Multi-Physics Digital Twin Architecture for Smart Manufacturing

Abstract

Digital twin technology has attracted substantial interest in smart manufacturing, yet the integration of multi-physics coupling, reliability prediction, and organizational standardization into a unified architecture remains underexplored. This review synthesizes advances across these domains, examining how thermo-mechano-electromagnetic coupling analysis, physics-informed neural operators, and enterprise informatization standards collectively contribute to digital twin implementation. The analysis reveals persistent challenges—trade-offs between model fidelity and computational efficiency, difficulties in transferring coupling assumptions across contexts, and organizational barriers to standardizing digital workflows. Evidence suggests successful implementations emerge through iterative calibration rather than a priori design, with closed-loop compensation critical for maintaining predictive accuracy. Methodological limitations include reliance on deterministic assumptions and insufficient attention to data provenance. Future research should prioritize empirical validation across diverse industrial settings and human-in-the-loop approaches that combine algorithmic recommendations with expert judgment. The review concludes that a hybrid methodology—neither purely data-driven nor purely physics-based, offers the most promising pathway forward, acknowledging the irreducible uncertainty inherent in complex manufacturing systems.