https://doi.org/10.1140/epjs/s11734-026-02411-x
Review
Artificial intelligence and digital twins for failure prediction in data center cooling systems: a comprehensive literature review (2018–2026)
1
Institute of Industrial Ecology UB RAS, S. Kovalevskoy Str., 20, 620990, Ekaterinburg, Russia
2
Ural Federal University, Mira Str., 19, 620002, Ekaterinburg, Russia
3
HSE University, Myasnitskaya Str., 20, 101000, Moscow, Russia
a
This email address is being protected from spambots. You need JavaScript enabled to view it.
Received:
22
April
2026
Accepted:
23
May
2026
Published online:
8
June
2026
Abstract
This paper presents a review of artificial intelligence (AI) methods for failure prediction in data center cooling systems, with a focus on the integration of digital twins (DTs), physics-informed learning, and graph-based models. Positioned within complex network science, this review addresses a limitation of conventional graph approaches—their reliance on pairwise connectivity—whereas real-world failures often arise from multi-component interactions. A systematic literature search (2018–2026, with emphasis on 2023–2026) is conducted using predefined criteria. Studies are categorized according to modeling approaches, including classical machine learning, recurrent and deep neural networks, graph neural networks, transformers, physics-informed neural networks, and hybrid DT frameworks. The results indicate a shift from traditional data-driven methods toward hybrid, physics-informed architectures that improve predictive performance and interpretability. Graph and transformer models demonstrate strong capabilities in capturing spatio-temporal dependencies, while DTs enhance adaptability through continuous system synchronization. We highlight that pairwise graph representations remain insufficient for capturing collective failure patterns, pointing to an underexplored intersection of network science and predictive maintenance. The contributions of the review are a taxonomy of AI-based failure prediction methods, a critical analysis of practical limitations, and a dedicated analysis of benchmarking datasets (experimental, simulated, industrial) which reveals data scarcity and lack of standardized public datasets. This review concludes with research directions toward scalable, generalizable, and autonomous predictive maintenance, explicitly identifying higher-order network analysis as a promising pathway for future work.
Copyright comment Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
© The Author(s), under exclusive licence to EDP Sciences, Springer-Verlag GmbH Germany, part of Springer Nature 2026
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
