How Evident NDT Equipment Caught a 0.6 mm Crack Our Calipers Missed

7:42 a.m., receiving bay, 9°C

The coffee had already gone cold. I was standing in the receiving bay with a clipboard, a 300 mm caliper, and a printout of the dimensional report our supplier had emailed the night before. Everything on paper said pass. That's usually where the trouble starts.

The lot was 1,200 anodized aluminum housings, machined for a customer who builds proximity-sensor modules for packaging lines. The housings hold 18 mm barrel sensors—their spec, not ours. We were the contract manufacturer in the middle, which meant if the housings were off, the customer's assembly line would find out before we did. That's the part nobody puts in the quality manual.

I've been doing incoming inspection for a little over four years now. In that time I've reviewed maybe 300 lots, and I've rejected somewhere around 6% of first deliveries. Not a heroic number. Just the one I keep in my head when a supplier tells me their process is "stable."

The check that wasn't on the original checklist

Our standard incoming inspection for machined housings is dimensional plus a visual. Bore diameter, mounting hole spacing, flatness, thread depth, anodize thickness on a sample basis. That's it. It takes about 40 minutes for a lot this size, and up until 2023 it had never missed anything that mattered.

What changed was the customer's end use. These modules were going onto a line that runs 22 hours a day, and a sensor that drifts out of position creates false triggers—which creates scrap. So in Q3 2023 we added an internal-channel inspection step to the control plan. Not because of a failure. Because of a near-miss at the customer's plant on a different part number that looked identical.

The tool we used for that step was a videoscope we'd bought in early 2022. It arrived in a case with the Evident logo on the lid, and I'll be honest—I thought the distributor had shipped the wrong unit. We'd been running an old Olympus-era scope for years, and the rebrand caught me off guard. I called the rep, who explained the science division had been spun out under the Evident name. Same optical bench, same probes, different label. Fine. We kept it.

That decision came back around 18 months later, on the batch in front of me.

Where the Evident NDT equipment earned its keep

We pulled 50 pieces for the internal check—tightened sampling, not full inspection. The scope probe was 4 mm, which fits the sensor bore with about 1 mm of clearance. On piece 47, the screen showed a fold on the inner wall of the mounting bore, roughly 0.6 mm long, running diagonally. Not a burr you could feel. Not something a caliper or a visual would ever flag on an anodized surface.

I called the line lead over. She looked at the screen, then at the part, then back at the screen. "That's inside the wall," she said. Which was exactly the problem.

The drawing note was specific: no visual cracks, porosity to ASTM E505 level 2 maximum in the section shown. The supplier's first response—I still have the email—was that the defect was "within normal machining variation for this alloy." I sent them the screen capture with the scale bar. The conversation changed after that.

This is the part where I'm supposed to say the Evident NDT equipment saved the day. It didn't, exactly. It just made a problem visible that we'd been structurally unable to see. There's a difference. The tool sat on a shelf for 18 months before we built a step around it. That's on me, not on the equipment.

The 336 clamp meter, the 23 multimeter, and a dying battery

While the internal inspection was running, we were also validating the test fixture for the customer's next build. The fixture checks sensor current draw and ground continuity before the module ships. For current, we use a 336 clamp meter—600 A range, true RMS, been in the tool crib since before I started. For voltage drop across the connector pins, a bench 23 multimeter that's older than my employment record. Both of them work. Neither of them is fancy.

The afternoon of the batch inspection, the 23 started reading 0.8 V where we expected 0.4 V. I spent twenty minutes convinced the fixture wiring was wrong before the second shift tech pointed out the battery icon on the display. Fresh 9 V, readings back to normal. Twenty minutes of my life I'm not getting back, and a reminder that the instrument is part of the measurement. We now log battery status on the fixture check sheet.

That same week, our customer asked why we weren't specifying Keyence sensors for the module instead of Omron. Fair question, and I didn't have a real answer—so rather than quote a datasheet, I ran a side-by-side test. Same housing, same target, same ambient conditions. I wanted to see how sensors compare with Omron and Keyence on the two things that actually matter for us: installed sensing distance margin and repeatability after a 30-minute warm-up.

The short version: Keyence gave more margin, Omron was tighter but consistent, and the housing bore tolerance interacted with both more than either datasheet admitted. We kept Omron for that build and wrote the bore tolerance into the drawing so the next machinist didn't guess. To be fair, this was one fixture and one target material—take it with a grain of salt.

The call I didn't want to make

Back to the housings. I rejected the lot. All 1,200 pieces, not just the 50 we'd sampled, because the defect was in a machined feature produced by the same tool on every part.

Even after I hit send on the rejection notice, I kept second-guessing. What if I was being too strict on a 0.6 mm internal fold? The supplier's sales manager called within the hour, polite but direct: rework would push delivery five working days, maybe seven. Our customer had a build scheduled. If we slipped, we'd be the reason their line went idle. The two days between that call and the supplier's written response were stressful, and I slept badly.

What made it bearable was that the decision wasn't really mine—it was the control plan's. The drawing note existed. The inspection step existed. All I did was follow it. That's the entire point of putting a check in writing before you need it.

What actually happened

The supplier re-machined the lot at their cost. Delivery landed five days late—two days better than their worst estimate. The customer's planner was annoyed for about ten minutes, then asked what we'd found. When I sent the scope image, he forwarded it to his own quality team and asked us to add the same internal check to the next two part numbers.

Rough numbers, because I keep them: the batch was worth about $28,400. Our internal estimate for a field failure—line stoppage, module rework, expedited replacement—came in around $40,000 to $55,000, based on a similar event at the customer's plant in 2022. Not a precise figure. But the direction is clear.

The supplier now includes an internal-channel inspection record with every lot. Their scrap rate on this part went up for two months, then came back down once they changed the boring bar and added a deburr pass. They were not a bad supplier. They were a supplier who had never been asked to look inside.

What I'd do differently—and what I now do every time

It took me about four years and 300-plus lots to understand that "meets spec on paper" and "works in assembly" are two different claims. The dimensional report was accurate. Every number on it was true. The part was still wrong.

Three things changed after that batch:

• A 12-point pre-delivery checklist that includes one non-dimensional check per critical feature. The internal bore inspection is now point 7. It adds about six minutes per lot.
• Instrument status logging—battery, calibration date, last verification—on the fixture sheet, after the 23 multimeter incident.
• A side-by-side test requirement for any sensor or meter we specify for a customer build. No more quoting datasheets in meetings.

When we expanded the inspection cell in Q2 2024, we bought a second videoscope. Same brand, partly because the existing probes and training carried over, partly because the distributor's tech answered a Saturday email in under two hours when the first unit's articulating tip got stiff. That kind of support is hard to price, but it's real.

I'm not sure the checklist has saved us $8,000 or $80,000. I haven't run the numbers that carefully, and I don't want to pretend I have. What I do know is that six minutes of looking inside a bore is cheaper than five days of explaining why we didn't.

Five minutes of verification beats five days of correction. The trick is doing it before the paperwork says "pass."