Pick & Place MachineTVM920

[zszabo]

$60 for 45 that comes out 1.33 And probably free shipping.

Removed my feeder plates.
Waiting for results from Peter or QiHe.

[zszabo]

Pic again, as I see cant be displayed for some reason.

[48X24X48X]

I think that $250 was the DHL shipping cost.

Okay, I reread the sentence again. 

I know there's quite a number of TVM920 owner here, does anyone know like for a particular size of tape width, how many slots it occupies?
On a TVM920, it has 56 slots that can hold up to 56 8 mm CL feeder.
So, how many slots does a 12 mm, 16 mm and 24 mm feeder needs?

[RobK_NL]

$60 for 45 that comes out 1.33

Oh dear, I stepped right into that one, didn't I 

[dtf]

I think that $250 was the DHL shipping cost.

Okay, I reread the sentence again. 

I know there's quite a number of TVM920 owner here, does anyone know like for a particular size of tape width, how many slots it occupies?
On a TVM920, it has 56 slots that can hold up to 56 8 mm CL feeder.
So, how many slots does a 12 mm, 16 mm and 24 mm feeder needs?

12mm takes up 3 slots, but if you place it next to another 12mm, you can place feeders every 2nd slot.

16mm takes up 3, but if you place it next to another 16mm, you can again place feeders every 2nd slot, they takes up 5 slots for 2 feeders.

24mm takes up 3 slots, but can only be placed on the front stack. Maybe a 12 or 16mm feeder can be squeezed in next to one, I don't know.

Don't know about larger feeders I didn't purchase any.

[Spikee]

That also depends on your feeder spacing. This does not seem to be a standard. I have seen feeder plates where the distance is 18mm, 20mm or even 24 mm between two slots.

[glenenglish]

a morning's work... 6 hours to be exact

straightforward.  right down to the limit switches.

attached

it is written as a journal as I figure stuff out with commentary,

ADDENDUM

ADDED control of up and down light

There is a startup readout from the MCU board
all comms are on a single UDP port, nothing else...
The motion commands all but rotation are absolute
rotation is a bit different.
all the on/offs are bitwise...

[zszabo]

Great work Glen !!!!

[glenenglish]

Now is Open PNP time....

What are people going to be running- Linux or Windows ?

The local PC will run either... you could even dual boot it.

[zszabo]

As Im Altium/Win user I preffer Win.

[thommo]

There's no need to complicate the 920 world anymore I feel - Windows is likely to be more 'universally understood' by a broader range of ultimate users, so I say stick with Windows for sure.

Now is Open PNP time....

What are people going to be running- Linux or Windows ?

The local PC will run either... you could even dual boot it.

[glenenglish]

Will be both.  as my OpenPNP runs here on linux...

Will be a java HAL  with network loopback interface to C++
so will work for win32 or pthreads.....

needs to be at least windows because it will NEED to run on the QiHe windows box- HOWEVER  it is only an N2600 ATOM and this represents some bottlenecks for good fast vision processing, so we will have to wait and see.  It is an older Atom without many of the more (desirable  and modern) MMX/ SSE extensions....

[ar__systems]

All of the atoms came out way later than MMX-SSE appeared, so they all have all of these supported.

[anfang]

On the "Sys Config" screen there is a "save" button towards the bottom. Does anyone know where that saves the settings? I know you can import/export. I took a quick look with ProcMon and didn't see anything obvious. If I click Save, there's a ton of registry read/write, but they are to microsoft keys.

Would someone mind exporting their Sys Config and sharing it? It is an easy to read text file. I know I'll have to re-do most settings. That is OK. But I have hosed mine pretty good.

Thanks

BTW, 0603 parts pick up fine on the head made for 0402.

[anfang]

a morning's work... 6 hours to be exact

Very impressive!

BTW, ProcMon (microsoft sys internals) will capture all the UDP send/receive at the app level. Pretty handy in a pinch

[glenenglish]

The new 16mm thick feeder plates arrived this morning from QIHe,

I tried a range of feeders, new, old, FV, CL, vibration, - they fit all the feeders perfectly and precisely.

very good QiHe ! This was all at QiHe's expense. So, bravo.

I will fit them into the machine after lunch and then advise QiHe if they are all right, to proceed...

[glenenglish]

I've now got an app going that runs on the internal PC on the qihe TVM920. IE the TVM920 PC application is no more.

it talks directly to the TVM920's microcontroller PCB. (UDP port 8701)
It's the TCP  socket interface to OpenPNP, which I think you  will want to run OpenPNP on a stronger computer than the internal N2600 (AR_systems - it does not have SSE4.x )
it's written in C++ Builder. but I am using no special C libs.

Just working through interface stuff with openPNP right now like feeder opening, pick, advance etc . all very basic . maybe a few more hours of work.
see attachment

UPDATED

[dpenev]

Hi Glen,

I am very new to OpenPnP but just to get the concept.

Do we need a GUI at the TVM920 Atom board?
Without having experience with OpenPnP I feel running it on an external powerful PC
and heaving Ethernet connectivity to the TVM920 will be pretty convenient.
(Porting everything to the internal Atom board will be always an option later I think)
   
Or is this GUI you have shown mostly for convenience during your OpenPnP integration?

