The connected welding robot

The problem

A modern cell, an open API, and a person typing anyway.

The welding cell is the last physical step before final inspection. A wrong program selected by hand can mean hundreds of out-of-spec parts.

• The robotic welding cell is the last physical step before final inspection: after it, the structural assembly is welded and whatever went wrong is already inside the part.
• It has a documented interface and it is perfectly capable of receiving its program from outside. It lives isolated from the MES all the same, because the integration project never reached the top of the list.
• So on every part-number change somebody selects the program by hand on the panel, with data that already exists in the ERP.
• One wrong selection means hundreds of structural parts welded to the wrong program: scrap, or sorting piece by piece, and a nonconformity handed to the customer if the pallet already shipped. And the exposure repeats on every changeover, several times a day, with the same person under the same pressure to get the line running again.

How it fits the IRIS system

Connect stitches the two systems through APIs, replacing neither.

Connect stitches two modern systems via API: when the ERP marks a work order as started, it automatically publishes the welding program and target parameters to the robotic cell.

What was actually executed comes back confirmed to the system: not the program that was requested, but the one the cell reports having run, with the parameters it ran it at. That distinction is what turns welding into evidence instead of an assumption, and evidence is what an OEM asks for when a structural assembly is questioned months later.

See the full IRIS architecture →

Before and after

The isolated cell versus the stitched cell

Aspect	Today	With iLEAN Connect
Selecting the welding program	By hand on the panel	Published from the active work order
The same data typed twice	Once in the ERP, once at the cell	Once, in the ERP
Wrong program on a batch	A monthly occurrence	Eliminated at the source
Welding parameters	Assumed to be the right ones	Targeted and returned confirmed
Changeover time at the cell	Minutes on the critical path	A work order status change
The robot and the ERP	—	Neither is replaced

From monthly manual program-selection errors to zero human error at the source.

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, counting only the changeover time recovered at the cell.
• The real value is the error that disappears: a drastic reduction of structural nonconformity risk from a program selected by hand.
• From manual program-selection errors every month to zero human error at the source.
• And what was actually welded comes back confirmed from the cell, so it can be used as evidence in front of the customer instead of assumed from what the work order asked for.

Estimated payback 3-6 months; drastic reduction of structural-nonconformity risk. Estimate to validate.

And the fair question from the production manager

“Wouldn't a proper integration project be cleaner?” — it would, on the day it reaches the top of the list, and that is exactly the problem: this is the classic modern machine still isolated because there was never a window. The stitching replaces neither system, needs no shutdown and no reprogramming of the robot, so it does not compete in that queue. Matching the work order to its welding program is an anchored task, where the best models drop below 1.5% error [1], and the cell returns what it actually ran, so the match is verified rather than assumed.

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