[stewi]

Implementing some of Nails motor sources in a new project and thread:
http://www.cnczone.com/forums/rc_rob...ml#post1013780

[scsi]

Haven't posted videos lately, so here we go. I'm making some good progress in PPM vision development. On the video below (do a full screen), you can see my test program that performs basic alignment of an 0805 resistor on the up-looking camera. Pay attention to the angle alignment indicator, it seems to be limited by motor resolution, which is currently set to 40x microstepping or 0.045 degree/microstep.

[ame=http://www.youtube.com/watch?v=9bODRom1jrs]Camera Alignment Test 0805 - YouTube[/ame]

I'm using a 3rd-party industrial grade vision library, which is not cheap but kind of affordable. This was just a mock up of the alignment process to evaluate the capabilities of the algorithms. Turns out it is very much possible to achieve a +-0.050 degree alignment and I'm sure I can do better with finer microstepping.

Since the test seems to be rather successful, I'm going to go ahead with the development of the PPM software that I'm planning to market when all is said and done. Under $1K should be doable. This is with cameras and software licenses. And yes, that camera on the video is a cheap ($90) USB camera with a decent Tamron lens on it. Same results will be achievable with pretty much any webcam, which will help to bring the prices down considerably.

I'm also using an 8-axis motion controller that costs $245 and performs beautifully. Can drive steppers in open and closed loop as well as servos and other stuff. Leaps and bounds beyond smoothsteppers or DeskCNC gizmos.

Now, if you want to talk me out of it, please say something here. If not convinced, I will start a separate thread to discuss requirements, feature requests and other stuff. It will be at least six month before I can produce a meaningful beta version, so you guys think about the features you need and I will try to accommodate as much as possible. The target platform is Win32/64, no Linux. Sorry about that.

And almost forgot. The idea is to develop a software package that can be used to turn a CNC router table into PPM machine that can handle 0201's and 0.4mm pitch packages. I see lots of great ideas in this thread on how to build the machine and attack mechanics of it. Nothing meaningful so far has been demonstrated in software area. Madell's package works, but is expensive and uses expensive cameras. Also, I'm hoping to meke my software a lot more user friendly and flexible to accommodate various machine configurations.

P.S. Yes, I looked closely at all the open source CV libraries out there. Unfortunately, there is nothing there that can be used as is. Most of the advanced algorithms that are included in OpenCV, like SURF, SIFT or Haar are heavily patented and cannot be used without permission. They also do not provide the accuracy necessary in our business. NCC correlation can be adapted to detect rotation, but performance is nowhere near to where we need to be. Commercial libraries are fast, but expensive.

Regards,
SCSI

[flippersplace]

@SCSI, I would be very very interested. And by that I mean as in paying customer interested

Please start a separate thread

And while I'm online, a big shout to Stewi, who has helped me significantly with a purchase of 10 x used Siemens feeders for my own machine. Which will be a major improvement

I'm bogged down doing some time critical PIC32 firmware development on another project at the moment so once that's done will detail the feeders on my own thread

[nige_uk]

I've been lurking on here for a few days but have only just signed up in order to post.
I have now read all 54 pages of this thread! Very interesting information indeed.

To get up and running quickly I purchased an entry level Madell P&P complete system last year with the software and nozzle arrangement mentioned previously by scsi. Given the price of their equipment it was no suprise to me that I had to make alot of upgrades & alterations to the machine in order to improve its accuracy to an acceptable standard including a complete re-wire but I'll leave that whole saga for another time (though I will say that Madell's support is very good).

Despite my best efforts, I still think the camera system supplied with it is reducing its potential. On their website they show high end industrial cameras on their machines but that is not quite what they supplied with my machine.

@SCSI, Could you please advise what up looking and down looking cameras you have found to work well with the madell software, as well as where you purchased the LED illuminators from for your machine (or if you made them, then how many LEDs / what diameter you found works well)? I see you mentioned Tamron in an earlier post but you did not mention which specific model(s) or which types of lens for up and down you have found work well.

Many thanks to scsi in advance and best of luck to everyone with all their projects.

