# pidtuner

## Forum Replies Created

Viewing 15 posts - 1 through 15 (of 18 total)
• Author
Posts
• in reply to: Model identification algorithm #90
pidtuner
Keymaster

Hi,

Thank you for your interest. The identification algorithm uses this solution:

https://math.stackexchange.com/questions/1428566/fit-sum-of-exponentials/3808325#3808325

Basically consists in integrating the inputs and outputs of the system (according to the desired model order) and then solving a couple of (linear) least squares problems.

The delay in nonlinear, so given that the (linear) least squares problems are “relatively cheap”, the delay is calculated by brute force initially using a grid of values. Then refinement is made using a simple newton method.

in reply to: Scale gains #87
pidtuner
Keymaster

Hi,
I am happy you like the tool, the “Scale Gains” slider simply implements the Skogestad PID tuning rules.

Cheers.

in reply to: Final values of the tuner #84
pidtuner
Keymaster

Hi,

Yes, they are indeed Ki, Kd and Kp, from the “parallel” or “ideal” form:

https://en.wikipedia.org/wiki/PID_controller#Standard_versus_parallel_(ideal)_form

Bear in mind that in most PID implementations, the “standard form” is the most common (the one that uses Ti and Td).

in reply to: Weird results #81
pidtuner
Keymaster

Apart from minimizing disturbances in the data, and making sure time is seconds, also make sure that the data must be the exact same values that go into the PID (process output) and out of the PID (process input).

I cannot stress this more, it is a very common mistake when using the tool, the data must be exactly what would come in and out of the PID block (of course in open-loop PID is disabled, but this is just to illustrate which are the signals that are needed).

Have a good tuning!

in reply to: Scale of things #78
pidtuner
Keymaster

Hi, all time must be in seconds, always. The scale of the axis must the exact same that is used to feed in and read out of the PID block that will be used in your process loop.

in reply to: Weird results #77
pidtuner
Keymaster

Hi, yes all time must be in seconds. Your input step data is definitively weird.

It seems you are not introducing an “clean” open loop step response of your process. This can clearly be seen because when your input is zero, your process is responding to something, let’s call it “unmeasured disturbance

What is even more worrying is that before the first step, the unmeasured disturbance seem to be affecting the process with a positive slope, and after the first step, it is affecting with a negative slope. So it not a disturbance that can be de-trended.

To get a good model of your process, you need to record a “clean” open loop step (without external disturbances). If there are external forces driving your system, it will be very hard to obtain an accurate description on how your process input drives the process output.

That being said, this does not mean the pidtuner will not be useful to you. What it means is that you should be careful and realize that the simulation given by it will not be an accurate one, because of the unclean data you gave to it.

If you have no way of performing a “clean” open loop step response experiment, I would use this data, but then de-tune the gains given by the pidtuner (slider to the left) and use it as a starting point for tuning. Then increase the performance slowly until satisfactory results are obtained. This is the price of uncertainty in control.

#69
pidtuner
Keymaster

Hi,

I have been thinking of adding this option in the simulation. Will probably add it in the near future as soon as I have the time.

Regardless, as long as your PID implementation has anti-windup mechanism, the PID tuning given by the tool should give you a very good starting point in spite of the limits. Then you can just fine tune online to achieve the desired response.

Thanks for the feedback! Let me know if you have more questions.

in reply to: What PID model is the best? #66
pidtuner
Keymaster

Hi Jacob,

If you really want to understand how a PID is implemented, I highly recommend you to read the following blog series:

Improving the Beginner’s PID – Introduction

They are short and well explained, also is the implementation used in the pidtuner tool.

Regarding the PID parameters, some PID implementations use PID gains directly (so called parallel or ideal form), other use times (so called standard form). That is why the tool provides both, then the user must use the ones that their PID requires. Read:

https://en.wikipedia.org/wiki/PID_controller#Standard_versus_parallel_(ideal)_form

in reply to: PID Tuner help tuning my system #64
pidtuner
Keymaster

Hi,

Sorry for the delay. Was offline for holidays. I tried to load your project, but the data saved seems to be the demo data. Can you please load the data again to that project and save it again? Else you can post the data here directly.

Cheers.

pidtuner
Keymaster

Then your test data is still incorrect, what you need is to do is:

• Wait until the process output has settled in one operating point
• Apply one single step in the process input.
• Wait until the process output settles again into another operating point.

The data you shared is not useful, because you make many small changes in the input and do not let the output settle. What we need is one single input step and record how the output settles from one stable value into another one caused by the input change.

pidtuner
Keymaster

• Your data is undersampled. You need to sample faster so the process output response dynamics due to a step change in the process input are clearly visible.
• Your output process curve does not seem to be correlated to your process output curve. I see no correlation between the input of the process and the output. Are you sure you have selected the correct variables in your data.

Take a look at the sample data provided by the tool, you can see a clear correlation between a step change in the input and a slow transitioning on the output to another level. This is what you should be able to see on your data.

