High Precision Control of linear motion

[spmvoss]

I am currently working on a controller that aims to control a cart that moves 200mm in x-direction. It is driven by a linear motor that is connected to a geared belt. The belt is connected to a cart (0.5kg) that slides on a rail. The cart has to make a cycle of 200mm in 0.9 seconds. It takes 0.05s to accelerate, then travels at constant speed until it reaches the end, decelerates and does the same thing on its way back. During the constant speed an accuracy of 10micron has to be achieved. I am currently stuck at 30 micron and am left with a high frequency vibration. I am using a PID controller and feed forward. The crossover frequency can not be raised due to stability. See the attached figure for the error.

Any ideas on how to remove the vibration? I was thinking it might be due to possible vibrations of the belt.

Thanks for taking your time!

 

[amanuensis]

Why don't you try to start with an FFT of your signal ?
It seems that the signal shows two phenomenons (one at low frequency and one at the high frequency)

 

[btrueblood]

IS the linear motor a stepping motor? Does the frequency of the 30 micron noise equal the velocity divided by the step size? If so, that is a clue. But I don't think it is fixable, unless you can decrease the step increment without reducing the peak velocity attainable.

If it's belt vibration, try a wider (stiffer) belt.

 

[BrianE22]

You can count the number of cycles from your plot. Looks like the low frequency has a period of about .05 sec (20 hertz). Does that correspond to anything? Maybe the rotational speed? I've found that in servo systems the two pulleys need to be aligned (square to each other) carefully.

 

[amanuensis]

20Hz, it's interesting...

Considering the system belt-cart as a spring-mass system, then

frequency*2*Pi=SQRT(Spring/Mass)

So 20*2*pi=SQRT(Spring/0.5)

Spring = around 126²/2 = around 7900 N/m

I don't have enough gumption to say anything about it...

 

[BrianE22]

I was looking at the zoomed in part. The time axis is probably for the large plot so it would be higher than 20 hz.