Re: [CAD_CAM_EDM_DRO] Re: Update on closed-loop to the computer with Gecko 320

Mariss Freimanis wrote:

>Jon,
>
>1) Do not use the HCPL-2530. It's gain is only 9%.
>

OOps. Well, I'll have to dig around, I may at least have something better
for a one-axis test.

>
>2) Do not use a 74HC part as a buffer. It's only good for +4/-8mA.
>Use 74AC parts; they are good for +/-24mA. Same pinouts (7414 =
>74HC14 = 74AC14).
>
>3) Meant as written. ENC+ (+5VDC) is also the virtual analog ground.
>This makes it a node, meaning current flows into and out of it. If
>current flows into it (from internal analog circuits), it causes the
>node voltage to rise above the regulated +5VDC. A 470 ohm resistor
>ensures that can't happen; the encoder load normally takes care of
>that.
>

Ahh, I'd missed that.

>4) It is symmetrical. Your brushes may have set in a direction. Run
>the motor a few hours in the opposite direction and try again.
>
>

If I don't have enough load on the ENC+, and the virtual analog ground
is bouncing around, I could imagine all sorts of odd effects in the
control loop!

>5) PID tracking is an unecessary luxury for this price drive. The
>drive is tuned for a particular system power supply that normally
>will not change over the service life of the system. That leaves
>ripple voltage which is an insignificant second-order effect.
>
>
>6) Fine-pitch placement requires machine vision. This slows a machine
>down from 15,000 parts / hr to about 3,000 / hr. and we have a
>Zevatech FS740 for that. I try to avoid fine-pitch packages for that
>reason; time is money.
>
>Whose FPGAs are you using? Xilinx? I ask because I'm learning to use
>their 9500 series CPDLs which I will eventually use to replace the
>SOIC glue logic in all our drives.
>
>

Yes, I bought the Xilinx Foundation package, and was later updated to
the ise package. I haven't updated since, because I need the 5V devices,
which the new Xilinx software does NOT support! Incredible!
I use Xilinx's Spartan 5V FPGA, the best deal for the kind of stuff
I'm doing. This involves lots of counters, FFs and gates, but pretty
light on memory. The XC9500 CPLDs are great in the 500 -1500
gate range, but become very expensive (and hot, too) when you get
much above that. The 10,000 gate Spartan XCS10 runs about $12,
the 30,000 gate Spartan XCS30 runs about $30 in small quantities.
The XCS30 has all the logic on my Universal Stepper Controller
board.

>7) Eliminate the D term by turning the DAMP trimpot full CCW.
>Eliminating the I term requires shorting an 0805 capacitor.
>
>

Well, I really don't think I'm going to try that route, as the scheme
I tried first works so well.

Jon