HARMONY

The HARMONY project aims to develop interpretable, robust, and real-time capable hybrid process models for the vacuum recirculation (RH) process in steel production—a key technology in the transition toward sustainable and resource-efficient metallurgy. By integrating symbolic (physics-based) and sub-symbolic (data-driven) AI into a hybrid modelling approach the project will enable the extraction of non-measurable process parameters, improve process understanding, and support operational optimisation.

The project combines domain-specific knowledge engineering, advanced sensor systems, and data pre-processing to create a unified, high-quality data pool. This foundation supports the development of hybrid models that are modular, easily retrainable, and beneficial for future industrial control applications. A particular focus is placed on model robustness, including tolerance to noisy or incomplete data and generalizability across operating conditions. 

A methodological innovation of the project is the use and further advancement of knowledge-driven symbolic regression. By incorporating so-called structural building blocks, symbolic models, such as physical laws and domain-specific equations, which can be directly integrated into the otherwise sub-symbolic, i.e., data-driven, modelling process. This approach enables the creation of hybrid models that are not only accurate but also physically consistent and interpretable. The incorporation of prior knowledge not only supports handling noisy data more effectively but is also expected to enhance the extrapolation capabilities of the models, thereby improving their performance in previously unseen scenarios.

The project emphasizes human-AI collaboration, ensuring that models remain comprehensible and usable by non-experts and adaptable to evolving plant conditions, supporting long-term usability and responsible AI adoption in industry.

Validation case studies, sustainability assessments, and industrial applicability evaluations in a controlled environment will demonstrate the impact of the developed models. 

HARMONY aims at creating hybrid process models for a vacuum recirculation process that will become a key process during the transformation towards sustainable steel production routes.

The hybrid models will:

1. Integrate domain-specific knowledge with symbolic and subsymbolic modelling techniques.
2. Enable the extraction of non-measurable, process-relevant parameters,
3. Contribute to a deeper understanding of complex metallurgical processes,
4. Lead to improved process models with higher predictive accuracy,
5. Support resource-efficient process optimization, contributing to SDG 9 and SDG 12.