[scsi]

...@SCSI, Could you please advise what up looking and down looking cameras you have found to work well with the madell software, as well as where you purchased the LED illuminators from for your machine (or if you made them, then how many LEDs / what diameter you found works well)? I see you mentioned Tamron in an earlier post but you did not mention which specific model(s) or which types of lens for up and down you have found work well.

Many thanks to scsi in advance and best of luck to everyone with all their projects.

Thought I reply to you in PM, but this may be interesting for the folks on the thread...

I believe Madell offers upgraded cameras as an option now. I suggested them to upgrade their optics and cameras some time ago and recommended Sony XC-ES30 or ES50 as that's what I have on my machine.

These are regular analog cameras with no fancy features but cost quite a lot because of the brand. Pretty much any B/W analog camera will do as long as it has 1/3 or 1/2" CCD and C-mount for the optics. There is a lot of these on eBay at a fraction of the cost and I suggest you look for a camera that has BNC output and more or less usable power input connector. It will be easier to interface it with Dalsa grabber cables. Those Sony cameras come with a standard high-density 12-pin Hirose connector, which is very hard to find. They are intended to be used with a standalone power supply or a grabber that has this same type of connector and can also supply 12VDC power to the camera.

As far as optics goes, I'm using 25mm 1/1.6 Tamron lens for both cameras simply because it's very compact and produces great picture. They also have locking screws for focus and iris, which is a nice feature on a fast moving machine. They are abouth $200 new. Search for "25mm C-mount" and you will find a bunch of them on ebay. Does not have to be Tamron for VGA resolution. Anything with manual focus and iris that goes down to 1/8 or 1/16 will do the job just fine.

You will also need some extension rings to bring close objects into focus. 5mm ring seems to work well for the down-looking and 10mm for the up-looking cameras. "c-mount ring" search will work for that. In case if you need wider field of view, go for 16mm or even 12mm lens. You can attach up to six cameras to that Dalsa grabber and use different ones for different parts for the up-looking video alignment. Try 16mm for the large IC's and 35mm for small passives. One problem with the up looking cameras is that they usually have cables sticking out the back and it will be quite challenging to mount them properly under the machine table. I ended up buying some industrial no-name camera on eBay that have right-angle lens mount, so that you can lay them down horisontally. That's the one sitting under the table that you can see on some of my videos.

The cheap ring lights for microscopes are in the $50-$100 range and most of them work fine. Look for a smaller one with 40-50 LED's that would be able to clamp onto the lens, unless you figure another way to mount it. Unfortunately, I don't have a link for the lights I'm using. I got them off eBay two years ago and lost track who I bought them from.

And by the way, since I'm now busy trying to develop my own PPM software and vision system, I won't be needing Madell's stuff for quite some time. Shoot me a PM and we'll see if I can maybe offload some of my spare cameras and lenses to you.

Regards,
SCSI

[dandumit]

The cheap ring lights for microscopes are in the $50-$100 range and most of them work fine. Look for a smaller one with 40-50 LED's that would be able to clamp onto the lens, unless you figure another way to mount it. ..................
P.S. Yes, I looked closely at all the open source CV libraries out there. Unfortunately, there is nothing there that can be used as is. Most of the advanced algorithms that are included in OpenCV, like SURF, SIFT or Haar are heavily patented and cannot be used without permission.

@SCSI - huge amount and verified information's. Thank You !

Regarding ringlight : I was attending to a COGNEX seminar - they have spent about 1/4 from seminar time to demonstrate and emphasis on importance of lighting : even in their handheld cameras have included a composed lighting system that sends out direct , tangential, indirect and 4 ways configurable light.
More than that on all images that I have seen for industrial machines have lighting under an angle. Probably this is improving results of recognition algorithms.

Regarding OpenCV - are really needed so fancy object recognition algorithms ?
I was thinking to build the feeder from a milky-plastic illuminated from bottom and recognize the component with up-looking camera with a Blob and CvMinAreaRect. Also a Contour function would do the trick. But again - I am just a beginner here in image recognition.

