Advanced Search

Have a look at this, no AI was harmed during that retrofit !
forum.linuxcnc.org/show-your-stuff/32784...-15-retrofit?start=0

Hi all,
Hoping this is the right place for a conversation about machine controller conversions. I'm a hobby user, not a production shop. I've got a  Mazak VQC 20/40 mill with M2 control, it's a nice fully functional machine with good tolerances for it advanced age. It's lack of memory is a real pain point. Programming in Mazatrol CAM M2 was probably great back in the day, not so much now. I can send G-Code files to it, but hit file size limits rather quickly. It's not capable of drip feed. I think I'm ready to go for the Mesa upgrade. With AI help I've input all the info I can think of, this is where I'm at... realistic? reasonable?

Servo Motors & Drives:
X-axis: Mitsubishi HD-81-12 DC permanent magnet servo motor with analog drive accepting ±10V command (terminals: AX, BX, G1X, G2X, G)
Y-axis: Mitsubishi HD-81-12S DC PM servo with analog drive (terminals: AY, BY, G1Y, G2Y, G)
Z-axis: Mitsubishi HD-101-12 DC PM servo with analog drive (terminals: AZ, BZ, G1Z, G2Z, G)
All three drives are MELDA-8 or similar Mitsubishi analog cards that accept ±10V command signals—these are REUSABLE
Position Feedback:
X-axis: Magnescale SR-921RH linear encoder (outputs quadrature signals directly) + MS3100B 20-29P tacho generator
Y-axis: Tamagawa Seiki resolver (MS3108B 20-29P, pins: RS1, RS2, RC1, RC2, RO1, RO2) + tacho generator
Z-axis: Magnescale SR-921RH-450-R-0 linear encoder + MS3108B 20-29P tacho generator
Spindle:
Mitsubishi AC motor SE-EV-FV with FREQROL FR-SX VFD controller
Already accepts 0-10V analog speed command and direction signal—fully LinuxCNC compatible
Tool Changer:
20-tool carousel ATC with CAT40 tooling
Nachi proximity sensors for position detectionMesa Hardware StackCore Cards 
7i92H  - Ethernet-based motion controller, main brain
7i77  - 6-axis ±10V analog outputs + encoder inputs + 48 I/O points
7i49 - 6-channel resolver interface (for Y-axis only)
7i73 - Pendant/operator panel interface

Servo Drive Connections:

Existing 24VDC power supply (HR-11F, outputs P24 and G24) remains in service:
Mesa 7i77 I/O is 24V compatible
Existing Contactors (3S-1, 3S-2, 3S-3, 3S-4):
Control servo drive enable sequencing
Spindle power control
Lube/coolant pump control
These stay and get controlled through Mesa 7i77 I/O outputsATC IntegrationInputs to LinuxCNC:
Tool-in-position sensors (Nachi proximity switches)
Carousel rotation feedback
Gripper open/close confirmation
Drawbar clamp/unclamp sensors
Outputs from LinuxCNC:
Carousel motor control (forward/reverse)
Gripper solenoid
Drawbar solenoid
Air blast control

Oh, I almost forgot – I wanted to let you know if I succeed.I’ve been gradually increasing the C00.05 stiffness parameter.

Are you measuring the encoder signals against ground?

(you cannot measure differential signals on an oscilloscope  without a differential probe)

Unfortunately HDD is dead. A colegue of mine will bring a laptop he was using with simillar machine but on Windows+Mach setup, maybe i will get some info from there

I guess the H11L1SM optocoupler could be damaged but that's
much less likely as it does not connect to the outside world.

I wanted to share an approach that may interest anyone running CNC on a Linux SBC without a Mesa card or a real-time kernel.

### The setup

Hardware: Arduino UNO Q — a single board that combines a Qualcomm QRB2210 (Debian 13 Linux) with an STM32U585 MCU connected via an internal UART (`ttyHS1`). The STM32 runs grblHAL and handles all real-time step generation. The QRB2210 handles G-code interpretation.

### The key insight

`linuxcnc-uspace` ships a Python API for `rs274ngc` that works perfectly outside of a running LinuxCNC instance:

```python
import gcode
from rs274.interpret import Translated, StatMixin

result, seq = gcode.parse(filename, canon, "G21", "")
```

This gives you the full rs274ngc interpreter — canned cycles (G81, G83…), O-word subroutines, G92/G5x coordinate systems, polar coordinates, error checking — with zero HAL setup and no real-time kernel requirement.

The `canon` object intercepts the resulting primitive motion calls (`straight_feed`, `arc_feed`, `spindle_on`…) and translates them to grblHAL commands streamed over serial with `ok`-based flow control.

### What rs274ngc handles automatically

- Canned cycles expanded to primitive G0/G1 moves
- Incremental → absolute coordinate conversion
- G5x / G92 work coordinate offsets
- Polar coordinates
- O-word loops and subroutines
- Unit conversion (G20/G21)

### Architecture

```
G-code file
    ↓ gcode.parse() — rs274ngc Python API
canon calls (straight_feed, arc_feed …)
    ↓ Python bridge
G0 / G1 / G2 / G3 over /dev/ttyHS1
    ↓
grblHAL on STM32U585 — real-time step generation
    ↓
Stepper drivers / CNC shield
```

Note: rs274ngc canon calls are always in inches (internal unit). All coordinates are multiplied by 25.4 before sending to grblHAL in mm mode.

