Projects
Active2025–2028Distributed MLDigital twinsSteel

ProcTwin

EU Horizon project on digital twins for steel: physics-based models, soft sensors and distributed machine learning across a chain of process steps.

Map of the ProcTwin consortium partners across Europe
The ProcTwin consortium: research institutes, steel producers and technology companies in Sweden, Spain, Germany, Austria and Belgium.

ProcTwin, short for Integrated modelling for sustainable and optimized steel manufacturing processes, is a four-year EU Horizon Europe project within the Clean Steel Partnership. It builds a demonstration platform that predicts and visualises the best way to run a whole chain of processing steps in a steel plant, with energy efficiency and product quality as the two goals.

The problem

Steel is made in a sequence of steps, such as continuous casting, reheating, hot working, quenching and levelling. Each step has its own control system and is usually optimised on its own. But the steps are strongly connected: how a slab is cast affects how it must be reheated, and how it is rolled affects how it can be quenched. Optimising one step at a time leaves energy and quality on the table.

Doing better needs a model of the whole chain, built from data that sits in different systems, at different sites and in different companies, and that cannot simply be pooled.

The approach

ProcTwin couples the process steps through a combination of three kinds of models:

  • physics-based models of each process step, adapted so that they can be connected to each other;
  • soft sensors and new measurement techniques, which estimate quantities that are hard to measure directly on the production line;
  • distributed machine learning across the process chain, with secure data sharing, so that models can learn from data held by different steps and different partners.
A steel process chain from continuous casting through reheating, hot working and quenching to levelling. Each step has its own physics-based model and soft sensors; a distributed machine learning layer connects them, and a digital twin predicts quality and energy use for the whole chain. Two use cases run at Celsa in Spain and SSAB in Sweden.
ProcTwin couples separately controlled process steps into one digital twin of the chain, tested in two industrial use cases.

The platform is demonstrated in two parallel industrial use cases, at Celsa in Spain and SSAB in Sweden.

Our part

At the University of Skövde, our work focuses on the distributed machine learning that connects the process steps: learning from data that stays with each step and each partner, in the spirit of federated learning. PhD student Abed Al Bakkour works on distributed AI in the steel industry within the project.

The consortium

The project is coordinated by Swerim (Sweden) and brings together research institutes, steel producers and technology companies:

Partner Country Type
Swerim (coordinator) Sweden Research institute
University of Skövde Sweden University
SSAB Sweden Steel producer
Prevas Sweden Technology company
VDEh-Betriebsforschungsinstitut (BFI) Germany Research institute
Materials Center Leoben Austria Research centre
CARTIF Spain Research centre
Celsa Spain Steel producer
Global Steel Wire Spain Steel producer
Aquiles Solutions Spain Technology company
Ingeniería y Sistemas de Ensayos No Destructivos Spain Inspection technology
ESTEP, the European Steel Technology Platform Belgium Industry platform

Results so far

The project started in January 2025 and runs until the end of 2028. Publications, software and demonstrators will be listed here as they appear; follow News for updates.

PhD students in the project

AB

Abed Al Bakkour

University of Skövde, Sweden

Distributed AI in the steel industry.

Interested in this work? Questions, collaboration or a thesis on a related topic: I am happy to hear from you.

open↑↓ move