The 8,000-Piece Rejection That Taught Me to Trust the Zeiss Coordinate Measuring Machine

Tuesday, 9:15 AM

I was standing in the receiving inspection area holding a serological pipette that looked perfectly fine. Under the Zeiss stereo microscope, the tip was clean, the graduations were sharp, and there was no flash or parting-line damage. My technician said, 'This looks good. Are we sure about the CMM?'

I wasn't sure. That was the honest answer. The Zeiss Contura coordinate measuring machine had flagged the same dimension on 11 of the first 32 pipettes, but the readings were only 0.06 mm off nominal. The vendor's quality engineer said it was within industry standard. I put the entire batch on hold.

I work as a quality compliance manager, not a metrology expert. I review 200+ unique items per year—incoming lots, outgoing certificates, calibration records. Over the last four years, I've learned that most quality escapes aren't caused by a bad part. They're caused by a broken assumption about how the part is being measured.

The lot that looked perfect

We make lab consumables and specialty molded components for medical and research customers. A lot of our work involves serological pipettes. For an incoming lot of 8,000 pipettes, 5 mL volume, the drawing called out a tip bore tolerance of ±0.05 mm. That's not super tight by machined-metal standards, but it matters for how the pipette seats on a pipette controller and how accurately it dispenses.

Everything I'd read about injection molding QC said the same thing: if the part looks correct under magnification and passes a functional fit test, accept it. In practice, I've learned that's only true when the critical dimension is on the surface. The tip bore is inside the part. A stereo microscope, even a very good Zeiss microscope, can't tell you if that bore is perfectly round and straight. It can only show you the surface.

We sampled 32 pipettes. Under the Zeiss microscope, all 32 looked acceptable. On the Zeiss CMM—a Contura we use for dimensional first articles—11 of them had tip bores 0.03 to 0.06 mm under nominal. The average was within tolerance. The individual parts were not. The supplier challenged us on that point, and they were right to.

Why the microscope couldn't catch it

I had to explain this to myself before I could explain it to anyone else. The microscope is for surface attributes: flash, cracks, parting lines, contamination. The CMM is for geometry: diameter, roundness, taper, position. Both are necessary. Neither can replace the other.

From the outside, this looked like a supplier quality problem. The reality was that our acceptance criteria were incomplete. We gave the vendor a tolerance, but our incoming inspection never verified that tolerance with the right tool. The vendor wasn't just outside industry standard—the vendor was working to a drawing we never actually enforced.

The thermal twist

We held the lot and ran a longer study. This is where the story turns in a direction I didn't expect. The CMM readings weren't stable. The same part measured 0.04 mm below nominal one hour and 0.01 mm above nominal the next. I almost blamed the Zeiss machine. But it had passed its calibration six weeks earlier, and its length measurement error was below the acceptance limit per ISO 10360-2 (source: ISO 10360-2:2009). The machine wasn't lying. The room was.

We pulled out the TG298 thermal imaging camera we use for electrical inspections. I don't know if you've seen one—it's a handheld thermal camera with a built-in IR thermometer, usually found in electrical maintenance toolbags. That afternoon, I pointed it at the wall behind the CMM and saw a temperature band shifting from the floor up to about waist height. The HVAC vent was blowing directly onto the lab's exterior wall. The sun hit that wall through a window I'd forgotten to include in our environmental risk assessment.

The temperature near the CMM varied by about 2.1 °C over three hours. Actually, I'm not 100% sure of that number; I'd have to check the thermal image log. What I remember is the visual: a clear hot band that explained why the readings drifted. Even a compensated CMM struggles with a large horizontal temperature gradient. The granite table bends a little. The probe heats up. The measurements shift by a few microns—enough to turn a borderline part into an out-of-tolerance part.

The lab was temperature-controlled in the building-management sense. It wasn't stable in the CMM sense. Without the TG298 thermal imaging camera, I suspect the CMM would have taken the blame and the pipette lot would have been released pending investigation. That would have been a mistake. The defect was real; the borderline readings were a symptom of an unstable lab.

The Fluke vs Klein multimeter tangent

While the pipettes sat on hold, I was dealing with a separate request: standardizing our handheld electrical test equipment. My maintenance supervisor asked me to settle the Fluke vs Klein multimeter debate. I ordered one of each and connected them to the same calibration source for a week.

The honest conclusion: both are good meters. The Klein was easier to use and more affordable; for general troubleshooting, it's a great tool. The Fluke had a tighter DC accuracy specification and a calibration certificate with NIST-traceable validity, which matters when you're making pass/fail decisions on equipment safety. We ended up buying Fluke units for calibration-critical checks and Klein meters for field technicians.

That sounds like a tangent, but it's the same principle. You don't pick the best tool in the abstract. You pick a tool whose uncertainty is small enough for the tolerance you need to hold. A microscope sees surface defects. A CMM sees geometry. A thermal camera sees environmental drift. A good multimeter with verified calibration sees electrical truth.

What we did about it

We rejected the 8,000 units. Maybe it was 7,800—I'd have to check the final disposition record. The vendor reworked their mold, ran a new first article, and we measured 50 parts across the cavity layout on the Zeiss CMM. The reworked lot passed with all readings within ±0.03 mm. The redo cost them somewhere in the range of $18,000, though I'm guessing; they never sent us the bill.

That rejection also produced a permanent process change:

1. Every new supplier lot gets a CMM first article before the visual inspection is allowed to pass the batch.
2. We run a thermal audit of the inspection lab every quarter, using the TG298 to map gradients around all long-term measurement equipment.
3. All handheld meters, including the Fluke units, go on a 12-month calibration cycle with documented traceability.

We also automated the CMM report. The software now exports measured values straight into a PDF and emails it to the quality folder. That cut our reporting time from 45 minutes per lot to about 15. More importantly, it eliminated the one step where a technician had to re-type numbers into Excel. That tech rarely made a mistake. But rarely is not never, and data entry errors are invisible until they aren't.

What I tell people now

Conventional wisdom says trust, but verify, supplier quality. My experience with this batch says something stronger: verify the verification. The part looked good. The paperwork looked good. The supplier was experienced. The only reason we caught the defect was that a Zeiss coordinate measuring machine was sitting in our lab—and that, for once, we listened to the measurement instead of the part that looked normal.

If I remember correctly, the whole investigation took about three weeks. We completed it in 12 working days, but don't quote me on that. The important number is what happened next: our supplier-related escape rate dropped in the next two quarters—about 60%, maybe 55%—not because suppliers magically improved, but because we finally had a process that could catch a subtle defect before it became an eight-thousand-unit problem.

I still use Zeiss microscopes for surface inspection. I still think the CMM is overkill for simple parts. But for tolerance-critical work, overkill is exactly the right amount. And before I run any measurement program now, I check the room temperature. The TG298 makes that easy—point, pull the trigger, look at the gradient. Note to self: we still haven't documented that thermal audit procedure in the work instruction. I should do that before next quarter.

If you're molding serological pipettes or doing precision inspection on anything with a ±0.05 mm tolerance, take this with a grain of salt: the tool that gets blamed is rarely the one that actually failed. Start with the environment. Then check the data. Then question your assumptions. The part can look perfect and still be wrong.