Lagun mill retrofit with Mesa cards

A friend of mine retrofits a Lagun mill that he owns. He has already made some work with Gecko drives connected to the parallel port, servo motors from Kelling (www.kelinginc.net/KL34-170-90.pdf) and Mach3, but he's not satisfied with the result.
He now wants to change to EMC2 (loved it when we tried it with his equipment) and a closed loop setup and I will help him accomplish this, so I would like your opinions and advice on what I'm thinking.

Our goal is to keep the cost as low as possible, without sacrificing quality of the results. We'll keep the motors and encoders and I'm thinking using a Mesa 5i23, two 7i29s for PWM driving of the motors and one or two 7i37s for general I/O.

The general idea is closed loop control with the motor encoders, but I'm also thinking of trying velocity loop with the motor encoders and a second position loop with the linear scales (5 um, Taiwan made) already fitted on the machine. I read in another thread here that this is kind of difficult to achieve though.

I would like to have some opinions on the proposed setup and the two closed loops.

Thanks in advance!

Thanks for the cents John!

5i23 is a little more expensive than 5i20, so I thought that it would be good to have a 400K instead of 200K gates card. Even though I don't know what this means in practice. PCW, could you please explain what is the difference?

I wonder why it is hard to close the loop with scales. Is it because of the backlash, or something else? Andy mentioned something about dither in another thread. What does this mean?

BigJohnT wrote:

I think a DRO type of scale is in the neighborhood of 0.0005" of resolution... but I might be wrong about that assumption.

And yes backlash if you have it will make servo tuning impossible if you try and use a linear scale due to the normal hunting of a count or twe of a servo drive.

Yes, it is in this neighborhood of resolution. Actually, as I mentioned in the first post, this one has a resolution of 0.005mm, so roughly 0.0002". I hope that if we manage to get a tight enough pid loop, we'll have a max error of +-4 scale steps (+-0.0008"). I don't know if this is possible though.

Backlash will be eliminated by using two ballnuts and some belleville washer springs, so I guess we will be ok for this kind of problem.