Pick & Place MachineTVM920

[dtf]

the actual placement, and quality of machine has been excellent.

the software, sadly, is better than hobby-grade, but frustratingly non-pro.

[Spikee]

Ever tought about retrofitting the controller with something else ?
Their closed loop stepper / servo probably just uses normal stepper input. It would probably be the fastest way to get it working.

The thing that holds me back is that these pnp TVM920 , Neoden4 are
expensive enough that it should just work (good) enough.

It still seems that OpenPNP or modding the liteplacer sw (.net) is one of the best way to go at this moment.

[dtf]

Sure if something was available I'll try it. But, currently the machine works, is under warranty, and is used for production. So any modifications needed or down-time required means it won't happen.

Last post in the openpnp thread was a month ago, I'm not holding my breath.

[Smallsmt]

If we receive the interface description we build a DLL to operate the machine with our VisionPlacer software.
I think there is a similar controller inside like TVM802.
Do they use ethernet connection or USB to talk to the controller?

[Spikee]

People who have the machine can you show some of the software (screenshots) and describe where
the major problems are.

I'm trying to figure out if this machine is usable "enough" for some 0402 passives and some qfn / bga work.

[dtf]

If we receive the interface description we build a DLL to operate the machine with our VisionPlacer software.
I think there is a similar controller inside like TVM802.
Do they use ethernet connection or USB to talk to the controller?

Ethernet, ar-systems has started work on reverse-engineering the TVM802 protocol already. Cameras are via a cheap CVBS capture card.

[Spikee]

I asked Qihe and they are going to make a video from setup to placing of a board with mixed components.
As in passives and IC's. They are not working on openpnp support at this moment.

For the people that are really interested in this product just contact them and ask them questions.
If you want openpnp support communicate that to them. If they have enough demand for that then
they probably will work on supporting it.

[Spikee]

And just a tip. If you find a problem or a software bug than it is important
to write a very simple step for step procedure of what is wrong. Pictures with arrows showing directly
what is wrong goes a long way with any developer.

I do these step:
1.
2.
3.

This happens:
[picture with arrows]
[some text what is happening]

This should happen:
[picture with arrows]
[some text about it]

-----
Edit:

New TVM920 software will have specific vision settings per feeder.
Vacuum setting is just one value for all things.
release date tbd.

[ttsthermaltech]

Heard that a new software version will be out in the next few days... I am just starting production on mine, likely tomorrow. Found one little annoying bug in the vision, and it appears that if it can't find the center (bad image / lighting), it just keeps trying and never fails. Found that knocking the component off the nozzle will make the machine do a dump operation and then re-pick, but gotta get a bug report to Qihe on that one. Had pretty good results the other night using nothing smaller than 805s, but I want to get things a little better yet.

Vision settings per component / feeder would be a big step up.

R.

[Spikee]

Update list:
-specific vision settings per feeder
-component rotation setting per feeder

I'm trying to get the whole list of changes that have been made.

[dtf]

Heard that a new software version will be out in the next few days... I am just starting production on mine, likely tomorrow. Found one little annoying bug in the vision, and it appears that if it can't find the center (bad image / lighting), it just keeps trying and never fails. Found that knocking the component off the nozzle will make the machine do a dump operation and then re-pick, but gotta get a bug report to Qihe on that one. Had pretty good results the other night using nothing smaller than 805s, but I want to get things a little better yet.

Vision settings per component / feeder would be a big step up.

R.

I had this same problem and it was quite frustrating, until I realized the camera was picking up a reflection instead of the part outline, and simply keeping the shades down eliminated the issue

[thommo]

I'd recommend also removing the piece of Acrylic which originally sits above the camera [and the LEDs].

I have removed it completely on my machine and have replaced it with a 47mm UV filter, which just sits directly on top of the lens itself. It has the benefit of protecting the lens still, but does not suffer from the flair or the  potential for reflections which can be caused and captured in the original Acrylic which was mounted on the top of the housing.

Cheers - Peter

[Spikee]

There are so many auto threshold kind of vision operators. I have implemented various methods (by hand so no opencv) before.
Also finding the shape of a components is pretty easy. Especially since the component body's area is already on the center of the cam.
you just need to work inside out instead of finding a shape that's somewhere in the camera view.
But it seems very difficult for them to implement things like that.

[thommo]

Yes, agree Spikee, but none of this helps or makes any difference if the 'image to be recognised' is contaminated with irrelevant shapes such as reflections, as DTF points out and experienced.

Cheers - Pete

[Spikee]

you could just use a "fillholes" kind of operator but yeah ... less processing is better.
I was not talking about reflection by the way but in general PNP vision operation.

One might almost just make their own software ...
--

If they just would opensource their software or their API 
It is not like they sell thousands of these things. It is probably in the hundreds. Better software will get them more sales
and a better name. Current SW is probably meh at best. A decent programmer could replicate it in a short time.
So being afraid of getting copied is not a good argument. But yeah...

[vonnieda]

you could just use a "fillholes" kind of operator but yeah ... less processing is better.
I was not talking about reflection by the way but in general PNP vision operation.

One might almost just make their own software ...
--

If they just would opensource their software or their API 
It is not like they sell thousands of these things. It is probably in the hundreds. Better software will get them more sales
and a better name. Current SW is probably meh at best. A decent programmer could replicate it in a short time.
So being afraid of getting copied is not a good argument. But yeah...

