The platform

The autonomy layer for the converting plant

One plant-edge stack turns perception into bounded machine control across the corrugator, presses, die-cutters, folder-gluers and end-of-line, with a digital twin that tunes makeready before the run.

The problem, precisely

A genuine black art, run at enormous speed on razor-thin margins

Turning giant reels of paper and board into strong, square, beautifully printed packages depends on dozens of coupled variables that drift with humidity, board lot, ink and job.

  • Corrugator: warp, washboarding, delamination, bond & crush strength
  • Press: registration within a fraction of a millimeter, exact brand color
  • Die-cut & fold: clean cuts, sound creases, no skips or fishtails
  • End-of-line: high-mix feeding, stacking, bundling and palletizing
Plant · live stateAUTONOMOUS
Corrugator warp0.6 mm/m
Press ΔE0.9
Die/anvil wear12% life
Glue integrity99.98%
Plant OEEon target

Perceive → decide → act

Perceive → decide → act, on every machine

The same closed loop runs at each stage, orchestrated by the twin.

  1. 01

    Sense

    Edge vision, inline spectrophotometry and PLC telemetry fuse into a live state for board, print, cut, fold and stack.

  2. 02

    Predict

    Models forecast bond and crush strength, color ΔE, cut and crease quality, joint integrity and stack stability in real time.

  3. 03

    Decide

    Process agents choose bounded setpoint moves that hold target quality and yield within operator-approved limits.

  4. 04

    Act

    Closed-loop control adjusts steam, heat, glue, moisture, tension, registration, ink keys, fold speed and robot motion.

  5. 05

    Simulate

    The converting-line-and-box twin auto-tunes makeready before the run and validates every change first.

  6. 06

    Learn

    Outcomes flow back into the plant-edge flywheel so every agent, and every plant, gets better.

The stance

Four brand pillars, built into the architecture

The platform is designed to earn trust on the plant floor, not just in a demo.

Trust

Grounded, auditable, quality-grade outputs. Every control decision is logged and traceable.

Autonomy

It performs the work, not just assists, bounded closed-loop control, not another dashboard.

Outcomes

Priced and measured on value: waste, makeready, claims, energy and OEE.

Vertical depth

Built for packaging converting specifically, not a generic industrial AI shell.

The data flywheel

A cross-machine dataset no single-vendor tool can assemble

Board strength informs fold and stacking; print and fold defects inform makeready; the twin calibrates against every run. Each product’s outcomes become training signal for the others.

  • Image + spectro + PLC + outcome telemetry, unified at the edge
  • Shared cross-vendor machine and MES connectors
  • Common Jetson / Triton / NIM serving fabric
  • Omniverse converting-line-and-box twin as system of record
0.6mm/m

Warp control

0.08mm

Print register

+6% nest

Board yield

−9%

Energy

Deployment

What runs where

LayerWhere it runsWhat it does
PerceptionPlant edge (Jetson)Real-time vision & spectro inference at line speed
Process agentsPlant edge / on-premBounded setpoint decisions per machine
Control bridgeMachine PLC / OPC-UAClosed-loop moves within approved limits
Digital twinOVX / RTX (Omniverse)Pre-run simulation and makeready auto-tune
Flywheel & modelsDGX / cloudTraining, evaluation and fleet model updates

Foldeon closes the loop from perception to corrugator, press and gluer control, lifting quality and OEE while cutting waste, makeready, claims, energy and manual checks.

Answers

Questions converters ask

Can we start with a single line?

Yes. The Line tier deploys one agent on one press or line. As ROI proves out you expand to more lines, more modules and the full plant loop.

How does the twin reduce makeready?

The converting-line-and-box twin simulates corrugation, print and die/fold to hit target quality and yield before the run, so the line reaches spec with less setup waste and fewer trial sheets.

What about operator trust and control?

Every plant chooses its autonomy level (shadow mode recommends only, assist mode requires approval, autonomous control runs bounded) per machine and per workflow.

Put the whole line on one autonomy layer

Start on one wedge, prove the ROI, and expand across the plant at your pace.