Notes & strategies
An anonymized, growing wiki of how I tackle real machine-vision and inspection problems - the approaches I tried, what failed, and the lessons worth keeping. A reference for me, and hopefully useful to others.
No notes match your search.
Foundations
How I think about problems before touching hardware.
How I approach engineering problems
Most "technical" problems are really constraint problems in disguise.
The inspection pyramid
A defect travels through a long chain before it becomes a decision. The later you catch a problem, the more it costs.
Why requirements matter more than hardware
Buying hardware before understanding the requirement is one of the most expensive mistakes in engineering.
Learning through failure
Experienced engineers do not know all the answers. They have just seen more ways things can fail.
Detecting vs measuring vs classifying defects Coming soon
Three different problems that get lumped together as "inspection".
Lighting & imaging
Why contrast usually beats resolution.
AOI lighting cookbook
Lighting is often more important than the camera. Many inspection problems are actually lighting problems.
Overexposure as a tool
Some defects only become visible when you deliberately drive the image into saturation.
Multi-image inspection
One image almost never contains all the information. The best systems combine several.
Four lighting strategies compared Coming soon
A side-by-side of four lighting setups on the same difficult substrate.
Optics & resolution
Comparing optics, fighting depth of field, and pushing toward sub-micron.
How I compare optics
Comparing optics by magnification alone is one of the most common early-career mistakes.
Focus stacking in production
High magnification destroys depth of field. Focus stacking is never purely an optical problem.
Building a sub-micron inspection strategy
Near the resolution limit, everything matters - and the camera is usually the least interesting component.
Designing a calibration target
Off-the-shelf targets rarely cover the full range you need. So I designed one.
2D vs 3D: choosing the right dimension
Not every problem needs 3D. Not every problem can be solved in 2D. The best solution is often a combination.
Line-scan vs area-scan cameras Coming soon
When each wins, and the integration trade-offs nobody warns you about.
Applications
Anonymized case studies from real inspection problems.
Conductive substrate inspection
Anonymized case study: finding subtle defects on ceramic/metal power substrates with multiple lighting strategies.
Reticle and pellicle inspection
Anonymized case study: inspecting a transparent membrane over a patterned surface is harder than it looks.
Wirebond inspection
Anonymized case study: "measure the wire height" turns out to be the easy part.
Comparing laser-triangulation sensors
Anonymized lessons from evaluating several 3D profilers. The best datasheet rarely wins.
Evaluations & production
Proving where a solution stops working, not just where it works.
How to run customer evaluations properly
A successful evaluation proves where the solution stops working, not just where it works.
Every sample lies
Customers send their best and worst parts. The dangerous gap is everything in between.
Understanding production variation
Most inspection systems fail because reality changes, not because the software was wrong.
False defects: the real production killer
Finding defects is easy. Finding only real defects is hard.
Designing AOI systems for manufacturing
Lab success is not production success. Production introduces a whole new set of forces.
Designing around cycle time instead of resolution Coming soon
Why throughput, not pixels, should anchor most system designs.
Support & infrastructure
Troubleshooting, recovery, and documentation that survives.
Symptom is not cause
The real support skill is not fixing problems - it is identifying the correct problem.
Database recovery in manufacturing systems
When data is at risk, the first priority is preserving it - not troubleshooting.
Documentation systems people actually use
The solution usually existed already. Nobody knew where it was.
Self-hosting and infrastructure notes
Outside machine vision I build infrastructure. The recurring lesson: simple systems survive.