### Repos

- Bridge (Python, runs on QRB2210): github.com/hoshigarasu/canon-grbl-bridge
- grblHAL firmware (STM32U585 port): github.com/hoshigarasu/grblHAL-STM32U585

Demo video: XY pen plotter drawing a 50mm circle, G-code interpreted by rs274ngc on the QRB2210, executed by grblHAL on the STM32U585. [photos.app.goo.gl/vnDY3CgpnHprdv5dA]

Happy to answer questions about the rs274ngc Python API or the canon interface.

I'm at this step to. But I feel like it's been worse after the autotune than it was before.
That software is such a mess that I regret not going the Leadshine way.
I also have a slightly noisy Z axis servo but I can't find the gain parameter... Any clue ?

Mark I thought about your issue some more and I have a  solution. Not terribly invasive a little bit of work but saves the variator. 

Return the speed control pid tuning to linux cnc

Put a closed loop stepper motor on the variator. Here's why

1. step counts are easy targets, so 10 steps = 1 point of gear ratio for example.

2. build a table of Frequency from command to stepper that optimizes torque.

So on Y you have commanded input from Lcnc as frequency
on X you have steps iun the table itself you have gear ratios. Build you selector logic to Always shoot for best torque.




Speed Command from g code
encoder feedback
min max home positions.
then spindle at speed output.

the look up table doesn;t even have to be complex could just be along switch statement, if <200 && >300 spnd cmd gear = 4

that's it. then let linuxcnc handle the fine control. your just grabbing the spnd cmd speed and setting the ratio.

See attached

www.alibaba.com/pla/IHSS57-36-20-Nema23-...rGX0_eUaArgzEALw_wcB


 

Good Day

Does anyone know how to re-install GPlus 450 software on a Knuth Mark Super CNC running windows XP SP3
On the Gplus 450 software my machine type is not even listed
Got all the Settings Files and Software Licence No's

First time Round got this error Message (Attached Jpg - Licence Install problem) 

Controller does not want to connect to internet - So Cannot even get Online Help From Knuth Germany
Any advice would be Very Much Appreciated
Many Thanks
Otto 

… continued…

Not being able to improve on that version, I started over in  Anthropic AI (free). This has been less successful. It is a more complex solution, and have had more issues with stability, and the ratio calculation freezing.  

However what is very positive is how easy it has been to have it generating data logging and analysis files. With better data, hopefully I can make faster improvement.   

I think I will do one more attempt to control this without adding hardware.  I will go back to my original concept for this.  

Rather than trying to control the CVT closed loop off that dirty inferred gear ratio, just command it open loop, with fixed ratios. Considering the CVT as five discrete gear ratios, one second apart, I’ll do a a survey of the system to create a lookup table of VFD Hz, CVT “gear” position, back gear position verse spindle speed, overall gear ratio, P and I values. My idea being that on spindle start, it sets the gearbox, then accelerates to the VFD and CVT nearest defined speed in the table, open loop, and then the PI closes the loop to control the VFD to control to that speed. Whenever the spindle is at a steady state long enough for the inferred ratio to become stable, the ratio inferred from commanded motor speed, could be used as a sanity check whether the CVT is still at approx the ratio expected.
Cheers,
Mark

Hi,

I am having the same problem, with the same controller , only 4 axis . Same id, and same symptoms. I am running linuxCNC 2.9.8, I have tried to change the ID in the .rules file, but no luck.
Did anyone resolve this, any suggestions?

Since I was able to create a usable tool changer control comp with the coding assistance of AI, I figured I had better confront the final boss beast of this CNC conversion... the spindle control. I was already crastinating on this before I turned pro.

If you remember, I retained the original motor, CVT and back gear, and added a VFD instead of the Dahlander motor winding switching. So far I was just manually controlling the back gear and CVT, and guessing what VFD frequency to command, with a single ratio hard coded on the INI, which is roughing right for about 1000 rpm.

I want to believe that I can use a comp to have the system follow S commands, including for CSS cuts.

 

So I have:
- huge variation in inertia, requiring PI gain scheduling based on overall reduction ratio
- no feedback of CVT or back gear position, which lead to the attempt to infer gear ratio from spindle encoder / analog motor speed command.

I got it sort of working with the assistance of ChatGPT(free) for coding. However, the need to freeze the ratio calculation during CVT bumps, or commanded speed changes means it very laggy, with speed corrections happening long into cuts. 

I have attached the ChatGPT created comp and hal, for general interest...
... continued next post, to separate the ChatGPT attachments from the Antropic AI generated files.
 

I had a quick look at the PortingPlan.md. Am I right to presume that this was AI assisted coding? It's insane how this thing is a game changer for non coder; I have some understanding of coding, read a lot of source code, but am not able to write more than a couple of functions to adapt an existing code (and even that part is mainly copy-pasting form existing libraries). Yet, within three days had a CANopen motor connected to a RPi spinning smoothly in CSP mode. I wouldn't say with Claude-code's help, but rather than Claude-code had it working with some of my contribution for hradware connection and debugging. 

Back to the topic, I think that eraserhd is one the right path when trying to generate pulse from PIO. Guess that's just a prompt more for the RP2350-(not PIO as it was there for Platform-IO, right?). The stepper-ninja  for LCNC claims a 1MHz per channel step generator using PIO.

Cheers

Thank you very much. Is there anything else that can be damaged down the line? I mean the DAC?