I'll reiterate my offer that if someone sends me one of these I will get OpenPnP running on it. Then there will be a completely open source software stack that can be used and improved for these machines. If there are enough people interested enough to throw a couple hundred bucks at the problem we can make it work.

Jason

[ttsthermaltech]

Jason,

#1. Thanks again for writing a great piece of software. I know it is a tonne of work, and you deserve a pile of respect for it.

I think the key issue at the moment, is that the firmware in the Qihe's is good enough, and works well enough that spending the time to create a driver is low on the totem pole. I want to do it myself, but the problem is that Qihe won't give up the protocol. Odd, as it only adds value to their machine. It is not like we want to open source what they have done so far in terms of software, but OpenPNP is orders of magnitude ahead of them, and they should recognize that it would only strengthen their market by either writing their own driver, or just releasing the protocol, so we can do it ourself.

I my opinion, and based on the motion driver in the current control board (not linearly interpolated), and since it lacks certain hardware assisted features (vac pressure specifically), it would be faster and easier to just rewrite the firmware for the existing board and expose everything to openPNP over ethernet.

What i want to do (in the new year), with some help from a few of my friends (Thommo and GlenEnglish) that we design a complete new control board, with extra IO capabilities (specifically analog for sensors, encoder interfaces for X & Y, and serial / CanBus) that is plug and play with the current wiring, yet fully openPNP compatible, so that it would be a 15 minute swapover from the existing board, and offer a future proof control system.

The Qihe 920 hardware is brilliant, and I would not hesitate to purchase a second unit when the need arises, but they need better software. not that it is buggy, just lacking in features, that a machine base of this capability can truly offer.

R.

[dtf]

I'd be thrilled to try some new software. And spend time on development. But I don't think anyone is going to send you $10k of equipment.

[Spikee]

I assume this machine just uses closed loop steppers that have the normal step input.
It would be rather easy to connect this to TinyG controller board or gecko G540 4-axis motor controller.

I assume the feeders are "stupid" and only require a solenoid to be switched on for x time.
This could probably be solved with the TinyG or just make another arduino shield that can control the solenoids which you
control by sending a simple uart command like "Feed 1".

By doing it this way it would also be easy to convert any of the other Chinese pick and place machines with feeders.
Instead of making / supporting a separate controller for each machine just swap out the controller.

Using this method you also don't have to risk breaking the machine and have warranty issues because it is more or less plug an play.
And you can plug it out with ease.

If this is the best machine to choose at this price-point in feb/march than I will get one.
IF the software still sucks or I need special features than I will definitely consider doing a retrofit using the described method
and use openpnp.

It has been done before using this method:
https://github.com/openpnp/openpnp/wiki/Retrofitting-a-Zeva-460---Part-1

[dtf]

They are indeed regular 4-wire steppers (with feedback) and regular china stepper drivers.

I don't know why you guys are so interested in modifying your machines, imo the hardware is fine. Actually pretty good. And any hardware modifications needed to run 3rd party software would pretty much be a non-starter for me, introducing downtime to the machine. Also adding time needed to go back to working state.

[ttsthermaltech]

Its not that I want to modify the machine, but considering the exceptional build quality, this machine begs to do more than it's software is capable of. I know it pretty much works out of the box, but there are some things that could really benefit the machine (and the user)...

1) Vacuum sensing of components at each of the four heads
2) Brightness / Color / Exposure control of the up vision LED's
3) Auto nozzle changer
4) Multiple cameras and or zoom control

Those are my big wants... All of which could be easily added to a new control board, and most are already featured in openPNP. Yes, it could be done with a bunch of tiny-G or other, but the 9 steppers on the 920 make if difficult to use a single motion unit. A custom one would be ideal, as you could likley port much of the motion code necessary from one of the open source variants, but having the ability to switch axis within a single controller just makes things simpler.

My $0.02 worth.

R.

[zszabo]

Just arrived mine.

Well packed.

[mikeselectricstuff]

4) Multiple cameras and or zoom control

You don't really need zoom, assuing you start with a camera with decent resolution for the smallest parts.
 The issue of parts that are too large for the field of view can be fixed by imaging multiple times & moving the part - although slower, chances are there won't be enough large parts to make a big enough speed difference to warrant a different optical FOV.

 

[Spikee]

Just arrived mine.

Well packed.

Please keep us updated

[Smallsmt]

Its not that I want to modify the machine, but considering the exceptional build quality, this machine begs to do more than it's software is capable of. I know it pretty much works out of the box, but there are some things that could really benefit the machine (and the user)...

1) Vacuum sensing of components at each of the four heads
2) Brightness / Color / Exposure control of the up vision LED's
3) Auto nozzle changer
4) Multiple cameras and or zoom control

Those are my big wants... All of which could be easily added to a new control board, and most are already featured in openPNP. Yes, it could be done with a bunch of tiny-G or other, but the 9 steppers on the 920 make if difficult to use a single motion unit. A custom one would be ideal, as you could likley port much of the motion code necessary from one of the open source variants, but having the ability to switch axis within a single controller just makes things simpler.

My $0.02 worth.

R.

It really has no vacuum sensing on each head?
How does it know a part was picked up?