How continuous height measurement can catch a production anomaly — before it reaches the customer

A manufacturing line was running within spec on average. But something was changing inside each unit. Only millisecond-resolution camera data from the Owl platform revealed what weekly QC audits had completely missed.

The customer and the challenge

Our customer operates a high-speed production line where each unit must meet a precise dimensional profile. The line runs around the clock, producing thousands of units per day. Traditional quality control relied on random sampling — measuring a handful of units per shift and checking whether the average height fell within tolerance.

This approach had served them well for years. Averages looked fine. Defect rates were low. But field returns had started to tick upward, and the complaints all described the same symptom: units that measured correctly overall but behaved differently in the middle of their profile.

Nobody could explain why. The sampling data showed nothing unusual. The line operators had no anomaly to point to. So the customer came to us.

Deploying Owl: cameras on every unit, not every hundredth unit

We installed the Owl measurement platform inline on the production line. The system captures height at twelve measurement positions along each unit as it passes — generating a full geometric profile for every single unit produced, not a statistical sample.

Over the first three weeks of operation, Owl collected profile data on over 24,000 units — roughly 290,000 individual height measurements.

Here is what traditional sampling would have told the quality manager: the line was fine. The average height across all units was 119.5 mm, sitting comfortably close to the 120 mm reference. Median almost identical. Standard deviation: 0.18 mm per unit. Nothing to act on.

The Owl platform told a completely different story.

Finding 1: measurement positions 10–11 were reading low — but the product was the cause

The first thing analysts noticed when reviewing the raw position-level data was that some positions consistently read lower than the rest of the profile. Initial instinct: sensor calibration issue, or a camera alignment problem in the measurement zone.

The investigation took a day. The answer turned out to have nothing to do with sensors: positions 10 and 11 sat at the trailing edge of the unit, and the product's own geometry meant those positions naturally measured lower. The shape was correct. The low reading was the product.

Moving from sampling to full-coverage measurement means seeing the real shape of the product for the first time. Some of what you see looks like a problem — until you understand it is simply the truth. With the true baseline established, any future deviation from it becomes immediately visible.

Finding 2: from day 13, a new dip appeared — in the middle

With a reliable baseline in place, the Owl platform's continuous monitoring started showing something the average never would. On day 13 of the measurement period, the average height across all units stayed flat at 119.5 mm — but the within-unit spread nearly doubled, from 0.64 mm to 1.23 mm (+92%), and kept climbing.

The added variation was not spread evenly across positions. It was concentrated at positions 5 and 6 — the geometric centre of each unit. A centre dip was developing: not present in the first twelve days, deepening steadily from day 13 onward. The jump coincided with a process change identified during root-cause investigation.

The before-and-after profile comparison makes the anomaly unmistakable: edges holding their position, centre sinking.

This profile change would have been invisible to traditional sampling.

What this kind of data makes possible

Precise timestamps, position-level profiles, and stable baselines give engineering teams something sampling cannot: the ability to pinpoint when a change happened, where on the unit it manifests, and how it has evolved over time. The difference between a shape that is fine and a shape that is drifting becomes visible in the data — not weeks later in field returns.

An average can be within spec while a shape is failing. Owl measures both — on every unit, continuously. If you are running a line where dimensional consistency matters, we would like to show you what your line actually looks like.

All charts use illustrative example data, not customer production data.