Virtual-physical collaborative intelligent fault diagnosis for marine rotating machinery under data scarcity towards digital twin

Abstract

In the domain of marine rotating machinery (MRM), the scarcity of high-quality fault data poses a critical barrier to the development of reliable and generalizable intelligent fault diagnosis (IFD) systems. To address this challenge, virtual-physical data collaboration under the digital twin paradigm has emerged as a promising direction. This review examines 117 representative publications, offering a comprehensive analysis of virtual-physical IFD approaches for MRM under data-scarce conditions. The review identifies key failure modes in critical MRM components and introduces a virtual-physical collaborative IFD framework, integrating high-fidelity virtual data and limited physical measurements to construct robust diagnostic models. Three categories of virtual modeling techniques are analyzed, along with fidelity validation strategies for ensuring model reliability. It further compares three collaborative learning strategies: parameter sharing, domain adaptation, and adversarial transfer learning, assessed for diagnostic accuracy and data augmentation effectiveness. Comparative results reveal that parameter sharing fits aligned domains, domain adaptation improves generalization under imbalance, and adversarial learning supports diagnosis when fault data is entirely absent. This review concludes by outlining key challenges such as system-level virtual modeling and lightweight deployment, and recommending future directions to support scalable DT-driven IFD in maritime applications, offering critical insights for managerial decision-making on adopting these technologies.