Every safety tightening, tied to its part

The problem

The curve exists. It is just not attached to the seat.

Those controllers are relatively modern and almost always already expose a programming interface or an open protocol. But in many plants they remain islands. It is the classic almost-integrated-but-not case: the data exists, the tool generates it, the interface is there, and the project to stitch it properly never rises up the priority list. The result: the evidence that the safety joint was tightened correctly does exist, but disconnected from the part. When the customer asks about a specific serial number, the answer is built by hand cross-matching files by timestamp, with the margin of error that implies. There is a second, more serious problem. When the variant changes and the torque program changes with it, today that change is made by hand on the controller. A wrong program means a safety joint out of specification.

• The belt anchorage, the frame joint, the side airbag module fixing in the backrest: safety joints tightened with torque- and angle-controlled nutrunners, and every tightening produces a curve.
• The controllers are modern and expose an interface or an open protocol, and they remain islands. The project to stitch them never rises up the priority list, because nothing is visibly broken.
• So when the customer asks about a specific serial number, the evidence is built by hand, cross-matching files by timestamp, with the margin of error that implies. The data exists, the tool generates it, and it is still disconnected from the seat.
• Worse: when the variant changes and the torque program changes with it, that change is made by hand on the controller. A wrong program is a safety joint out of specification.

How it fits the IRIS system

Connect stitches controller and manufacturing system through their APIs, in both directions.

Connect stitches the two systems through their programming interfaces, in both directions. Inbound: every tightening event, torque, angle, result, retries, is captured and tied to the serial number active at that station. Outbound: when the manufacturing system announces the next variant in sequence, the correct torque program is published to the controller with no keying in between. And let me be clear about what is not needed: neither the tools nor the manufacturing system get replaced. We stitch what you already have.

Inbound, every tightening — torque, angle, result, retries — lands tied to the serial number active at that station. Outbound, when the sequence announces the next variant, the correct torque program is published to the controller with no keying in between. Neither the tools nor the manufacturing system get replaced: we stitch what you already have.

See the full IRIS architecture →

Before and after

The isolated controller versus the stitched controller

Aspect	Today	With iLEAN Connect
Torque curve	In the controller, disconnected from the part	Tied to the seat serial number as it happens
Evidence for a serial number	Cross-matched by timestamp, by hand	In seconds
Torque program at variant change	Keyed by hand on the controller	Published from the sequence
Wrong program on a safety joint	A permanent exposure	Eliminated by construction
Retries and NOK tightenings	A count, at best	Each one, with its curve
Tools and manufacturing system	—	Neither replaced

Impact estimate

Impact estimate — to be validated with your numbers.

The block below is an estimate to be validated against your plant's actual data. We put it forward so the committee has an order of magnitude; we refine it during the assessment.

• Estimated payback 3-6 months, mostly from engineering and quality time spent today reconstructing evidence by timestamp.
• The wrong-torque-program error between variants disappears, and that is a failure mode with safety consequences, which is the big value here even if it never shows in the hours saved.
• Evidence for any belt anchorage or airbag module fixing in seconds, by serial number, instead of a hand cross-match by timestamp.
• And the torque program travels with the sequence, so a variant change stops being a moment of exposure.

Estimated payback 3-6 months · removes wrong-torque-program errors between variants Fast payback, three to six months, mostly from the engineering and quality time spent today reconstructing evidence, an estimate to validate. But the big value is risk: the wrong-torque-program error between variants disappears, and that is a failure mode with safety consequences.

And the fair question from the production manager

“Isn't a classic MES integration cleaner?” — it is, the day it reaches the top of the list, and that is the problem: the interface has been there for years and the project never starts. This stitching replaces neither system, asks for no shutdown window and is bounded to one station in the trial. And there is no camera reading in the loop here: the curve arrives through the controller's own interface, so the reliability question does not even arise — unlike the photo cases, where the best models already drop below 1.5% error on anchored tasks [1].

^[1] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.