Federated learning over industrial data spaces: an end-to-end operational pipeline framework for predictive maintenance
DYNA · 01.11.2026 · 101 (6) · 576-584
DOI: 10.52152/d11698 ↗Kurz gefasst
Fachartikel in der Zeitschrift DYNA, online erschienen am 01.11.2026. Verfasst von FELIX LARRINAGA BARRENECHEA und 8 weiteren Autor:innen.
Abstract (Originalsprache)
Current Industrial predictive maintenance depends on Data-Driven Artificial Intelligence. However, operational datasets are isolated due to privacy, sovereignty and regulation concerns. Data heterogeneity among manufacturers and the lack of a framework that enables data sharing introduce additional obstacles. To address these issues, this paper proposes an operational pipeline based on a generic sovereign and distributed manufacturing architecture through Federated Learning workflows within an Industrial Data Space utilizing standardized Digital Twins. The operational validation of the framework is addressed through a cyber-physical manufacturing use case involving distributed machining spindle heads, which provides a first demonstration of collaborative diagnostics under data sovereignty constraints. At the semantic and interoperability level, the Asset Administration Shell standard is instantiated via Eclipse BaSyx. On the one hand, to translate heterogeneous edge vibration signals into uniform digital representations and on the other, to enable collaborative FL model and parameter sharing using the IDTA 02060-1-0 AI Submodel Nameplate specification. In addition, Eclipse Dataspace Components connectors orchestrate the policycompliant and auditable exchange of model parameters, so that every training round is bound to negotiated usage contracts. • Keywords: Federated learning, Data standardization, IDTA Digital Twin, Asset Administration Shell, Predictive maintenance, Data Spaces
Zitieren
APA
BARRENECHEA, F. L., Porto, A., Montejo, E., Paz, I., Legaristi, J., Charramendieta, S., Bartolome, M. G., Ulloa, M. I., & IRIONDO, I. R. (2026). Federated learning over industrial data spaces: an end-to-end operational pipeline framework for predictive maintenance. DYNA, 101(6), 576-584. https://doi.org/10.52152/d11698
Vancouver
BARRENECHEA FL, Porto A, Montejo E, Paz I, Legaristi J, Charramendieta S, et al. Federated learning over industrial data spaces: an end-to-end operational pipeline framework for predictive maintenance. DYNA. 2026;101(6):576-584. doi:10.52152/d11698
BibTeX
@article{BARRENECHEA2026,
author = {BARRENECHEA, FELIX LARRINAGA and Porto, Alain and Montejo, Elena and Paz, Iñaki and Legaristi, Jon and Charramendieta, Santiago and Bartolome, Mikel Garcia and Ulloa, Michel Iñigo and IRIONDO, ITZIAR RICONDO},
title = {Federated learning over industrial data spaces: an end-to-end operational pipeline framework for predictive maintenance},
journal = {DYNA},
year = {2026},
volume = {101},
number = {6},
pages = {576--584},
doi = {10.52152/d11698}
}
EUVIMED