The Set-and-Calibrate Step Is Hiding Your Real Defect Rate

Written By Mitch Robinson

Every electromechanical relay that comes off your line gets a set-and-calibrate step. Ten to twenty minutes per unit. Spring tension gets adjusted for proper return. Contact pressure and gap get set to design parameters before the unit moves on to functional testing. That adjustment work is a normal, documented part of relay assembly. Everyone treats it as fixed overhead, the price of building something with moving parts. It isn't. In most shops running this step, the technician is often correcting the same handful of defects, unit after unit, and nobody upstream knows it's happening. GEYA

What set-and-calibrate actually is

On paper, set-and-calibrate means fine-tuning a good unit to meet its final spec. In practice, on a lot of lines, the technician isn't fine-tuning. They're correcting. A stamped part upstream is running loose, so the contact gap is wide. A supplier lot shifted, so spring tension reads low. A winding step isn't holding tolerance, so coil resistance is variable. None of that gets traced back. The tech just sees a bad reading and knows how to make it good, because they've done it a thousand times.

The calibration bench is absorbing defects created earlier in the build, and nothing records that it happened.

This has a name, and it's older than your process

Quality engineer Armand Feigenbaum named this in the 1970s. He called it the hidden factory: the share of a plant's capacity that goes into unofficial rework, workarounds, and correction that never shows up on a scrap report or a defect log. He pictured it as an iceberg, with the costs everyone tracks (rejects, warranty, scrap, formal rework) sitting above the waterline, and the real cost sitting below it, uncounted. His number, which sounded absurd when he said it and has held up since, was that the hidden factory eats somewhere between 15 and 40 percent of a plant's total capacity. iSixSigmaAIGPE

Set-and-calibrate on a relay line is a textbook hidden factory. It's not on the scrap report because nothing got scrapped. It's not on the rework log because nothing was ever flagged as defective. It's absorbed, quietly, by a skilled tech who has learned to compensate without ever being told what they're compensating for.

The sharper problem: compensation isn't correction

Here's the part most quality systems miss entirely. W. Edwards Deming spent years demonstrating something called tampering, using a simple setup: drop a marble through a funnel at a target, and adjust the funnel's position based on where the last marble landed. He built the experiment specifically to show what happens when you adjust a process in response to variation without first understanding what's causing it. The finding, confirmed over and over: adjusting based on individual data points doesn't tighten the results, it multiplies the variation, sometimes by two or three times over. W. Edwards Deming Institute

A calibration tech nudging spring tension because the reading came in low is doing exactly this. They're not correcting a known cause. They're reacting to a symptom and moving the setting until the number lands in range, one unit at a time, with no record of which direction they moved or why. It works, in the narrow sense that the unit passes. But the upstream cause that put the tension out of range in the first place is still there, still drifting, and will show up again on the next unit and the one after that.

What it actually costs

An automotive parts supplier once calculated their scrap and rework at 5 percent of revenue, using the numbers everyone tracks: material and direct labor. When they added the capacity tied up by rework, the engineering hours spent chasing the same problems, and the expediting needed to replace what got scrapped, the real number came out above 12 percent of revenue. They'd been fighting the same quality problem for years while measuring less than half its actual cost. Set-and-calibrate time on a relay line works the same way. If a clean unit needs three to five minutes and your line average is twelve, those extra seven to nine minutes are rework, they just never got the name. Rework

Three costs hiding inside the same ten minutes

Time. Every minute past the clean-unit baseline is labor spent correcting a defect that was never diagnosed.

Scrap risk. Not every correction lands. A unit that's out of range far enough eventually gets scrapped instead of saved, and that one does show up on your report, disconnected from the ninety-nine before it that got quietly fixed.

Dependency on one person. The fastest techs aren't fast because calibration is easy. They've built an unwritten pattern library: this reading, this fix, every time. New hires take longer and miss more, and it gets written off as a training gap. It isn't. It's the same Best Person Bottleneck that shows up anywhere critical knowledge lives in someone's hands instead of on a spec sheet. Your line's real ceiling is however sharp that one person's memory is on a given day. I’m beginning to wonder if we should just ABOLISH all specific production training so that every employee can do every job, regardless of how good or bad their memory is. Probably not going to happen. But what if we made it easier for everyone to do the work. Like LEGO instructions. Like the Betty Crocker Cookbook.

How to open the black box

Log what's actually getting adjusted, not just how long it takes. For one to two weeks, have techs note which parameter needed the most correction on every unit that ran long. Not root cause yet. Just which reading was off.

Sort by frequency. In most quality data, a small share of defect types accounts for the large majority of the cost, typically in the range of 20 percent of types driving 80 percent of the total. Relay calibration will follow the same pattern. Two or three recurring issues will account for most of your extra minutes. Rework

Trace each one to its actual source. A wide contact gap traces to a stamping or forming step. A resistance drift traces to a winding process or a supplier lot. Go stand at that step. Don't read the traveler, watch the part get made.

Fix the source, not the symptom. Once the upstream cause is controlled, that defect stops showing up at the bench, and your standard calibration time drops for every unit that follows it, permanently, not just for the batch you happened to catch.

The takeaway

A hidden factory doesn't announce itself. It looks like a normal step on a normal traveler, running at a time everyone's accepted as standard. The floor already knows which units take longer and which reading they're always fighting. That knowledge just never left the calibration bench. Go stand there for a week and find out what your best tech has been quietly fixing for you, for free, this whole time.

Sources

Mitch Robinson
Previous

DIY Changeover Streamline - SMED
Next

What a Bottleneck Breakdown Assessment Actually Finds on Your Assembly Line