What is the standard overshot for a servo?

[InsaneEPP]

I am deciding on what I want to drive and control my servos and was looking at the Rutex site and came across this page:

http://www.rutex.com/us/cart.php?tar...&category_id=4

It is showing a servo overshot of 295 steps. I don't know much about the other info on the graph and I doubt this servo is optimally tuned, but I was thinking that 295 steps would be WAY off and you wouldn't want this in normal practice. Gecko throws a fault when the servo is off 125 steps from where it should be. That also seems like a lot. If you were using a 5 tpi screw and a 2000 steps per rev encoder, 125 steps would be .0125 inches off. I guess I'm trying to find out what the typical error is. It doesn't make sense to make a machine within .001" with precision ball screws and slides if the servos aren't that accurate. Thoughts?

Andreas

[arvidb]

That curve is a step response curve of motor speed. No motor can respond to that without lots of error (no motor can accelerate instantly).

A CNC controller would keep acceleration and speed within reasonable limits so that the motors have a chance to do what they're told. Then you won't get such a big following error.

Step response curves are useful to set the PID controller parameters, though.

I'm not sure how big following errors one can expect from a well tuned real life machine. If anyone knows, it would be interesting to know.

Arvid

[HuFlungDung]

Arvid is right. That graph is not typical of the final tuned system. That's just an example of what you would see while fiddling with the PID settings, and Acc/Dec ranges.

[ynneb]

I guess there are two areas that you can use to tune your servos. One is by tuning the driver cards as best as you can. The second is adjusting the acceleration and deceleration of the servos (software). Obviously with a smoother acceleration and deceleration you will avoid overrun.

As stated b4, it is an impossibility for a motor to reach maximum speed in an instant, and also stop completely in an instant. I think that is what the tuning software is trying to do, and it reports on those stats.

Is this correct fellers ? This is my current newbie understanding ?

[Mariss Freimanis]

Benny,

On the money.

Mariss

[mmjpotter]

A correctly tuned servo system should hold position to within +/- 1 encoder pulse. The following error will vary with the required velocity but an overshoot of more than 1 encoder pulse is not acceptable.

[InsaneEPP]

Thank you for all the info guys. I think I understand it now. Basically the servo will never be at the position it is supposed to, but it should be close. There is some lag and that is ok, because all the axises are lagging the same amount of time. It's good to hear that the overshoot is 0 to 1 pulses.

[ESjaavik]

I think +-1 pulse is a bit optimistic. The PID controller uses the position error to adjust position. The output to the motor is the error multiplied with P. If the error is just 1 then P have to be large to get a current to the motor that will actually move it (and the axis). But then the output will "go through the roof" if the error is larger, and the machine becomes unstable. Think of it as the steering of a gokart having a high P, while a van with too low air pressure in the tires have a low P. So the P should not be too high, or you'll just weave about.

But, then we will have a residual error of some counts that will not be eliminated because the force is not high enough. Enter the I factor. If a small error persists over time I will increase the output to the motor and thus try to get "exactly there". But it will take some time (relatively speaking). So increasing the I factor will improve the final position. Still there must be an error for I to add it up. It's like after rounding a corner with your gocart you will not immediately go in a straight line, but your heading will very soon be fine as you correct it.

The D factor acts as the damper and can be used to bring down the overshoot and hunting back and forth until finding the correct position. It has the same purpose as the dampers on your car. If they are not working, your car will keep on going up and down after hitting a pothole or a bump.

So to make it easy, just set all of them P I and D to a high number?
Sorry, if it was only that easy. That's why you need to tune these numbers on your machine with your motors and your drives. And you'll never get to zero error because the measured error is the basis upon which all this works.

Now you see why some servos have a high number of pulses (or other discrete steps) per revolution (up to 128000). Because then the measured error is quite large even for a small error on the motor (and machine).

So based on this, the error is a result of several factors where the ability to tune in on the best P I and D is important. And the resolution of the encoders is one factor that sets the limit where it can be said no additional tuning will improve it.

