ALCHIMIA

ALCHIMIA project develops a system for Federated Learning to improve the performance of optimisation models by letting these learn from other models with an analogous logic. The domain is steel production, and the optimisation has targets, such as efficient operation, efficient scrap reuse and minimal environmental impact. The models are data driven, and the techniques vary, including but not limited to neural networks, ensemble learning and physical models. For the models operate, they need data from online production systems. For the Federated Learning concept, model-related weight information is exchanged with a centralised server outside of the production plant.

Regarding standardisation, the project has KPIs to deliver 3 contributions, form 3 liaisons and present project results in 3 standardisation group meetings.

Reason for applying to HSbooster.eu services

The project partners are currently still learning standardisation. Two ALCHIMIA partners have already memberships in ISO standardisation committees or groups via DIN and BSI, four in total: ISO/TC 17/SC 3, ISO/TC 17/SC 21, ISO/TC 17/WG 28 and ISO/IEC JTC 1/SC 42. However, the partners experience that they lack the specific expertise required to contribute to the current work items. Additionally, we consider it would be most appropriate to contribute from project results rather than from the generic knowledge of the partners, which is not always directly related to the project.

On the other hand, it is unclear how well the project results could be generalised enough for a standard. The project is already preparing a to-be-open-standard Smart Data Model for message-based communication, however, the KPIs necessitate more contributions. A CEN Workshop Agreement (CWA) has been considered, however, the topic is missing. We hope to receive an expert viewpoint about the standardisation potential, especially the topics suitable for standardisation. 

Additionally, now that the partners alone are ineligible for an (ISO) liaison, the liaisons would realise on the project level. This seems difficult now that the project results do not seem to suit directly for any active work item and the potential to generalise the results seems low.