data I recorded for the PID value calculation.
PIDLoopSimulator was used to estimate a controller reponse, although the loop is still much slower.
I used the guides from Engineers-Excel.
The gain was the factor between the change/full range, i.e. 1.
The 336 seconds is the time it takes for the temperature change starting to 63,2% of the final value.
The delay is the time it takes from the 15%-90% step change to notice a temperature change in the temperature.
PIDValues are the factors I inserted.
Setting Integral and Derivative to 0.
Set the Controller Gain until the simulator oscillates (not getting stronger or weaker).
Set the Controller Gain to half that value.
Set the Integral value until the Process value gets as close to 1 (I took the value at 100 seconds)
Set the Derivative value to reduce overshoot (second sheet take maximum value from the Process values)
PIDControl is the controller in action while running Heaven Benchmark on the GPU.
There are still some things I might do better, but for now I'm starting to get progress.
The reponse is not as agressive or fast, but it seems to handle a Heaven Benchmark.
Step Response 90 is the data I recorded for the PID value calculation.
PIDLoopSimulator was used to estimate a controller reponse, although the loop is still much slower.
I used the guides from Engineers-Excel.
The gain was the factor between the change/full range, i.e. 1.
The 336 seconds is the time it takes for the temperature change starting to 63,2% of the final value.
The delay is the time it takes from the 15%-90% step change to notice a temperature change in the temperature.
PIDValues are the factors I inserted.
Setting Integral and Derivative to 0.
Set the Controller Gain until the simulator oscillates (not getting stronger or weaker).
Set the Controller Gain to half that value.
Set the Integral value until the Process value gets as close to 1 (I took the value at 100 seconds)
Set the Derivative value to reduce overshoot (second sheet take maximum value from the Process values)
PIDControl is the controller in action while running Heaven Benchmark on the GPU.
There are still some things I might do better, but for now I'm starting to get progress.
Speedy-VI
Senior Member
dimanche 6 mars 2022, 02:27
Well AQS does provide the ability to manually enter values for Factors P. I, D, Hysteresis, and a Reset Time for Factor D if you select Target Temperature mode and then User Defined. I found that if you select one of the presets first (Fastest, Faster, Normal, Slower, Slowest) and then User Defined, you can see what values AQS sets for each preset. I guess you could try them all and graph the resulting curve for each one. If you still felt that all of these presets were not acceptable, once you see how the values affect the curve you could make an educated guess of what values to enter to get whatever response you are after. I think OP is saying that with the PID Loop Simulator he should be able to predict the outcome of entering these values. Then if you know exactly what response you want you can calculate the values to enter to get that response. I may be confused about this. It's interesting, but I have found the curves AQS provides are sufficient for my needs. This is one of the many features in AQS that you can tinker with, and I guess OP is compelled to tinker with it. I am curious how this will develop.i asked because from the curves i see, it seems you're basically recreating the existing. By controlling the fans from water temperature, or the delta between ambient and water temperatures (quieter if you have temperature swings in the room), you get to the same result.
WinstonWoof
Full Member
dimanche 6 mars 2022, 08:48
I approached things a slightly different way using virtual sensors.
For my Water Pump Control I have a Virtual Sensor configures so that it outputs one of two values dependent on load & time.
The lower value is the default and runs the pump at just under 3k rpm.(based on its very simple fan curve with no tweaks)
The higher value runs the pump at its max speed of ~ 4.5k rpm
The Higher speed is triggered
IF either Average CPU/GPU load as measured over 15 seconds goes above threshold (80%) And it then stays above this for a further defined period (120 seconds) THEN use High Value.
In this scenario I also have a 60 second delay on shut off(which actually I need to add in for the next scenario as welll now I think about it)
If However the Average of both CPU & GPU go above threshold then High Value is used straight away.
WaterPump_Sensor.jpg
For Fan Control I have something similar but with a twist.
