ArcelorMittal Belval: a deep dive into the simulation behind the modernization

The project

ArcelorMittal Long Products Luxembourg is investing in the largest modernization at its Belval site since switching from blast furnaces to EAFs in the 1990s. The goal: produce CO2-reduced steel locally instead of sourcing it from Poland and Germany. That means a 15% productivity increase, new steel grades, a revamped EAF with higher tapping weight, a shift from two-bucket to one-bucket scrap charging, and a brand-new vacuum degassing plant.

Every area of the plant is affected — from scrap delivery to billet storage. The question isn’t just “will it work?” but “where exactly will it break?”

The simulation approach

We modelled the entire production chain as a discrete event simulation: scrap yard, melt shop, casting, and run-out area. The model was validated against the highest-output period from a full year of production data, then continuously re-validated throughout the project as we iterated on the future concept.

Three distinct production plans were created, each combining different format combinations at each casting strand. Plans were designed for a full month and scaled to a year — ensuring the results hold up over long-term operation, not just cherry-picked good days.

The production schedules define what needs to happen — using static process and transport times. The simulation determines whether it actually can — accounting for ladle availability, crane conflicts, delayed tappings, longer-than-expected scrap loading, and every other dynamic interaction that static planning ignores. Critically, the deviations between the static plan and the simulation output were fed back to refine both the model and the production concept itself. This makes the simulation a design tool, not just a validation step.

Scrap yard: why pile height matters more than crane speed

The switch from two-bucket to one-bucket charging changed the entire scrap yard dynamic. Two portal cranes share a common rail track in a longitudinal pit, so they cannot operate independently — one crane’s position physically constrains the other. Scrap arrives by truck (one every 8 minutes on weekdays) and by 20-wagon train deliveries. Mobile material handlers transfer truck-delivered scrap to the main piles, while portal cranes can charge directly from piles, from the pit, or straight from train wagons. Piles are modelled as height-sensitive up to a maximum of 10 meters.

We first ran the future production with the existing pile layout. The result: because the two cranes share a rail, the east crane’s bucket had to travel much further than the west crane’s, causing late arrivals at the EAF.

An optimized layout repositioned scrap types within the pit. The impact:

• West crane bucket: 3 minutes faster on average (2.5 min minimum)
• East crane bucket: 7 minutes faster on average (3.9 min minimum)

But the bigger insight came from pile height. Loading times vary significantly based on how full the piles are. When the average pile height drops below about 8 meters — roughly bucket height — loading time increases dramatically because the crane must lower further and make more repositioning movements. Above 8 meters, no further improvement.

This means pile replenishment strategy is as important as crane speed. Keeping stock high and piling up scrap whenever there’s idle crane time directly reduces loading times.

Melt shop: can the cranes handle 15% more?

The melt shop model covers everything from EAF to casting — the complete hot ladle cycle with all full and empty transports. Two cranes handle ladles, slag pots, tundishes, pallets, casting molds, and segments. Their movement logic includes shortest-path routing around structural obstacles and inter-crane coordination with context-dependent yielding.

The concern: adding VD transports, deslagging on the crane hook, and 15% more heats — would the cranes become the bottleneck?

We identified two critical phases:

• Phase 1 — fastest steel productivity, shortest casting times, maximum heats per day. Crane utilization increased by 12%. Easily manageable.
• Phase 2 — every heat requires VD and deslagging, adding significant crane tasks even though heats per day are lower. Crane utilization increased by 23%. Close to maximum, but still no production delays found.

The key was analyzing utilization during specific peak periods, not averaging across the full simulation. An annual average looks comfortable; the stress points tell the real story.

Run-out area: the hidden bottleneck zone

The casting and run-out model tracked every billet from strand cutting through the cooling bed to storage or rolling mill entry. At Belval, different formats can be cast on each strand, making cooling bed loading from the pusher table complex.

We measured actual pushing and lifting speeds on the physical equipment and implemented them in the model. Various logics determine which format-length combinations can be pushed together onto the cooling bed.

This area — between strand cutting and end of cooling bed — was where we expected bottlenecks, and the simulation confirmed specific constraints that needed addressing in the future layout.

The bottom line: 1.3 MTPY confirmed

The simulation validated the production target of 1.3 million tonnes per year across all three production scenarios. Depending on the scenario, 4 to 13 unscheduled buffer days remained at year-end — flexibility for maintenance windows or unplanned downtime.

The simulation results were instrumental in securing the necessary investment for the modernization project. ArcelorMittal Luxembourg proceeded, with the first heat expected in 2025.

This study was co-authored with ArcelorMittal and presented at ABM Week 2024 in São Paulo and the European Electric Steelmaking Conference (EEC) 2024.