Also make sure your PID is turned off during the experiment and that your process input is exactly what would get out of your PID and the process output is exactly what would get in to the PID.

pidtuner
Keymaster

Then go and convert your time values to seconds. Then fill up the table in the pidtuner, placing the values on each corresponding column. Once the table is complete, verify the charts display the data as expected to confirm that you introduced the data correctly.

Then click the “Save” button and copy the link that the pidtuner shows you. Paste that link here and then I can help you move forward.

pidtuner
Keymaster

It should always be in seconds, and as time goes, it should always be increasing. To see an example, simply load the test data included in the tool. See the tutorials as well:

in reply to: Differential drive PID tuning #52
pidtuner
Keymaster

Assuming you have setup your hardware correctly, the PID controller does not care about the details of all of it. To tune the PID controller, we just need a simplified “model” for everything that is outside of the PID controller. So for a moment, please forget about your hardware details, and think as if you were the PID controller and know nothing about what is outside.

We are the PID controller, and to know what gains to use, we will try to guess what it is that we need to control. So we will try to obtain that simplified “model” of the outside. The input of such “model” is the output of the PID controller, and the output of the “model” is what we feed as input to the PID controller. In your specific case the input of the “model” is the what gets out of the PID controller, so the TurnSpeed. The output of the “model” is what we are trying to regulate or steer, which your case is the GPSHeading.

I understand that you also use a “Setpoint” (what you call “Aim”) to form the error that is then fed to the PID controller, but we will forget about that “Aim” as well because the “Aim” does not get affected by a change of the TurnSpeed, only the GPSHeading is affected by the TurnSpeed, so the “Aim” is like an external disturbance and is not relevant for our model, because we are only interested in what we can actually control.

So we are the PID controller, how to we model everything that is outside us? Well, we do as any person would do. We make experiments and see what happens. Since the TurnSpeed is the only variable we can manipulate, me make a small change to it and see what happens. We go a little bit up and little bit down and we see how the GPSHeading responds, because that is the variable we want to regulate.

Control theory tells us that if we design that experiment to be a step change on the input (TurnSpeed), we get enough information to create a “model” that is good enough for control (good enough to make a good PID controller tuning).

And this is what the pidtuner tool helps you to do. It takes your step response experiment and makes a “model” of everything that is outside of the PID controller, and then uses that “model” to give you some PID gains that will be good enough to regulate that “model”.

So you give your data to the pidtuner in Step 1 of the tool, then on Step 2 you select a time range in your data where you can select where a step change in the input (TurnSpeed) occurs, so when your click next, the pidtuner can generate some models for you.

Notice that I said before; “to regulate that model”, meaning that the “model” has to resemble reality somehow, for the PID tuning to be successful. That is what you see in the Step 3 of the tool where it says “Select Model”. It compares the output generated by the models that the pidtuner has created for you, against the actual data of your experiment. Your job in that step is to select a model that resembles the actual data. The closest the model output looks like the data, the better. It does not has to be perfect though, there is some room for error in the gain, as long as the relevant dynamics are captured.

If no model resembles your data then you have to get back to any of the previous Steps. Either go back to Step 2 and select a different time range that contains better information about the step response. Or even go to Step 1 and import data from a better experiment. This is what I proposed to you, to make a better experiment by making manual steps in the TurnSpeed. If you want to see how a good experiment looks, take a look at the sample data that the pidtuner has by default. That data is actually from a drone pitch angle response.

Once you have a good model-data match, you can get to Step 4 and get the PID gains that you need. In Step 4 you can simulate how the “model” would respond to a change on the “Setpoint” (for you is the “Aim”). You can use the “Scale Gains” slider to make the PID gains more or less aggressive. Then you can see how much time it would take for the model to reach the desired “Setpoint” (“Aim”). Beware that the more aggressive you tune the PID gains, the less robust it will be. That can be seen in the Bode Plot that is shown also in Step 4, but understanding robustness trade-off against performance and how it relates to the Bode Plot is too complicated to explain here and requires university level of engineering studies. You can get a grasp of it by watching some youtube videos though. If you are interested, checkout the youtube channel of Brian Douglas:

But you really don’t need all that advanced knowledge just to tune your PID, just follow the instructions I gave you for the tool and you should be good to go.

in reply to: Differential drive PID tuning #47
pidtuner
Keymaster

Hi,

The input to the pidtuner should be the “TurnSpeed” and the output should be the “GPS Heading”. I tried to import your data to the pidtuner using the “Loop Timer” as the time, but divided by 1000 to get the seconds.

Sadly your data does not contain clean manual steps on the input (TurnSpeed). These should be performed in open-loop (PID turned off). Some steps up and down while recording the output (GPD Heading).

Still the pidtuner gave somewhat of a match, see the following link:

https://pidtuner.com/#/0eDJOt49JM

I would try some conservative PID gains first like the ones shown in the link, then if it works, move the slider right to mak the PID gains gradually more agressive.

Still the bets thing to do would be to repeat the experiment, providing some clean steps on the “TurnSpeed”.

Viewing 15 posts - 1 through 15 (of 18 total)