In this case it uses the H2O delta as a Normal value and a first fixed value (Temp Output) as the standard High.
This then runs the Rad Fans & Case fans on their own curves which are configured so that the Rad fans are more sensitive (ie can run from 0-100% )than the case fans which always run at pretty much a fixed speed based on their curve (my rads are mounted externally).
The next added twst is that once an hour for 5 minutes all fans (inlduing case fans) are spun up to 100% by setting a Max value (100) to be used as the output temp.
Fan_Curve_Sensor.jpg
For my Water Pump Control I have a Virtual Sensor configures so that it outputs one of two values dependent on load & time.
The lower value is the default and runs the pump at just under 3k rpm.(based on its very simple fan curve with no tweaks)
The higher value runs the pump at its max speed of ~ 4.5k rpm
The Higher speed is triggered
IF either Average CPU/GPU load as measured over 15 seconds goes above threshold (80%) And it then stays above this for a further defined period (120 seconds) THEN use High Value.
In this scenario I also have a 60 second delay on shut off(which actually I need to add in for the next scenario as welll now I think about it)
If However the Average of both CPU & GPU go above threshold then High Value is used straight away.
WaterPump_Sensor.jpg
For Fan Control I have something similar but with a twist.
In this case it uses the H2O delta as a Normal value and a first fixed value (Temp Output) as the standard High.
This then runs the Rad Fans & Case fans on their own curves which are configured so that the Rad fans are more sensitive (ie can run from 0-100% )than the case fans which always run at pretty much a fixed speed based on their curve (my rads are mounted externally).
The next added twst is that once an hour for 5 minutes all fans (inlduing case fans) are spun up to 100% by setting a Max value (100) to be used as the output temp.
Fan_Curve_Sensor.jpg
Ce message a été modifié 2 fois. Dernière modification effectuée par "WinstonWoof" (6 mars 2022, 08:51)
WinstonWoof
Full Member
dimanche 6 mars 2022, 09:00
In this scenario I also have a 60 second delay on shut off(which actually I need to add in for the next scenario as welll now I think about it)
Quickly modified them to the Delay Off applies evenly
WaterPump_Sensor.jpg
Fan_Curve_Sensor.jpg
Ce message a été modifié 1 fois. Dernière modification effectuée par "WinstonWoof" (7 mars 2022, 08:15)
ronzino
Newbie
lundi 13 juin 2022, 01:39
It is not clear to me neither in the Engineers-Excel guide nor in your calculations why Gain=1.
"I used the guides from Engineers-Excel.
The gain was the factor between the change/full range, i.e. 1."
In you calculation the gain should be G=deltaPV/deltaMV so (25,6-35,1)/(90-0) = -0,1.
as well as in the web site mentioned it should be (110-100) / (55-50) =2
Speedy-VI
Senior Member
samedi 11 février 2023, 01:26
This is concerning. Can you elaborate a bit more on your findings?After a lot of hours spent to study again PIDs, and hours spent to make a step response of my system, I have figured out that Aqaucomputer had chooesen a PID gains rapresentation with non pratical sense.
K gains or T have a standard representation - form. Aquacomputer chose to adopt its own convenction for those coefficent (they cannot be negative) so it is impossible to a real PID expert to calculate the proper coefficent of its system just beacuse the aquacomputer coefficent representation are custom and not documented.
So don't waste your time with equations and step response..... it will lead you to a death end path.
Discussions similaires
-
Software, allgemein »-
NFS:U2 - tuning
(9 novembre 2005, 17:17)
-
- Off-Topic »
-
Pimp my Ride
(1 août 2004, 22:18)
-
- Off-Topic »
-
Fotostory vom Einbau eines 57i Performance Kits
(29 juillet 2003, 02:54)
-
- Off-Topic »
-
auto gefunden!
(22 juillet 2003, 02:16)
-
- Off-Topic »
-
Krasses Gerät
(12 juin 2003, 19:38)