Rod's Spaceship Scratch Built Plasma Table

[robertspark]

reverse run which is used for error recovery which is probably a not a major impediment. Plasmac will auto sense if reverse run exists and configure itself accordingly.

Unfortunately, the release manager for LinuxCNC did not want to release the reverse run feature until there had been further testing so it will be in V 2.9.

hmm, now that is an interesting feature to add.

How does it work?

In UCCNC it works by
A) providing THC is enabled
B) providing M3 is active
C) tracking the arc OK signal

if the ArcOK drops out with both A+B are active, then the torch will decelerate to a stop, and backup to the ARC OK lost position and await a restart.

[Rodw]

To tell you the truth I don't know as I've never needed it to recover from a fault yet. I think it allows you to move back and forth as far as the last M3 while the machine is paused (which loss of ArcOK triggers). There are some controls on the GUI to control this.

There is a video on Youtube of a guy using a MPG wheel to move back and forth thru the gcode with reverse run active which is cool. Search for user Samco from memory. Because I have 3 spare axes on my pendant, I was going to try that one rainy day...

[robertspark]

Thanks,

Interesting
https://www.youtube.com/watch?v=3aYaHxT6ZnQ

[Rodw]

You found it. I was actually thinking about the UCNC algorithm.

I don't think the travel distance after loss of ArcOk is significant with LinuxCNC. Say I'm cutting 2mm steel at 2100mm/min (35mm/sec). Linuxcnc will sense the loss of ArcOK in 1 millisecond and if my X/Y Maximum Accelleration is 1500 mm/sec/sec (which I think is the rate we will stop at), I've not moved further than about 0.8 mm from the loss of the arc (Just enough to have some material to restart on). I guess thats one of the advantages of running in real time.

[beefy]

Hi Rod,

The ultimate would be reverse run to a certain distance BEFORE the flameout point. Then restart the motion but without torch switched on. You'd also need THC off for this of course. Then just a few mm before the flameout point, manually turn the turn on (assuming this would not cause a pause in motion). Once cutting again, manually enable THC again.

[Rodw]

This does not work for our config (and probably not for LinuxCNC). Our torch is turned on with an M3 and with it all of the probing etc. When a fault ocurs, we pause motion so to restart it is a resume, not a full restart,

[beefy]

This does not work for our config (and probably not for LinuxCNC). Our torch is turned on with an M3 and with it all of the probing etc. When a fault ocurs, we pause motion so to restart it is a resume, not a full restart,

Not sure if I'm understanding that correctly Rod. Are you saying that in Lcnc, an M3 is not simply a torch fire but also initiates a bunch of other things happening. For that matter is there no way to manually fire the torch while doing a dry run of the gcode.

In Mach3 and UCCNC for example there are things like "Run From Here" or "Set Next Line" which can be used to start a program from any point in the gcode file. Of course you may need to manually touch off and set the Z height first, and possibly manually set feedrate before starting these mid file runs, because certain modal commands may not be in place.

Just trying to get a feel for how Lcnc runs compared to what I'm used to.

[islander261]

Guys

I have used the reverse run feature in the PlasmaC branch many times now. It is very nice when recovering from torch break aways when you hit tip ups. I just back up to the point the cut ended (a manual button on the GUI) and start again. PlasmaC always probes after an error so no special height setting or manual torch firing required. I think the docs Phill and Rod have posted now capture most of the functionality.

PlasmaC shares the general LinuxCNC problem of poor run from here support. It should always work if you are starting a file over but I wouldn't call its functionality robust. The good reverse run saves much of the need for this. There is an open source fix for this but I have had no luck in persuading any developers to make it part of the main stream configurations ( it is far beyond my coding ability).

PlasmaC takes full control of the Z axis away from the Gcode. When setup correctly a M3 S? will start the probing and torch control sequence. It is hard coded in plasmac component, something Mach and UCNC don't have. The S block is need in all LinuxCNC configurations because of some hard coding, the value of the parameter can be anything when plasma cutting.

You can email me with questions Kieth as I have been using PlasmaC with great success since early May. I was initially very wary of letting PlasmaC have total control over the Z axis and torch firing but Phill C. (one of your country men) put in incredible work and made a really good product. This is the solution that I was looking for 3+ years ago when I started trying LinuxCNC.

John

[Rodw]

John, thanks for helping Keith out there. John knows far more about error recovery than I do. John has had a major influence on Plasmac and Phill in Melbourne had this knack of capturing all sorts of different ideas into one cohesive unit.

[beefy]

Thanks very much John & Rod,

and John I much appreciate your offer of emailing. Got a bit better grasp of Lcnc operation with the M3 now.

