I thought we'd found an enormous scratch on a bezel for our FR5 camera trial. Then I turned it slightly and the scratch developed the corner of the window behind me. So that's embarrassing. How do I stop this contaminating the reference set before we start tuning anything?
My bezel has a scratch shaped exactly like the window frame
TheoArcher0410 · 2025年3月16日 11:18 UTC
主帖中文翻译暂不可用，以下显示英文原文；回复保留原文。 查看英文原文
17 条回复
Can I ask who has inspected the actual bezel? Put the image labels aside until someone establishes whether there is a rejectable defect on the part.
16分Quality looked this morning. Accepted part. Apparently I'd photographed an excellent little mirror.
22分Keep it as an accepted example. Control what the surface reflects, then check whether a real rejected part remains distinguishable in the resulting images.
16分Don't delete the embarrassing pictures. They're useful examples of false labels.
8分回复 AlexChen1168
Agreed. Relabel them with the confirmed part status and explain why the original annotation was wrong.
16分We had a board behind the camera that everyone called temporary. Six months later it was still held up with packing tape, and cleaning staff kept moving it. Whatever you use for this trial, plan how the finished station keeps that background repeatable. Otherwise your training shift will spend its life putting the window reflection back where the software expects it. I had a photograph taped to the bench showing where the board belonged, which was our very sophisticated vision system for a while.
17分回复 LeahAli0234
That's a useful warning about the installation. For now, record the trial lighting and surroundings well enough to repeat the capture.
7分We've screened off the window for the bench trial. Much cleaner images, but the actual rejected bezel's fine scratch is harder to see now. Removing my window didn't magically build an inspection system.
11分Who chose that rejected bezel, and is its scratch the sort production actually rejects?
16分Quality chose it from a held batch. Real handling scratch, with its location recorded. Not something I attacked with a screwdriver for the demo.
24分Good. Is the vision supplier looking at both pieces together?
12分Any progress with the supplier's imaging trial? The accepted reflective part and the genuinely scratched one are the useful pair to keep comparing.
6分回复 LeahAli0234
And let the person who'll clean it see the proposed enclosure before it's made. Our tape arrangement started because the proper cover blocked access to the lens.
10分回复 LeahAli0234
Cleaning access should be part of that design discussion. Opening and closing the enclosure also needs to leave the imaging setup repeatable.
12分Have you paused threshold tuning while this is unresolved? No point polishing numbers against changing images.
10分回复 AlexChen1168
Yes. Supplier has both bezels and is comparing light positions before we pick hardware. Cleaning access is on their sketch now too. The window-frame pictures are staying in the folder with corrected labels.
12分Leave the defect-detection result open until they show repeated presentations of accepted and rejected samples. The cleaner background is progress, but you already found why that alone is insufficient.
12分讨论已关闭
此讨论已连续六个月没有活动，现已关闭新回复。 最后活动： .