971's Control System

[Ether]

Kevin,

What value have you measured for L for CIM ?

Oh, also: Ke should be V/(rad/sec) or V∙sec/rad no?

[AustinSchuh]

Originally Posted by adciv

That does help, and I look forward to your white paper. Thank you. Gathering the physical parameters is the part I've been trying to figure out, since I don't have access to a system analyzer like I did in college.

For flywheels and stuff like that, I've been starting by recording a step response. I'll then feed the same step response to my model, and plotting the results on top of each other. I then fiddle with the parameters until it matches well enough.

[NotInControl]

Originally Posted by Ether

Kevin,

What value have you measured for L for CIM ?

Oh, also: Ke should be V/(rad/sec) or V∙sec/rad no?

I would need to look up the measured value from my old models, I do not recall off the top of my head what the measured inductance value was. I'll provide it as soon as I can.


As for the units for Ke, you are correct. The units must satisfy this equation:

emf [Volts] = Ke [Volts/(rads/sec)] * AngularVelocity [rads/sec]

Looks like I left off the parentheses in my original post. I will edit my post to add parentheses to make it clear. Thank you for catching the typo.

Regards,
Kevin

[Ether]

Originally Posted by NotInControl

I would need to look up the measured value from my old models, I do not recall off the top of my head what the measured inductance value was. I'll provide it as soon as I can.

If anyone else has any test data from which CIM inductance could be calculated, would you be willing to post it?

Thanks.

[Bryce Paputa]

Sorry to necro this thread, but I haven't had time to look into this stuff in detail until now and I have a few questions. Looking at the 2013 flywheel controller, the state matrix for the model looks like:

Xdot = [ speeddot ] = [ 0     1          ][ speed ] + [ 0      ] U
       [ inputdot ]   [ 0 -kt/kv/(JG^2R) ][ input ]   [ kt/JGR ]

y = [ 1 0 ][ speed ] = speed
           [ input ]

(that's correct, right?)

The part I don't understand is the -kt/kv/(JG^2R) term. That represents how the wheel slows down over time, right? How is/was it derived?

Another question I have is about arms and gravity, when you're modeling an arm like your claws last year do you just ignore gravity in the model and treat it as a disturbance? It doesn't seem like a linear force to me with how it changes with respect to the arm angle and I can't figure out how to get it into a linear state matrix model.

[adciv]

What you could do is add it as a forcing function and cancel it out in your input, along the lines of:
u2 = u1+sin(angle)

[James Kuszmaul]

Originally Posted by Bryce Paputa

Sorry to necro this thread, but I haven't had time to look into this stuff in detail until now and I have a few questions. Looking at the 2013 flywheel controller, the state matrix for the model looks like:

Xdot = [ speeddot ] = [ 0     1          ][ speed ] + [ 0      ] U
       [ inputdot ]   [ 0 -kt/kv/(JG^2R) ][ input ]   [ kt/JGR ]

y = [ 1 0 ][ speed ] = speed
           [ input ]

(that's correct, right?)

The part I don't understand is the -kt/kv/(JG^2R) term. That represents how the wheel slows down over time, right? How is/was it derived?

You are correct; that term does describe how the wheel slows down over time. This comes from the equations that are use to describe an ideal motor. I have attached a pdf with the derivation of the A and B matrices.
Edit: I didn't look carefully enough at your matrices; what I said is correct, but the state matrix (X) is of the form:
[[position]
[velocity]]
for the 2013 shooter and Y would just be the position.

Originally Posted by Bryce Paputa

Another question I have is about arms and gravity, when you're modeling an arm like your claws last year do you just ignore gravity in the model and treat it as a disturbance? It doesn't seem like a linear force to me with how it changes with respect to the arm angle and I can't figure out how to get it into a linear state matrix model.

We did ignore gravity in our final model (we ran some simulations with gravity included and the differences were relatively minor; unfortunately, I don't have any of the graphs on hand). This is also complicated by the fact that gravity's influence on our arm would be non-linear (as you mention) and due to the wide range of motion of the arm, the small angle approximation can not be used.
The way that we handled the constant force was through the use of the delta-u controller (previous posts in this thread should have some information on this), similar to how the I term of a PID controller would handle it.

If you have any more questions, any of us would be happy to answer.

Another Edit:
I realized that no one mentioned this explicitly since the last post, but you can find some more controls documentation and our 2014 source code here.

[Michael Hill]

You make your kids do root-locus plots by hand, right?

*shudders*

[AustinSchuh]

Originally Posted by Michael Hill

You make your kids do root-locus plots by hand, right?

*shudders*

I don't think I've done a root-locus plot since I got out of school. We tend to use the pole placement tools (and LQR and a bit of practice/intuition) to do the final tuning. I never ask the kids to do anything I wouldn't be willing to do myself.

[Michael Hill]

Originally Posted by AustinSchuh

I don't think I've done a root-locus plot since I got out of school. We tend to use the pole placement tools (and LQR and a bit of practice/intuition) to do the final tuning. I never ask the kids to do anything I wouldn't be willing to do myself.

I remember loathing any assignment requiring root-locus plots. I never really got much out of them anyway (maybe it was just how it was presented). I'm pretty sure I learned more about the control system while making the root-locus plot than anything the plot actually told me when it was done. I've wondered why more teams haven't done LQR or lead-lag compensators. I seem to remember a good LQR was somewhat easy to implement if you knew the dynamics of your system pretty well, but I believe PID is probably more forgiving if you don't fully understand it. Is there any truth to that?