So like me, Phill is also in Melbourne. I think there is something wrong with us poms. We escape the crappy British weather and come to the cold winter rain of Melbourne LOL. I was supposed to go to tropical Queensland (barrier reef and all that stuff) but somehow I ended up here.

Keith

[robertspark]

nothing wrong with the weather here.
crackin the flags at the moment

[Rodw]

This isn't my handiwork, but here is a sheet of 16mm cut with the Linuxcnc plasmac config.

fullsheet.jpg

Stefan who converted from another well known commercial controller often used on this forum said the THC was still very much required even on material this thick. I think the right hand corner shows why....

[Rodw]

Its been a while since I did an update here. Our LinuxCNC Plasmac config is jumping ahead in leaps and bounds. I've had the parts laying around for years and never made my scribe mechanism but I think I better get onto it soon because scribes are now supported in Plasmac. This is done a little bit differently then in other systems becasue LinuxCNC supports multiple spindles so its just a matter of loading a scribe tool from the tool table that applies the tool offset from the torch by using standard tool change M6 commands and then specifying the spindle to use in the M3 command.

Plasmac has also gained inbuilt support for hole cutting where overburn and velocity slowdown is supported.

For guys cutting thick material, there is also a pierce only mode so you can do all pierces in the job and clean the dross off the plate before cutting out the parts. Stefan has used this to cut full sheets of 16mm and it all worked perfect! Here is a simulated demo of this in action
https://youtu.be/2R8zMYsL_A8

While this has been going on, I've been working with Peter Wallace from Mesa Electronics and today I finally got some real cutting done using a Mesa THCAD-5 card for Ohmic sensing. Peter wrote some new firmware for my Mesa 7i76e board so it can support up to three THCAD boards so I have one to sense the arc voltage and another for ohmic sensing. This is managed by a component I wrote that runs on LinuxCNC's servo thread 1000 times a second.

So here is the hardware:

hardware.jpg

Parts from the right of the unconnected blue relay in order:
24 volt power supply with 500 volt isolation
390 K scaling resistor to give a full scale of 24.5 volts (overkill mounted using a Mesa BUS strip that has 2 segments commoned together.)
THCAD-5 with 5 volt range (these can handle 500 volts overvoltage indefinitely)
Earth clamp for THCAD ground to table via enclosure bolted to it.
This circuit is powered up when the control box turns on it is powered all the time

So with this setup, everything in the ohmic circuit is 100% isolated so we don't even have to worry about protecting the circuit from arc voltages. I will say it took a long time for Peter to convince me of this!

So when I did some testing I found that the ohmic voltage rose from 0 volts to 24 volts over Z axis travel of 0.04mm and the 24V opto isolated relay I have been using turned on at 7.5 volts and goes off at 5 volts. This is likely too sensitive in some situations. So with the THCAD actually reading the voltage as the torch shield comes in contact with the material, I decided to enable a probe signal once the voltage exceeded an arbitrary 18 volts.
As an added bonus, I decided to add what might well prove to be a totally useless feature but I knew the THCAD can read a bit more than its full scale so by having a 24.5 volt full scale range, we can turn on a pin if the voltage exceeds 24.5 volts. Becasue we are using a 24 volt power supply, the only way that can happen is if the torch is actually cutting! So we have this cool and very wanky indicator we can use if we want to.

So there was only one way to see if this would work so I unleashed it on a job with 110 pierces and it worked flawlessly!

parts.jpg

Even more amazing is I did not blow anything up as the circuit was powered for the full job!
The next stage of development for this is to modify the component so it turns on say somewhere between 18-22 volts and stays on until the voltage falls to say 1-5 volts so there is a big range of sensor hysteresis for nice stable sensing. I will add some GUI controls so we can vary the sensing sensitivity for different conditions. It will be interesting to see how this will work with a water table. That is the next step.

So becasue this circuit does not rely on any resistors and has such fine control over material sensing, I have decided to name this hyper sensing (which one wag suggested was ohmic sensing on steroids!)

One thing we heard was that with a water table, leaving the sense circuit powered all of the time could result in corroded consumables. Does anybody have any knowledge about that?

[Rodw]

On this thread viewtopic.php?f=103&t=28781&p=174194#p174194 I said I would publish the ohmic sensing circuit used in the
last post and here it is:

HyperSense DIagram (1).png

I thought this is so vastly different to Ohmic sensing that we are calling it hypersensing which as one wag put it is Ohmic sensing on steroids!

The 390K resistor scales the 24 volt power supply to 24.5 volts. Because of the isolation on the THCAD, it would not matter if you were using a lower voltage for field power becasue the probe signal is generated in software.

