Human–AI-Assisted digital twins for advanced manufacturing: toward industry 5.0

Digital twins are increasingly recognized as key enablers of human-centric Industry 5.0 manufacturing. Yet existing reviews rarely examine digital twin architectures, AI methods, human-in-the-loop collaboration, and interaction technologies within a unified analytical framework. This paper systematically synthesizes eighty-two peer-reviewed studies, identified through a PRISMA-2020 systematic search of Scopus, the Web of Science Core Collection, and IEEE Xplore, into a comprehensive analysis of human–AI-assisted digital twins in advanced manufacturing. Four research questions structure the review, addressing digital twin architectures, human-in-the-loop collaboration models, AI methods, and manufacturing applications, as well as interaction technologies aligned with Industry 5.0. The qualitative synthesis is complemented by a quantitative analysis comprising a seven-dimensional quality assessment, Technology Readiness Level (TRL) analysis, AI-method-family frequency analysis, and ethics-coverage classification. The findings indicate that digital twins are evolving from machine-centric monitoring toward intelligent coordination platforms that integrate sensing, simulation, AI reasoning, and human interaction. However, most studies remain at laboratory-scale validation, and alignment with Industry 5.0’s resilience and sustainability pillars is often more aspirational than empirically demonstrated. The review identifies three priorities for future research. First, develop interoperable architectures that span the full perception–reasoning–interaction stack rather than individual layers in isolation. Second, integrate human-state-aware AI grounded in theories of trust calibration and shared autonomy. Third, validate proposed systems through industrial pilot deployments while explicitly addressing worker-data ethics. Per-paper quality scores, Technology Readiness Level assignments, and AI-method classifications are released as supplementary material to support reuse, benchmarking, and replication.