Hi, I am an electronics engineer by profession. During my course of work I found that there is a strong need for understanding the design of electronics behind the PCB boards for troubleshooting and performance enhancement. I wish to create a model using Computer Vision and Deep learning for generation of circuit schematics, PCB layout drawing from PCB boards.Those interested can share your thoughts.
@mgautam Personally I think one of the biggest challenges with this task is to ask if there is a way to do so without X-Rays of the board-- Not only to capture surface features but hidden layers (in the PCB, not neural network sense), via’s etc. Further such a corresponding dataset would be valuable for training-- Though I’m not sure I can think of one openly available.
Or, are you thinking of only examining (limiting your set) single/dual sided boards ? That task, at least, would be much more feasible to use CV alone to deduce your netlist.
This effort will struggle to cope with multi-layer circuit boards. It also will not capture important details regarding the effect of the circuit board material on the a circuit’s performance.
Also be extremely careful that such a project does not violate any entity’s intellectual property rights.
@Nevermnd Hi, Thanks for the reply. I wish to try it out for a dual sided board. Do you have reference to any project which has already been undertaken on this topic?
@mgautam Ah, honestly no-- But that is probably exactly for the reasons @TMosh points out;
I mean, for personal reasons I would say I enjoy hacking and pulling things apart to see how they work or re-engineering them for small projects. I mean if you purchased whatever it is you are working on, generally at least, I don’t think you lack the right to do that.
That said-- other than perhaps as an ‘interesting conceptual exercise’ or an open source project… I don’t think this is a concept that could ever be ‘sold’ without getting yourself into some hot water. I mean, really, if you yourself designed the board you’d have the netlist, schematic, layout and wouldn’t need to perform such an operation with computer vision.
Thus, in that case, I have to agree with Tom and for such a use case I don’t feel I can condone it.
Some entity’s survival strategies are to place IP traps everywhere, and it is so sad to see young people inventing obvious things themselves only to find out that somebody who was born before them already patented this line of thought.
If somebody has the skills to study, copy and apply stuff, we should celebrate these people, applause and support them - not to punish. We are monkeys who forgot how to put a person on the moon, and it is not the most sophisticated technology to retain out there. We can not even make the EU climate models run on GPU, because people who wrote the math could not pass the knowledge.
And now we are telling people to tremble in fear behind patent trolls, before the people even got a change to study the matter. How demotivating is that?
Active fear is a very hungry system process - you can not study and be creative with it.
Yeah… I mean honestly I am with @TMosh here; I admitted I enjoyed tweaking, hacking, even a little private ‘reverse engineering’… But at the same time we both (I think @TMosh ?) don’t want @mgautam to end up getting sued or in jail somewhere.
And again, not at all to put @mgautam down-- Us serious practitioners or those that have belief in these technologies are not about that. However for those of us that know a bit more a subject, all his original goal would be useful for is ‘copying things’, not learning things.
Thus I will even amend his idea into what I think could be a really good one.
So, again, let’s assume we don’t have access to X-Rays, yet magically we have access to a thermal sensor (i.e. as in camera to read cross board temps), and then also at minimum something like a ‘Faraday Cage’, where at least we can assume what ever RF signals we pull off the device are isolated.
So via CV perhaps you have the trace (without the company having to hand over all their IP, then you’ve got the thermal and RF layers on top of it)-- Now that would be a model, and that would be a business.
I mean @mgautam I think a lot of people would pay for a package like this. I probably can’t afford, but willing to Beta test.
I mean, ugh @abitrolly. Should I even acknowledge ?
Those in the land of ML are not all trolls but try to look out for one another.
I have a broken amp at my home, and my electronics skills a bit rusty, so I would really appreciate the AI+AR app to guide me how to fix it. Why the right to study and the right to fix is not a thing here? We already have a lot of waste, because companies need to sell things, but with AI systems that help people to learn and fix physical system there will be less of that.
@abitrolly I’m sorry, but I am not going to follow this inquiry much further. I mean the original poster @mgautam looked me up, and he seems smart and it was not like he had a totally bad idea, yet there are some warning signs.
I mean you speak towards ‘right to repair’ which I generally agree with— Yet unlike what the original poster was speaking about, so little of this happens at the ‘board level’. Usually what will happen is after a time your caps will fail out. Yet there is no AI for that. You have to really get in there with a meter and start poking around.
All the real action happens on the ICs, and so-- What, is he supposed to put them under acid and then the microscope and then decap them ?
I mean none of us are ‘dissing’ the original postern, yet just kind of stating ‘these are the challenges you will face’.
Hi All, I am an electronics engineer by profession. I am in the process of trying to build products that serves to solve problem points of people. Based on my experience, I feel that there is no point in copying a PCB board or electronic circuit because there are hundred of ways to achieve the same thing. In the age of SoC, this copying effort will be useless. My point is to create a generate Ai App that can generate schematics of a two layer PCB board based on images fed to it. This will be of great use to the society in understanding the functional issues, improve performance and reducing scrap. I am a software enthusiast, but new to Computer Vision, GenAi and Deep learning. If someone is really interested in my topic, do thinks for fun and improving skillset, let us join hands in building one. Let it be open source.
@mgautam My thought towards the RF/thermal stuff. Though I have not been through it myself I understand getting FCC certification is both very expensive/tough. I laud your learning and open source efforts regarding the tracing, and only was suggesting if you were able to design an ML/AI model for the latter… That would be a real product you could sell.
Hello all. I found this thread, and this forum because I discovered AI LLM development recently and I cannot tear myself away! I have built some special projects mostly in Google AI Studio and have begun deploying a self-hosted instance. Also I have a background in many technical and scientific fields, so I have been brain storming real life tasks and trying them out with the help of AI.
I have a pro grade DJ controller that has had its USB input jack broken. I had a chat with regular web hosted Google Gemini and was (surprised but not really) that it was able to predict all expected resistance, diode voltage drop, and voltage levels for every component lead i read to it. Far beyond that it contributed big picture insights, like how I should prepare for the possibility of replacing a near by IC. warning me that SMD desoldering and soldering can be tricky. That’s when it hit me! My friends use AI models to debug their code, or create code, and I already know that the schematic for my device is unavailable to the public. I am not an attorney but I don’t think that what I am doing here is against the law, civil or criminal. Here is my chat so far [edit - no links allowed]