Kind REgards,
DAniel




REgarding

[scsi]

Regarding OpenCV - are really needed so fancy object recognition algorithms ?
I was thinking to build the feeder from a milky-plastic illuminated from bottom and recognize the component with up-looking camera with a Blob and CvMinAreaRect. Also a Contour function would do the trick.

Yes, those simple algorithms may help to locate the part or the pocket on feeder tape if the lighting is right, but they will have trouble telling apart an empty pocket from the one with a part in it.

For a real PPM action the component rotation problem needs to be solved. I wasn't able to find a free algorithm that would resolve even five degree rotation reliably and went with the commercial library. Still amazed with its capabilities. Today I changed microstepping on the nozzle motor to 1/256 and now the parts align to within +-0.005 degree. That's ten times better than on the video I posted yesterday and it shows 20x better angular resolution then advertised by the library vendor. All that with 640x480 webcam. Amazing!
It's also incredible to see how a regular 1.8deg stepper motor can hold a position at 1/256 microstepping.

[dandumit]

Today I changed microstepping on the nozzle motor to 1/256 and now the parts align to within +-0.005 degree. That's ten times better than on the video I posted yesterday and it shows 20x better angular resolution then advertised by the library vendor. All that with 640x480 webcam. Amazing!
It's also incredible to see how a regular 1.8deg stepper motor can hold a position at 1/256 microstepping.

This is really incredible ! - I was reading many times MAriss Freimass posts (in this site) and in his opinion there is no point to go under 1/10 Microstepping (that's why there are no Geko drives under 1/10). Also I have read that stepper motor has to be fabricated with very high precision on stator and rotor...

Would you please tell us what library have you used ?

10x
DAniel

[scsi]

This is really incredible ! - I was reading many times MAriss Freimass posts (in this site) and in his opinion there is no point to go under 1/10 Microstepping (that's why there are no Geko drives under 1/10). Also I have read that stepper motor has to be fabricated with very high precision on stator and rotor...

Well, it works and I am myself surprised quite a bit seeing how a regular Nema 14 motor can hold position at 0.05deg, let alone the 0.005 degree... The motor driver is a DM422C from Leadshine, running the motor at 0.5A

Would you please tell us what library have you used ?

I hate to do it, but won't tell. I will tell which ones I'm not going to use. Took me several months to find a library that costs less than $500 for a run-time license with just one algorithm included. The rest are way up there in the thousands and most only work with some upscale frame grabbers. Cognex works with 3rd-party images, but they want $4,500 for a run-time! Matrox is even scarier. Dalsa is more affordable, but not quite there yet. And the list goes on...

[scsi]

Forgot to mention that Mariss is probably right about the usefulness of fine microstepping. In most applications, the motor won't be able to hold its intermediate position if it faces some meaningful load, like in CNC applications. In our lucky case there is no resistance at all and the rotor just hangs in there.

Madell is using 1/40 microstepping for rotation and I believe the value is just a tradeoff between alignment speed and positioning precision. DeskCNC motion controller tops-out at 125,000 SPS and they wanted to align the parts quickly.

The motion controller I'm using can go up to 2.5 MSPS, so there is no tradeoffs there.

[phomann]

Microstepping won't provide any more accuracy. It's purpose is to provide smooth movement rather than "Clunking" from pole to pole. This helps to reduce any shifting of the component on the pickup nozzle.

Cheers,

Peter.

[Khalid]

Microstepping won't provide any more accuracy. It's purpose is to provide smooth movement rather than "Clunking" from pole to pole. This helps to reduce any shifting of the component on the pickup nozzle.

Cheers,

Peter.

hi Peter,
I think micro stepping provide high resolution in the system which the scsi require..

[dandumit]

I hate to do it, but won't tell. I will tell which ones I'm not going to use. Took me several months to find a library that costs less than $500 for a run-time license with just one algorithm included.

I respect your position ! In fact, at some point each one of us has to make a little profit.
More than that, you have published until now a lot of useful and directly applicable information. If each one of this thread would do the same probably each one would had an working machine at home and chinese ones would be under 1K$.

