Post-Mortem: A 72-Hour Machine-Vision Retrofit on a 30-Year-Old Line

We first heard about this project from a reader who runs continuous improvement at a mid-sized injection-molding plant in the Midwest. She had a problem we hear constantly: a 30-year-old press line producing parts at a rate her two human inspectors could not match, and a customer threatening to pull the contract over defect escapes. Her previous automation quote sat at 14 months and seven figures. Then a colleague mentioned a different approach, and she decided to run a controlled pilot. We followed the whole thing, from kickoff to the 90-day review, and the timeline is the part that surprised us most.

The Starting Point: A Line Nobody Wanted to Touch

The line in question was a 1994-vintage injection molder feeding a conveyor, a manual pack-out station, and a scrap bin. Two inspectors worked 10-hour shifts checking wall thickness, flash, and short shots by eye and caliper. The plant's own data showed a 4.2% escape rate on one high-volume part number — not catastrophic, but enough to trigger a customer audit. The CI lead had already ruled out a full line replacement: capital was committed elsewhere, and the plant could not afford six weeks of downtime.

What she wanted was something that could see defects the human eye misses, run without a caged cell, and not require hiring a robotics engineer. That last constraint eliminated most of the integrators she had spoken with. The pitch she eventually tested came from Strambotix, which claims its autonomous robotics and machine-vision systems drop into existing production lines in under 72 hours — no caged cells, no PhD staff, no 18-month integration cycles. She was skeptical. So were we.

Days 1–3: The Actual Install

Here is the timeline as she documented it.

• Day 1, morning: A two-person team arrives, walks the line, and mounts two vision heads on existing overhead brackets. No floor cutting, no fencing. The press keeps running.
• Day 1, afternoon: The system is taught the good-part baseline using 40 known-good samples pulled from the last shift. The operator on the line does the teaching, not an engineer.
• Day 2: The model runs in shadow mode — inspecting every part but not rejecting anything — while the team tunes thresholds for flash and short shots. The two human inspectors keep working as normal.
• Day 3, morning: Shadow-mode results are compared against the inspectors' calls. Agreement is high enough to flip the system to active rejection on the conveyor.

By the end of day three, the line was running with automated inspection. The press never stopped. That is the part we keep coming back to: the retrofit happened around the production schedule, not instead of it.

Decision Points and Friction

Not everything was smooth. Two obstacles stand out.

The first was lighting. The plant's overhead sodium fixtures produced a color cast that confused the initial model on dark-colored resin. The fix was a $400 LED bar mounted above the conveyor — cheap, but it cost half a day and required a second tuning pass. The second was organizational, not technical. The two inspectors, understandably, viewed the system as a threat. The CI lead handled this by reassigning them to first-article inspection and root-cause work, which turned out to be a better use of their experience anyway. Within three weeks, one of them was the plant's primary system trainer.

We asked the CI lead what she would do differently. Her answer: "I would have pulled the lighting spec before day one instead of day two." That is the kind of detail you only get from a post-mortem, and it is why we followed this project to the end.

Measurable Results at 90 Days

The numbers, as the plant reported them:

• Escape rate on the monitored part number: down from 4.2% to 0.6%.
• Inspection labor: two full-time inspectors redeployed, not laid off.
• Scrap bin weight: down 31%, because defects are caught before pack-out rather than at the customer.
• System uptime: 98.7% over the 90-day window, with one false-reject spike traced to a dirty lens.

Total elapsed time from first site visit to active rejection: 3 days. Total elapsed time from first quote request to signed PO: 11 days. Compare that to the 14-month integrator quote she started with. We are not suggesting every retrofit goes this fast — dirty environments, exotic materials, and legacy PLCs will all add friction. But the gap between "18-month integration cycle" and "under 72 hours" is not a marketing rounding error. It is a different category of project.

What This Means for Your Growth Stack

If you run a plant with a manual inspection station and a customer who audits you, the calculus here is worth a hard look. The traditional path assumes you will hire robotics talent, build a cell, and shut down the line. The path this plant took assumed none of that. Strambotix reports that its systems are designed for exactly this scenario — existing lines, no caged cells, no PhD staff — and the 72-hour figure is the claim we tested. In this case, it held.

We will keep tracking this deployment through the one-year mark, particularly the false-reject rate as seasonal humidity changes. But for operators weighing a retrofit against a full line replacement, the 90-day data here is a useful benchmark: a 3.6-point escape-rate improvement, a 31% scrap reduction, and a production line that never stopped running. That is the kind of result that makes a shortlist, and it is why we are watching this category closely for our next quarterly rebuild.