Neural network powered PCBa Visual inspection

[jpedroc]

We are developing a neural network capable of detecting faults on the electronics components placed on a PCB just from a photo taken of any PCBA.
 We are finally in the Alpha version of it: AgnosPCB.com

Please feel free to take a look at our site and the current visual inspection tool: the web App https://app.agnospcb.com/

This APP is currently in the ALPHA version. Within months, we will release and full Computer/Android version of this APP

This ALPHA version can detect missing/misplaced or rotated elements on any PCBA. It does not matter the type of elements placed/ PCB architecture/ color or size of the components.

How does it work?: The neural network uses a reference PCBA photo (a PCB that has been previously checked) and it will compare that reference with all the photos of PCBAs to be inspected.

The visual inspection takes seconds. Any mid-range smartphone can be used to take the photos (>8 Megapixels).
There is a short video explaining how to use the WebAPP here*
This fast visual recognition tool is extremely useful for low or medium quantity PCBAs batches

If you were interested in trying our APP, we will happily provide a USER ID (needed to run the WebAPP) for free to inspect boards. There is not equipment investment required.

The tests can be made using just a smartphone’s camera.

This service is still in an ALPHA (early testing phase). The more boards are sent to the neural network, the more it will improve identifying mistakes. An ALPHA version means as well that we can add features to the APP as they are requested by the users.

Anyone interested on trying the neural network, please contact me at: agnospcb@jjrobots.com

[OwO]

Are you reinventing AOI?

I mean, most of these are things functional testing would catch. I don't really care about checking for wrongly placed components, wrong rotation, or short circuits. Those will be rejected during functional testing. What I need is something that will find bad joints/intermittent faults that will lead to reliability problems. Pictures of the board taken from the top or just one angle is useless. I'm pretty sure existing AOI systems already take pictures from many angles and look for soldering defects, but it still isn't good enough and x-ray inspection/shock&vibration testing is often needed.

[jpedroc]

We assemble our own PCBs and the visual inspection is the first step after taking the boards out of the oven.
 Many times, for short or medium PCBs batches, programming the AOI is very time-consuming. So we decided to create a neural network capable of telling were tombstones, soldered pins, missing or rotated elements are, within second. A technician needs minutes to do the same task. This is a helping tool a AI-AOI

[OwO]

Ah ok, so you just wanted a simple sanity check before the boards go to actual programming and testing? In the cases I've worked on the boards usually go straight to functional testing out of the oven. Unless you have a high defect rate, I don't see how this would improve productivity. Feel free to ignore my opinions though as your process requirements are probably different.

https://www.sakiglobal.com/automated-optical-inspection-aoi-3d-technology.html
This is what the current state of art in AOI looks like. 3d inspection is essential as the whole point is to catch defects that functional testing wouldn't catch. If you want to improve on that with neural networks, I think this is the basis you should start from.

[jpedroc]

Yes, that is the idea. A quick check that will discard PCBs in the same way a technician would do, but much faster. Our experience says that a persons needs from 40 seconds to 4 minutes to check the general integrity of an assembled PCB. The neural network needs from 3 to 8 secs.

Of course, we have an AOI but for small /medium PCBs batches it does not worth the time to program it

[SMTech]

I like the idea, with low volume work you might have a whole bunch of parts you fitted as a separate process to the automated P&P and those designs can easily be too complicated to spot silly mistakes like missing or swapped components by eye. Many of those faults might get through functional testing, its quite a feat to create a functional test that catches every function let alone build errors that won't necessarily affect function like a missing cap or tie-up.

What we're looking at here is something that competes against FAI machines - flatbed scanner based systems that cost around £10k.

[SMdude]

This does look like a good idea indeed! I'm surprised that there isn't something like this out there already.
I do small batches of densely populated boards and it does take a bit of time to inspect them and it is easy to miss a part or 2.
A quick photo that shows up the errors almost instantly would be nice.

[jpedroc]

Thanks
 Everyone interested on trying the neural network, please email me to agnospcb@jjrobots.com
We are offering a trial for 200 PCBs inspections for free. Feedback and thoughts are more than welcome ;-)