Kind REgards,
DAniel

[phomann]

hi Peter,
I think micro stepping provide high resolution in the system which the scsi require..

Yes, it provides higher resolution, but that resolution is not accurate. If you have a 1/40 micro-step system, or any micro-step system, the angle between each micro-step will not be constant.

The purpose of micro-stepping is to provide smoother motion. Think about it, there is a relative large detent force trying to hold the rotor at the pole. I doubt that providing a single step in a 1/40 microstep driver will move it off the pole. Continuing to give it more steps will cause it to spring off the pole.

Try attaching a long needle to the motor and measure the angle each micro-step produces. You probably won't be able to see 40 micro-steps (resolution), let alone the accuracy of 0.45 degrees per microstep.

As I said, microstepping is only there to get smoother motion.

Cheers,

Peter.

[ad_bfl]

P.S. Yes, I looked closely at all the open source CV libraries out there. Unfortunately, there is nothing there that can be used as is. Most of the advanced algorithms that are included in OpenCV, like SURF, SIFT or Haar are heavily patented and cannot be used without permission. They also do not provide the accuracy necessary in our business. NCC correlation can be adapted to detect rotation, but performance is nowhere near to where we need to be. Commercial libraries are fast, but expensive.
SCSI

Question, to what degree of precision do you need?

OpenCV is amazing capable for a "generic" image processign library. Identifying corners, linear segments etc to identify a part & then determine it's position given that you should know what your looking for and not doing a from scratch identify and match approach would not too hard to implement.

Given that you have control of the lighting intensity, illumination angle, maybe add a background reference, I bet the system could be tuned pretty darn tight and still operate fast enough to not be a limiting factor.

The latest release (2.3) actually has a good tutorial, plus there is a new book just published that is a big help for getting the basics working.

However, if you already have a library pre-built and optimized for this use case, then by all means use it till it's cost gets big in your BOM.

[stewi]

The cheap ring lights for microscopes are in the $50-$100 range and most of them work fine. Look for a smaller one with 40-50 LED's that would be able to clamp onto the lens, unless you figure another way to mount it. Unfortunately, I don't have a link for the lights I'm using. I got them off eBay two years ago and lost track who I bought them from.

And by the way, since I'm now busy trying to develop my own PPM software and vision system, I won't be needing Madell's stuff for quite some time. Shoot me a PM and we'll see if I can maybe offload some of my spare cameras and lenses to you.

I like to sell (make offer) 2 digital camera/vision systems as well. DVT is now Cognex

[Khalid]

Peter,
I agreed... The more the microstep the less the accuracy between the microsteps but overall the system become more accurate...

[scsi]

Guess I will attach a needle and have some more fun with microstepping tonight.

[dandumit]

Guess I will attach a needle and have some more fun with microstepping tonight.

Discussion went away from PPM machines. IF you want to see errors on positioning of motor shaft. Put on shaft a laser pointer and measure positions on a wall. With a needle you won't see a big difference. (a similar discussion was on this site regarding a3986 stepper driver)

[RomanLini]

Guess I will attach a needle and have some more fun with microstepping tonight.

I did that back in 2000 when designing the Linistepper, with a needle 1 metre long. Some notes here; http://www.romanblack.com/stepper.htm

Microstepping does give quite a lot of improved accuracy, especially at light loads (as SCSI has said). Modern motors have much more precisely manufactured pole construction than early motors, and the microstep positions are averaged over 50 poles (on a typical 200 step hybrid motor like most of us use) so per-pole errors are largely negated by averaging.

The Linistepper was designed with microstep currents that gave accurate positional microstep sizes, and now most stepper driver chips etc also do this so all the microsteps are of a similar and comparable size.

Of course there is no guarantee that every microstep will be exactly the same size, but they are quite good and the positioning accuracy of an unloaded motor is generally better than +/- one microstep, which I believe SCSI was also stating, in his measured ability to turn the motor to hold within 0.05 degree (smaller than one microstep repeatability).

The view that "microsteps don't increase accuracy" is a very 1970's attitude and with modern motors and drivers they do increase accuracy by a huge amount over full step and half-step systems.