[InsaneEPP]

Esjaavik,

Thank you for your detailed post. There is a lot of good information there. My question after reading this is: does it matter that there is some error? Follow my logic. For example: If you had a one axis machine, your controller would tell the servo driver to go to position 100 (in encoder pulses) and it is currently at position 0. The servo would take some time to get there, but would eventually get there +/- 1 encoder position. During the movement of the servo, there is error (up to 100 pulses). The error does not matter, however because it is moving from one position to another, which is just what you want. If you move to a two axis system, it is the same except the two axises need to move in concert with the amount of movement being proportional throughout the whole of each movement command. Usually, the controller sends single step moves. For example, both axises are at 0 are are both commanded to go to 100. Each must take a step at the same time. The contoller will send a position of 1 to both drivers, and then a 2. The speed at which it does this is setup in the controller. As long as the interval is greater than the amount of time needed to accellerate, the servos will move together. One servo may accellerate faster than the other, in which case it will arrive at the position quicker, but it will have to wait for the next controller input to move again. There will always be a lag between the controller command and the servo, but as long as the two servos are moving proportionally and the accelleration is low enough, there shouldn't be a problem. Am I off here, or does this make sense?

[mmjpotter]

InsaneEpp

Sounds O.K to me. As the axis approaches the target the error becomes less, providing automatic deccelleration until the error is a minimum 1 pulse. Therefore the axis should not overshoot.

Esjaavik. +/- 1 encoder pulse has to be realistic or the machine will not have any accuracy. The PID loop gain is set to provide enough current to drive the machine axis to ZERO, but I agree we have to limit the MAX error. The key to this is the servo loop time constant or error sample rate. The time constant can be set (depending on the application and resolution) from say 1 to 300 milliseconds and the max error will be the distance travelled in this time so we can limit the error. (max error x loop gain= max velocity). This is a very basic example but I hope it explains why the velocity will not go through the roof.

[ESjaavik]

@insaneEPP: It makes sense. But what you describe with the 2 axis is a diagonal line. If they don't move synchronously the line will not be a diagonal while you meant it to be. If one axis arrives there before the other, it will be a diagonal with a "knee".

And Arvid touched the clue: you don't ask a machine axis to go to the endpoint abruptly except for tuning purposes. You feed the path into it nice and including the acceleration and retardation ramps. So it should ideally all the way be a small enough change that the error does not become large at any time during the travel. In a real machine (say a mill) it may not necessarily stop when "getting to the endpoint". It will continue into the next movement with only as much speed change as necessary. If I stick to my analogy with the car, you don't stop at each turn of a country road, you just change direction. And the speed at which you move north/south will only become zero when heading straight east or west.

Regarding your original question, I don't think an answer can be given. Yes, I also found that frustrating, but there are so many variables. For what it's worth, in a test setup I did where one step from the controller was equal to one count from the encoder I could not get better than +-6 without there being unstable states where full control was lost. But this was my setup, your mileage may vary.

@mike: I don't agree with you that it will not have any accuracy if error > 1. The MAX error is a limit we set that will occur if the system is badly tuned, or the controller asks so much more of the servo system that it can deliver, that we want it to stop trying, or returning a message to the controller. The servo system can keep within 1 step of the controller if it has an internal resolution better than 1 step. But not if the step size equals the feedback (encoder) from the motor shaft.

[arvidb]

Originally Posted by mmjpotter *snip* Esjaavik. +/- 1 encoder pulse has to be realistic or the machine will not have any accuracy. The PID loop gain is set to provide enough current to drive the machine axis to ZERO, but I agree we have to limit the MAX error. The key to this is the servo loop time constant or error sample rate. The time constant can be set (depending on the application and resolution) from say 1 to 300 milliseconds and the max error will be the distance travelled in this time so we can limit the error. (max error x loop gain= max velocity). This is a very basic example but I hope it explains why the velocity will not go through the roof.

I agree with ESjaavik that one encoder pulse is not a realistic maximum following error. This would give the input (error) signal a very low resolution (either no error or full error) and the controller will have a very difficult time to react normally to this. My guess is you would get oscillations.

The machine will have a well defined accuracy even if the error is (for example) up to 10 pulses, it's just 10 times worse than if the error was only one pulse. My machine will have 2048 PPR encoders, giving a resolution of <1 micron. So even with 10-20 pulses off, the limiting factor will be rigidity of the machine etc.

I'm not really sure what you're saying about error sampling rate. The sample rate should be high enough to approximate a continuous controller (I read >10 times the bandwidth of the system). As long as that is fulfilled I guess it should not affect the controller?

I don't understand what you write about the max error being the distance travelled in the time constant time. Can you elaborate?

Arvid