A Filter That Failed at 950 nm
On a Tuesday morning in March 2024, I measured the same 47-822 1000 nm shortpass filter for the third time. The curve on the spectrometer screen dropped at 950 nm. Not 1000 nm.
I tested five out of the eight filters in the batch. All of them had the same issue—cutoff point about 50 nm early.
We rejected the batch.
It was my third rejection in four years as quality manager at our laser shop. The first two were minor—a batch of lenses with surface flatness below spec, and a batch of apertures off by 0.2mm. This one was different. It affected a new client order for acrylic cutting.
Let me back up.
From metal to acrylic
Starting in late 2023, we kept getting inquiries about cutting acrylic sheets. Signage. Display stands. Protective covers. Light boxes. The first question was usually "how thick can you cut?" The second was "will the edges yellow?"
Cutting acrylic looks simple: set the power, pick the right lens, adjust the air, and you're done. That's the common oversimplification. In practice, edge yellowing comes from heat buildup at the cut point, not from the laser settings themselves.
To monitor quality live, we added a camera system to our laser cutter gun. The camera looks at the cut through a shortpass filter—the filter blocks the 10.6μm laser reflection and lets the camera see only the visible light from the cut edge.
That's where the 47-822 came in. 1000 nm cutoff, standard stock item. I checked the spec sheet: 1000 nm ± 5 nm. Looked fine.
The trouble was in the phrase "looked fine."
The assumption that cost me two weeks
We bought a few standard Edmund Optics lenses for the focus system and that 47-822 filter. I didn't re-test anything on arrival. My assumption was simple: Edmund Optics is a major brand. Standard stock items get tested before they ship. I don't need to repeat that step.
"Why test standard parts from a big name?" That was my exact line at the time.
That assumption cost me two weeks.
After installing the first batch, the camera feed came back too dark to resolve the edge. I assumed it was an exposure setting and spent two days adjusting. Then I thought it was the lighting angle and repositioned the source. Still no good.
Honestly, that part was frustrating. You know the problem is there, but you can't quite grab it.
Then I finally thought to test the filter itself.
One measurement and the answer was obvious. The cutoff started at 950 nm. Passband transmission was 12% below spec. The camera wasn't getting enough visible light through the filter, so the image was dark.
The gap in the supplier's sampling
I contacted the supplier. The sales engineer was polite. "Could be batch variance. Let us check." Three days later: "Our outgoing inspection sampled three pieces. All passed."
"Which three?" I asked.
"Pieces three, five, and seven."
"I tested piece one."
There was a pause. "Let me have the factory re-test the batch."
Two days later, the answer came back: of the eight filters, two passed and six failed. The out-of-spec pieces hit cutoff between 950 and 980 nm.
That's the hole in sampling. The pieces they tested passed. That didn't mean the batch would.
The supplier agreed to replace the batch. This time I asked for a full piece-by-piece test report. When the new shipment arrived, I tested every filter on the spot with our spectrometer. All eight passed, with cutoff points between 998 and 1002 nm.
Once the new filters were in, the camera feed cleared up. Acrylic cut edges went from fuzzy to sharp, from burred to smooth. A small moment of satisfaction—two weeks of chasing the problem, and it finally lined up. The client saw the sample and said, "That's what we need."
Three rules that came out of it
That experience changed our procurement process. We wrote three rules into our SOP:
One: a brand name is not a free pass. Edmund Optics has a solid reputation, but there are too many variables between production and shipping. A major brand lowers the odds of a bad batch. It doesn't eliminate them.
Two: don't skip full inspection on critical optics. Supplier AQL sampling is fine for general industrial parts. For optical components, one bad piece can tank the whole system's performance.
Three: write our own specs, not "industry standard." We used to say "meets industry standard." Now we say "1000 nm ± 5 nm." The supplier ships to our spec, and if something goes wrong, we have a document to point to.
Testing the process on a portable laser rust removal machine
This summer we bought a portable laser rust removal machine for cleaning old molds in the shop. When it arrived, I applied the same rule: get the factory test reports for the optical components, then re-test on arrival.
The field lens had a focal length deviation of about 2%. Not catastrophic on its own—but a portable laser rust removal machine has tight depth-of-focus requirements. A 2% deviation means uneven removal depth. The supplier swapped the lens and we were good.
Without the re-test habit, that deviation would have surfaced at the customer's site. The rework cost on a portable laser rust removal machine is a lot higher than on a single filter—and that's before counting the hit to customer trust.
Back to the question: how do you cut acrylic sheets?
Most people answer that question with "set the laser power right." That's not wrong. But it skips a critical assumption: optical system consistency. Lens focal length, filter cutoff wavelength, camera resolution—if any one of those drifts, cut quality goes with it.
I now spend about 10 minutes with clients explaining this before we even talk about an order. I'm not pitching products. I'm drawing the line between specs and system behavior.
"An informed customer is the best customer." The first time I heard that, it sounded like a sales line. Now I think it's just practical. A customer who understands the basics won't just compare prices. They'll ask, "What's the focal length tolerance on this lens batch?" instead of "Can you knock 10% off?"
One saves money. The other saves rework.
Side note on spare parts: a single 47-822 filter runs about $150–250 based on Edmund Optics' public pricing for standard stock items in 2024 (verify current rates). Compared to two weeks of downtime, that's not much. Keeping a spare isn't optional—it's part of quality control.
As for the rejected batch: the supplier took it back and ran a new production lot. We covered expedited shipping. They covered production. Nobody got a free ride.
That's a fair outcome.
Leave a Reply