I've now removed all the old ohmic stuff I had and am using this in production now its that good!

IMG_20190831_173258 (1).jpg

Somebody mentioned that there seemed to be a lot of wires but in reality most of them are hooked to other hardware

IMG_20190902_083822.jpg

The Mesa 7i76e is in the bottom right corner and it has another THCAD-10 board mounted beside it to the left to measure the torch voltage. Setting the voltage divider board in the plasma cutter to 30:1 is perfect for the THCAD-10 as then it needs no scaling resistor and provides a 300 volt range.

The THCAD will come into its own with high frequency start plasma cutters where EMI/RFI noise can cause havoc.

[cuttinparts]

Clean crisp wiring Rod Everything looks awesome. Nice job.

[robertspark]

rod, I thought there was only one encoder input on the 7i76e , where is the second?

what is the advantage of using an thcad card?

I understand it will have an accurate analog input (effectively) via the encoder.... but it will still have a digital threshold voltage / setting to determine if the plate has been touched or not?

what is the advantage over a bog standard commercially available ohmic sensor that has a potentiometer adjustment to set the threshold and a digital output ??

[Rodw]

Clean crisp wiring Rod Everything looks awesome. Nice job.

Thanks, It got a bit messy after I added Ohmic sensing so I still need a bit of tidying up and get the cover back on the conduit

There is actually (hopefully) a bit of science in the layout to defend against noise, all the mains power stuff is in the top left, all the 48v to the stepper drivers goes around the top and the logic wiring is in the conduit above the motion boards. If I did it again I'd downsize the 24v and 5v power supplies and make more use of DIN rail mounts and DIN rail terminal strips.

[Rodw]

rod, I thought there was only one encoder input on the 7i76e , where is the second?

Aha! The THCAD only uses a 2 wire output attached to Encoder A. So Peter at Mesa wrote new firmware for the 7i76e and 7i96 that places 2 additional encoder A's on the Encoder B and Encoder Index pins on the boards. The firmware is available for download from the Mesa product pages in a Zip file. The file names end in "pl.bit"

what is the advantage of using an thcad card?

I understand it will have an accurate analog input (effectively) via the encoder.... but it will still have a digital threshold voltage / setting to determine if the plate has been touched or not?

This project was based on a theory of Peter's that the THCAD might let resistance be calculated from voltage loss. He sent me the THCAD to experiment with. I found on my table there was no voltage loss with firm contact but approaching the material with 0.01mm increments, I saw the voltage increase from 0V, to 12V, to 18V, to 24V
I also looked at the 24V relays I was using and saw they turned on at 7.5 volts and off at 5 volts.

So I wrote a component that allowed an adjustable upper threshold (I chose 18 volts) and I am going to add another threshold at say 1-5 volts at which it will turn off to make sure we have a super wide switch hysteresis to avoid any spurious triggering.

To adjust sensitivity, it is intended that these thresholds will be able to be adjusted from the GUI screen.

what is the advantage over a bog standard commercially available ohmic sensor that has a potentiometer adjustment to set the threshold and a digital output ??

None of us are using a commercial ohmic control with linuxcnc, just a couple of relays. But as I see it,

1. Costs are not significantly higher ( I removed AUD $60 in opto isolated relays and added an AUD $100 Thcad.)
2. Much simpler circuit, no need for diodes to protect from reverse arc voltage
3. no voltage loss
4. Instant sensing for faster and most accurate sensing.
5.no relays which are often quite slow to respond (have a look at the opto22 specs for example)
6. bulletproof in relation to arc voltages,
7. High noise immunity due to the THCAD design
8. Logic circuit is decoupled from the noisy plasma environment so no possibility that the ohmic sensing wire being used as an antenna to introduce spurious faults in a HF environment (eg 200 amp and above).
9. precise sensitivity adjustment

With the plasmac config after probing, the Z axis is lifted at the rate of 0.001 mm per millisecond (from memory) until the sensor turns off. I found the THCAD appeared to be faster with this final part of the probe. eg it touched and instantly rose to pierce height and fired.

[robertspark]

thanks rod for taking the time to reply, I see where your coming from now plus have a bit more food for thought

[Rodw]

Well, its been a while since I updated this build thread but a few months ago, decided to turn it into a water table

waterpan.jpg

watertable.jpg

And of course, this brought with it the inevitable probing problems when water entered the torch. For the most part I could work around this but I did notice that the larger 120 amp nozzles had more issues than the 40 amp nozzles I usually use. I found on my machine, I could change the threshold to get this working but people using Hypertherm 45XP's found they could not get enough voltage drop to insert an on and off threshold plus the signal was very noisy by comparison.

