The Steel That Learns: Inside CELSA’s Journey with ALCHIMIA

By David Blázquez, CELSA Group

When we look back at the journey of ALCHIMIA, it’s easy to see how ambitious its vision was: bringing Artificial Intelligence to steelmaking through decentralised, collaborative learning — without compromising industrial data privacy. Three years later, that vision feels much closer to reality.

For CELSA Group, ALCHIMIA has been a unique opportunity to test what federated and continual learning really mean in a complex, multi-plant environment. The project connected our sites in Barcelona, Bayonne and Ostrowiec under a shared goal: to make our processes smarter, greener, and more resilient.

The development of analytical and neural models for our Electric Arc and Ladle Furnaces has shown how distributed AI can evolve across plants — learning from differences rather than ignoring them. And while the validation phase is still ongoing, the groundwork is there: a unified data model, robust interfaces, and people ready to use these new tools.

That last part — people — has been one of the most rewarding aspects. In October, we launched ALCHIMIA’s first global training sessions, bringing together operators and engineers from all three plants to explore the new dashboards and workflows. Around the same time, we presented our experience at the ESTEP Annual Event in Udine, joining the European community shaping the future of green and digital steelmaking.

As we approach the final validation phase, the project continues to evolve. The upcoming visit from Cardiff University to our Bayonne plant will capture the human side of this transformation: how digital tools reshape daily work, collaboration, and decision-making on the shop floor.

ALCHIMIA has proven that innovation in steel is not just about algorithms or data pipelines — it’s about people learning, adapting, and building together. The real success is not a model that predicts better, but an industry that learns better.