CASE 001 / INDUSTRIAL AI

MANUFACTURING · COLD ROLLING

ONE PROBLEM, DECODED.

Fewer passes.
Same standard.

A production-grade decision system that searches for the minimum feasible cold rolling schedule—without trading away quality, safety, or mechanical constraints.

BUILT FROM SIX YEARS OF PRODUCTION HISTORY

01

PRODUCTION HISTORY

Observe
02

REDUCTION MODEL

Predict
03

PROCESS LIMITS

Constrain
04

FEASIBLE SCHEDULE

Optimize

Every pass has a cost.
Every reduction has a limit.

Cold rolling reduces steel coils to a target gauge through a sequence of passes. The difficult question is not whether the target can be reached, but how to reach it in the fewest safe steps.

Pass schedules are shaped by interacting process parameters and are often built through rules or operator experience. That can leave efficiency hidden inside the process: extra passes, extra energy, and unnecessary variation.

Predict what is possible. Optimize what is feasible.

The model does not propose an idealized answer and hope the mill can execute it. Feasibility is part of the search itself.

We learned the relationship between process conditions and achievable thickness reduction per pass from historical production data. That predictive model became the engine of a constrained optimization framework that allocates reduction across the full sequence.

Candidate schedules are checked against the operation before they are accepted. The result is the shortest schedule the process can actually run.

HARD CONSTRAINTS / ALWAYS ACTIVE

  • 01Rolling forceVERIFIED
  • 02Current limitsVERIFIED
  • 03Tension rangesVERIFIED
  • 04Material strengthVERIFIED
  • 05Final gaugeVERIFIED
PREDICT×CONSTRAIN×OPTIMIZE×VERIFY

Small changes in sequence. Measurable changes in production.

15–20%

fewer passes

for selected coil scenarios

2–3%

lower total pass count

on average across production

≈2%

energy savings

from a more efficient rolling sequence

Beyond energy, the optimized schedules improve throughput, stabilize the process, and reduce dependency on operator-driven decisions.

LESS VARIATION · MORE THROUGHPUT · LOWER ENERGY

What looks fixed
may only be familiar.

Bring us the problem