Here is a plot of voltage during a probing event with a wet torch. Instead of getting a 0 volt to 24 volt range, I was getting a 22 volt to 24 volt range. Note two signals are shown here, the higher one is the actual voltage and the other one is the probe signal. This area is where the torch touches the material and when it breaks away when probing back away from the material. It is that trailing edge we use to set the material height.

wet probe original.png

The beauty of our hypersensing technique is we could change a voltage threshold to accommodate this. I changed our algorithm to use a moving average of the voltage to dispense with the low threshold that we had been using based on replicating a simple relay with hysteresis. NAd that semeed to make it more responsive and improve the chance of false triggering. But then Peter from Mesa suggested that we add a resistor to the sensing circuit to reduce the sensitivity. But the catch was the resistor had to handle 300 volts when the torch was on. At a rought guess, he came up with a 24k resistor at 3.75 watts (so use a 5 watt resistor). When I went looking for one locally, they did not have a suitable one, so I ended up with 5 x 4.7k 1W resistors in series to give 23.5k @ 5 watts.

ohmic.jpg

They look a bit untidy but I wanted to keep them well of the circuit board with plenty of air around them becasue I figured they would get hot!

So this had the desired effect to increase the signal hysteresis making it much easier to sense the change n voltage. Its not much but it does make it reliable to probe with

wet probe.png

So it might look quite theoretical but what we have here is arguably the first ohmic sensing circuit that can reliably probe the surface on a water table with water having entered the torch and shorting it out. Finally, its probably worth showing a probe event with a dry torch (eg downdraft table) whihc clearly shows that the issue we are dealing with is peculiar to water tables..

dry probe.png

We might need a bit more fine tuning of resistor values but I hope I've demonstrated that achieving reliable ohmic sensing on a water table is just one more reason to use LinuxCNC's Plasmac config instead of any other controller.

[robertspark]

[Rodw]

Its about time I added a bit of an update to my build thread. During the Coronavirus lockdown I was approached by Marshall in Texas who asked me if I could help him build a plasma table. He said he was a qualified engineer but realised after he decided to build a plasma table, he knew nothing about plasma tables but liked my idea of build a reference plasma controller using Linuxcnc.

Marshall wanted to harness the massive torque of stepper motors to attain high acceleration rates and had a feeling that some of the motor and controller packages on the market were not particularly well engineered with this in mind. So the goal became to produce a carefully matched motor and controller package and possibly extend that to a range of plasma tables. Man, we had no idea how hard this would become and we are not done yet!

So the first step which took quite a few weeks was to build a comprehensive stepper motor selection model. He built the model, and I reviewed it and compared the output to what I knew was possible in the real world. This proved particularly difficult becasue what he thought were authoritative white papers on the Internet contained errors. One caught us out for a long time and it took a lot of research to validate our suspicions of an error. However, finally we had a very comprehensive model that took the axis mass, pinion size, gearbox ratios and all of the friction losses into account. Then we extended that to allow entry of motor specs and then predict the voltage and amps required for a specific stepper motor and also determine the heat increase while operating. I profiled hundreds of drives and finally found motors in NEMA23, NEMA24 and NEMA34 that satisfied our model requirements.

One of the things we learnt from the research phase is that if we matched the motor to the required parameters, because plasma has no cutting forces, we could be 100% certain that the motor would never operate under conditions that could lead to missed steps as they are a very predictable device! The other thing we could see is that many motors did not need to operate at their maximum amp rating to achieve our criteria. For example the 2 amp NEMA24 only needs 1.2 amps peak at a minimum of 72 volts to meet our deign criteria.

Still to this day, I don't fully understand what constitutes a good motor except that the rotor inertia is the most significant factor.

Then we looked around at available stepper drivers and I approached Lam Technologies in Italy who put me in touch with their incredibly knowledgeable Australian distributor Ron Nollet. We looked at a number of alternatives but in the end, Lam just really knew their stuff, had a range of drivers that can handle 50 volts, 90 volts and 160 volts. Leadshine and others just did not even come close! One of the big advantages we saw is they had a range of AC drives that we could connect direct to an AC toroid transformer without the need for a rectification circuit which made for a much cleaner install.

Along the way, I decided that it would be a good idea to retrofit my table to get as close as I could to the model design.

The other issue was upgrading my 48v switch mode power supply to a higher voltage toroid. Here again I was constrained by what was available off the shelf and I settled on running two 32 volt Toroids in series to give 64 volts out at 20 amps from Tortech, an Australian manufacturer. These power supplies were multitapped for various voltages and Tortech set them up for me so I could get 32 volts out of them.