Thank you!
Dimitar   

[mikeselectricstuff]

Hi Glen,

I am very new to OpenPnP but just to get the concept.

Do we need a GUI at the TVM920 Atom board?
Without having experience with OpenPnP I feel running it on an external powerful PC
and heaving Ethernet connectivity to the TVM920 will be pretty convenient.
(Porting everything to the internal Atom board will be always an option later I think)
   
Or is this GUI you have shown mostly for convenience during your OpenPnP integration?

Thank you!
Dimitar

It would be best to run it on the internal PC if possible, and if not, an external dedicated machine. Sharing a machine running PnP with anything else is very risky, considering the potential loss of parts if a job stops and can't be continued.
 

[glenenglish]

My application 'TVMdriver program'  is the intermediate between OpenPNP (generic interface) and the TVM920 internals.

In the TVM920, there is a microcontroller PCB  that deals with everything. The internal  ATOM based PC runs QiHe's application.
Communications between the two devices is by ethernet UDP. In the OpenPNP system, we do not use the QiHe PC application at all.

The TVMdriver program has a socket interface (ethernet) to the Microcontroller PCB, and  another socket interface that connects to OpenPNP.

The TVM driver program has some manual controls, but these would NOT normally be used for open-PNP.
The only reason there is a GUI on the TVMdriver application is so that I could test all the functions and interface. It would not normally need to be accessed or used. All control in Open PNP mode is from the OpenPNP  GUI.  The TVMdriver GUi is only for test and verification. (but- the TVM driver can operate at the same time as OpenPNP no problem - there are multiple message queues and three threads) .

 It is my opinion that the internal N2600 ATOM PC in the TVM920 does not have sufficient power to provide a good user experience with OpenPNP busy.

So, there are two options
1) Run the TVM driver on the internal PC , and run OpenPNP on a external PC. Connect via LAN. ethernet.

2) change the internal Mini  ITX Atom N2600 PC to something like a Mini ITX (1151 pin cpu) with something like a G4400, or G3900 CPU.  So total spend would be about USD200. Of course you could put a small Intel i3-6100 in the same Mini ITX motherboard (USD99)  for another $100 etc... But I think  2 core, 4 thread processor is a good idea, even a low end processor. Support for SSE4.x is likely useful for the Vision stuff.  The low end older processors don't support these multimedia extensions.

I think an external PC is just easier, but some people might like the all in one integrated solution...
The internal PC has Chinese Windows . You cant do much with it, so you might as well just run the TVM driver on it.

[mikeselectricstuff]

 It is my opinion that the internal N2600 ATOM PC in the TVM920 does not have sufficient power to provide a good user experience with OpenPNP busy.

What do you mean by "user experience" ?
The only vaguely taxing thing ought to be vision - everything else is UI, almost all of which is only needed when not running a job

[vonnieda]

 It is my opinion that the internal N2600 ATOM PC in the TVM920 does not have sufficient power to provide a good user experience with OpenPNP busy.

What do you mean by "user experience" ?
The only vaguely taxing thing ought to be vision - everything else is UI, almost all of which is only needed when not running a job

Vision is not particularly taxing in OpenPnP. There's really very few actual vision operations and most of them are pretty simple. The most CPU heavy thing in OpenPnP is processing, rendering and displaying the images from the cameras. I typically tell people to run their cameras at 5 FPS or less to keep performance high.

It should be noted that OpenPnP has *never* had any optimizations done. I generally try to write performant code, but it's pretty much at the bottom of the priority list since most people run OpenPnP on powerful, modern desktop PCs.

[glenenglish]

right. machine interface  done....
not bad for an RF/analog man I guess, I have to do computer crap every now and then.

even got the mapping of the front three green, red, yellow buttons

and the command to lock/ unlock the cals and NVM

I'll add into the driver program the ability to pull out ALL the (cal) values out at boot , just like the qihe program does. (48 ints)
I want to see if they vary much between machines and understand any variations

[anfang]

I poked about a bit on the vision in the TVM920. I recall others had too on this thread, but I cannot find it now.

In the TVM920 is an analog capture card based on Philips SAA7134. It opened in a few lines of AForge in c#. At first I though there would be two streams, but instead it's a single stream and you just switch the camera input mux and keeping inhaling the same stream.

The native resolution looks to be 320x240. In the attached, I grabbed the 320x240 image, did a high quality upsample to 640x480, converted to grayscale, converted to binary (black/white based on pixel thresh), did a blob detection, recognized a quadrilateral, determined the corners and computed and drew normals.

At 640x480 it takes about 25 mS on my desktop without any optimization. Tomorrow I will try the Atom in the TVM and report the detection time there.

The accuracy is a bit better if I upsample to 2X resolution. If it don't and process at 320x240 it's still pretty good, and the recognition time is about 1/4 at 7 mS.

Placing at 3600 parts/hour is about 1 second/part. Even if the atom takes 100 mS at 640x480 it's probably pretty negligible. I think 0603 doesn't even need vision. So the question is mostly moot.

I really like the idea of the current PC running everything so that I could go back and forth between QiHe app and whatever comes next. I browse the web over a USB wifi dongle while the machine is placing now, and it's never stuttered a bit.

PS. In the attached picture is an 0603 capture run through the primitive recognizer noted above.

[dtf]

Nice. Very nice.