The Rejection
In the second week of March 2024, a customer rejected 14 out of 200 precision mounting brackets we'd delivered for a commercial aircraft program. The rejection notice said two critical datum points were out of tolerance by about 120 micrometers—roughly the diameter of a human hair. Every one of those parts had passed our in-house inspection on our hexagon CMM.
I'm the quality compliance manager at a mid-sized aerospace machining shop. I review every deliverable before it leaves our floor—roughly 200+ unique items a year. In 2024, I'd already rejected 14% of first deliveries myself, for things like surface finish deviations and documentation gaps. So I was used to being the one who sends parts back. Being on the receiving end felt different.
The bracket was a 2.5-meter-long aluminum component with datums at opposite ends that had to hold a 0.1 mm positional tolerance. Our CMM—a Hexagon model we'd maintained religiously, with annual reverification per ISO 10360-2—passed every one of those parts. I told the customer's quality engineer exactly that, and offered to share our inspection data.
He thanked me and sent back theirs. Both calibration chains traced back to NIST reference standards. Two calibrated systems pointing at the same parts, and they disagreed. That's when I stopped feeling defensive and started feeling curious.
The Blind Spot
The uncomfortable truth, which nobody on our team had wanted to say out loud, was that our CMM couldn't measure the full bracket in a single setup. The machine's measuring range forced us to work in sections. We'd measure one end, re-fixture, measure the other end, and then mathematically stitch the data together. Every section re-established datums using the part's own features.
That approach had worked fine on shorter parts. But on a bracket this long, with critical features at opposite ends, sectioned measurement can quietly wash out exactly the kind of cumulative error the customer was seeing. The CMM wasn't lying to us—it was blind in one eye, and we hadn't realized it.
We needed a system that could see the entire part volume in one pass. A hexagon laser tracker was the right tool for that. We rented a Leica Absolute Tracker AT960, which has a published accuracy of ±15 µm + 6 µm/m. At our 2.5-meter working distance, that's comfortably inside the precision needed to settle a 0.1 mm dispute.
There was one catch. Nobody on our staff was certified to run it.
Two Days That Changed Our Approach
We sent two senior technicians to Hexagon's training facility. Both had already gone through hexagon CMM training back when we'd purchased the machine. The laser tracker operations course was new territory. It turned out to be the most valuable training investment we've made in a decade.
The first day covered the hardware: warming up the tracker, establishing the measurement volume, setting reflectors, and running the standard performance verification per ASME B89.4.19. The second day was different. The instructor spent the entire afternoon on measurement uncertainty budgets, and somewhere in hour three he said something I'll never forget:
The instrument is only one component of the measurement system. Everything around it—the operator, the environment, the fixture, the floor—is the rest of the system.
I'm not a metrology engineer, so I can't walk you through the math behind interferometric length measurement or the SPC error model. That's outside my professional lane. But what I took from that afternoon was a shift in thinking. It's tempting to believe a ±15 µm laser tracker is simply "more accurate" than a ±2 µm CMM, and that better numbers on a spec sheet solve your problems. That's an oversimplification. The real uncertainty of any measurement is a system property, not a spec sheet figure.
The Root Cause Was on Our Own Floor
When the tracker arrived, our technicians set up the full bracket volume in a single coordinate system. They measured all 14 rejected parts plus a sample of the 186 we'd already shipped. The rejected parts lined up with the customer's data—not ours. They were genuinely out of tolerance. The sampled parts we'd passed were fine. The customer's inspection had caught the bad ones, and our sectioned process had missed them.
That was a hard moment. Not because the 14 brackets represented only about $12,000 in direct value against a $180,000 contract. The hard part was realizing we'd shipped non-conforming parts with genuine confidence. Our data had told us we were fine, and we'd believed it because we didn't understand the limits of what the data was seeing.
Then our lead technician did something that wasn't in the work plan. He used the tracker to measure the machining fixture—the cast iron assembly that held the brackets during production. He'd never have thought to do that before the training. The course kept repeating the phrase "verify your references," and he took it literally.
The tracker revealed a 200-micrometer distortion across the fixture's primary datum plane. The fixture had warped over four years of use. Not visibly. Not enough to show up on any single-point check with a dial indicator. But over the full 2.5-meter length, that distortion tilted every part just enough to push the far-apart datum features past the 120-micrometer threshold.
I'd assumed the fixture was still accurate because it had been accurate the day we accepted it from the toolmaker. I never verified it after that. We didn't have a formal fixture reverification process—the CMM got certified every year, and the fixture just got "trusted" until something broke. That's the most expensive assumption I've made in this job, and I own it.
An Unrelated Test, the Same Lesson
In parallel with the bracket investigation, I was running a durability comparison of ifm photoelectric sensors vs. two cheaper alternatives for our assembly tooling. One of the ifm units was a time-of-flight (ToF) model we needed for presence detection across a variable distance range of 100 to 500 mm.
Procurement wanted the cheapest option—the ifm sensors cost about 18% more per unit. So I ran a durability test before anyone signed a purchase order: temperature cycling from -20°C to 60°C per IEC 60068-2, continuous vibration at 4g, and a 3-million-cycle operational test. The ifm ToF sensor held its distance accuracy across the full 400 mm range and completed the entire test without a failure. One competitor's emitter output decayed below detection threshold at 800,000 cycles. Another competitor's housing cracked at 400,000 cycles.
The losing vendor's rep said I was testing beyond "typical" operating conditions. He wasn't wrong. But our plant runs multiple shifts around the clock, and a line stoppage costs about $500 per hour in idle labor. The 18% price premium disappeared the first time we didn't have to replace a sensor. Same principle as the fixture—whether it's a $200 sensor or a $40,000 fixture, the lowest-cost decision is the verified one.
The $38,000 Resolution
The 14 rejected brackets had to be re-machined from raw material. That redo cost $22,000 in material, machine time, and expedited freight. The tracker rental was $6,800. The training came to $9,200 with travel. Total: just over $38,000, not counting the eleven-day delay to the customer's program. Prices as of March 2024—verify current rates if you're budgeting for similar work.
I could frame that as a disaster. For a while, I did—kind of. But then something unexpected happened. We sent the customer a complete report: the tracker data, the fixture distortion scan, our new fixture reverification protocol, and a formal acknowledgment that their inspection had caught something ours couldn't. Their quality director told us later it was the first time a supplier had shown up with root cause analysis instead of an argument. They awarded us two additional programs the following quarter.
The fixture verification protocol we implemented now covers every fixture in the shop. Each one gets measured against its original datum before every production run, and the results go into a log reviewed at our monthly quality meeting. It took a $38,000 mistake to build a $500 checklist, but that checklist has already caught two other fixtures drifting out of specification. The math works out.
What I'd Tell Every Quality Manager
People ask me whether the CMM was wrong. It wasn't. The CMM reported exactly what it was positioned to see. The laser tracker reported exactly what it was positioned to see. The dial indicator checks did the same. The difference wasn't in the instruments—it was in the completeness of the reference frame each one used.
Customers don't buy parts. They buy confidence. When we shipped those brackets, we shipped our belief that they were good. The customer's rejection was a challenge to that belief. And it turned out our belief was built on an unverified assumption. Quality, in the end, is the sum of your assumptions—and the dangerous ones are the ones you don't think to check.
Now, when I evaluate a new instrument, a new supplier, or a new fixture, my first question isn't "how accurate is it?" It's "what's holding it in place, and when was the last time we verified the reference?" That question has saved us more money than any single piece of equipment we own. I do not say that lightly. I say it with a 200-micrometer scar on my pride that reminds me every day.