IMG_20200602_122110.jpg

They could have manufactured a toroid to our requirements but the timeframe was a bit too long for my liking...

I bought 3 x Lam DS1076a stepper drives and their Programmer

IMG_20200527_132012_MP.jpg

But later I decided it would be a bit silly to not replace the Z axis motor so Ron also sent me a Lam LS1073a 3 amp open framed driver.

phew, enough for one post. to be continued.

[Rodw]

Then I lucked on a batch of Sanyo Denki 2 amp NEMA 24 (60mm) motors. I ran them through our model and they were a perfect match. SO I quickly grabbed the batch of 5 drives for AUD $100 for the lot when the RRP is over AUD $200. I was able to purchase the wiring harnesses so I ended up with a $220 drive for $30 a piece...

They are quite an upgrade when placed beside the existing cheap motors I was using.

IMG_20200622_171216.jpg

My machine uses NEMA34's on the Y axis becasue I was lazy and did not want to make the reduction drives so I settled on some Moon NEMA34 7 amp motors. (Marshall had introduced me to this brand) If this was a new machine, I think I would have used the smaller Sanyo's.

IMG_20200702_132400.jpg

So then I had to work out how to mount the Toroids as there was no way I could fit them in my control box. I decided to plasma cut an enclosure that could piggy back onto the back of the existing control panel and have it folded up.

MVIMG_20200611_180026.jpg

Did I tell you I was a crappy welder?

But when it was powdercoated, no one will ever know

IMG_20200625_172806.jpg

Lam recommended placing fuses on every wire to the stepper controllers and I found some nice DIN rail mount fuse holders (the black things) So with AUD $1300 worth of drivers going in, I thought it was cheap insurance! Also I had heard of people having trouble tripping mains fuses due to the inrush current and having to change their mains breakers. I raised this with Tortech who provided me with a couple of Thermistors (small black disks on the grey din rail connectors). These things take a bit of time to heat up and pass current so it solves the issue (A lot cheaper than getting an electrician in after the fact!)

Then I got my mate Chris around and we fitted this 17kg (37 lb) monster

MVIMG_20200703_160019.jpg

Out of sight, out of mind they say.

And I swapped out the drivers

IMG_20200702_170351.jpg

I had to redo a mounting plate for the X axis motor to accommodate the bigger framed motor. When I pulled off the old motor, I found the flange bearing supporting the pinion needed replacing. Party becasue I lost a lock nut right back when I built it so I bought a complete new flange bearing assembly

IMG_20200511_102433_MP.jpg

This looks much better

IMG_20200625_172846.jpg

So I still have to swap out the NEMA34's and replace the NEAM23 on the Z axis with a NEMA24 but its back together again.

Just as we predicted in our model. 30 m/min rapids with Acceleration of 0.5 m/sec/sec (Just over 0.5G)



So stay tuned, more to come.

[xnaron]

Awesome build. You have really got me thinking about trying LinuxCNC. I started a thread here to get some input. viewtopic.php?f=3&t=30758

[Rodw]

So I've been still fine tuning this build after fitting the new motors and drivers. It seems I have a nest from hell that is a real torture test. Lots of pierces (about 1200 on 4' x 4' table and lots of small parts). The first time I cut it, one of the motors on the gantry overheated and that caused missing steps. So I decided to incorporate the LAM driver boost feature so that the current is reduced by 50% once it is at constant velocity as shown in this video.



You can see we are really smashing the table here. Time to repeat the torture test1

This substantially reduced the motor temperatures and then I got a driver over temperature error. This time, I had also connected the LAM fault signal to the Linuxcnc amplifier-in-fault for each motor. So the job stopped immediately. But unfortunatley the fault was trewated as an estop so I was unable to recover mid cut.

Its my own fault as I have placed the drives in a vertical stack so the top drive which errored had some heaters below it. I am running the drives very close to their max specs.

But one of the lInuxcnc guys suggested I connect the fault signal to the spindle at speed signal (motion never starts if the spindle is not at speed). So with a bit more trickery, I should be able to make the system recover and start cutting again automatically once the temperature comes back in spec.

I will say that this is definitely a torture test becasue I also had an over-temperature error from my 120 amp Thermal Dynamics Plasma cutter and it was only cutting at 40 amps! This was eaxy to recover from but I think it was becasue there were so many pierces and short cuts so the machine had to work a lot harder than if it was cutting condstantly.

SO the next chore that has been outstanding for a long time is to stiffen up the gantry ends to reduce the flex under accelleration.