Cloud-based digital twins: ow simulations can predict failures in industry

Abstract

The relevance of this study stems from the increasing complexity of industrial systems and the need to process large data streams in real time to ensure reliable monitoring, predict technical failures, and support decision- making. The aim of the work was to identify typical architectural configurations of digital twins in cloud environments and determine how the distribution of analytical functions across architectural levels affects the efficiency of such systems in production settings. The research methodology was based on a critical analysis of interdisciplinary sources using content analysis, comparative analysis, and SWOT analysis, which enabled thematic structuring of the material according to architectural, algorithmic, and organisational-regulatory parameters. As a result, it was established that a multi-level digital twin model provides a universal foundation for describing architectures in mechanical engineering, energy, and automated manufacturing. Hybrid solutions that transferred part of the analytics to the edge layer offered increased resilience to network failures and better adaptation to changes in the technical condition of assets. It was found that system efficiency depended not only on the topology of computational tasks but also on the ability of analytical models to process streaming data, maintain accuracy amid data drift, and remain interpretable in critical decision-making contexts. It was shown that key barriers to implementation remained the fragmentation of approaches to functional decomposition, the absence of unified standards, and sensitivity to unstable interactions between components. Based on cross-industry comparison, a typology of digital twin architectures was developed, taking into account the nature of analytics distribution and its integration with cloud infrastructure. The results provide a conceptual foundation for further empirical research aimed at practical verification of the stability, adaptability, and scalability of digital twins in production environments.