Visual fault finding, camera in endless loop

Hi

Working on a switch mode right now, and sometimes things blow up. After the magic smoke, it takes some time to find the cause and what blew up

I was just thinking about adding a camera to the lab, recording endless loop of the bench, so in case of a fault I could rewind and if lucky see the actual development of the fault

Any ideas?

(I have a NAS server that records endless loop for security cameras, but that is only 10 frames per second)

I would add a thermal camera alongside the visual camera.

Lighting may be critical to spot smoke. If you’re lucky, you’ll also see things glow or flash.

Good point

I just bought the DytSpectrumOwl:

https://www.amazon.co.uk/dp/B098JSNSC4?ref_=cm_sw_r_cp_ud_dp_8CA9ABFEYVV89CW0X4CJ

In fact the CA20 model

Really nice, great resolution, nice SW

image

I am not sure that can be setup in an endless loop with only 1minute cache

I wonder if you could use a dashcam?

Yes, I thought the same. We don’t use them in Denmark, so I don’t know how they operate. I guess it is just a loop, with a sensor to detect a collision, or maybe a button also, to transfer video to the flash storage

Can you use python and opencv to record alternate files in a folder, like a double buffer? https://youtu.be/ChWvD3C7SgE

Some have collision detection, but not sure on the details. Generally they just record video to a memory card in 30 minute segments, and when the card is getting full they remove the oldest video. Mine records at 1440p 60fps, and will store about 5 hours of video on a 64GB memory card.

My thoughts are, in order:

  1. GoPros do this. Since like the 4 or so, they all have a loop recording mode, I think mostly used by motorcycle riders. Helmet cam meets dashcam.
  2. It won’t probably do that much for you. The moment you blow up a board, it’ll take you a minute or two before you think to hit the HiLight button on the gopro, then like 10 minutes to dump the footage, then another 10 minutes to figure out where in the footage the board actually explodes, even though you know the absolute time to within a minute or two. Then once you actually see the board roast on video, what do you do with that? Do you play it in reverse for another 30 minutes until you find the moment you bumped the power supply to 6 volts instead of 3.3 or whatever?

All of this is from my experience with having a Wyze cam on our front door, so the lab use case may or may not be as bad. But even when we know almost exactly when something happened, it’s SUCH a tedious process to go back and try to find the event in the footage, and it’s often underwhelming or unrevealing when you do.

Yes, I know it is a long shot. And even if I find the sequence of what fails, I cannot be sure I can relate that back to the root cause. I was just thinking that there might be a way to optimize fault finding, since I find it to be a tedious grind

1 Like