A Data-Driven and Model-Driven Coupling Multiscale Model for Composite Structures

Recent advancements in artificial intelligence and big data analytics have propelled computational mechanics into an era of intelligent computation, marked by synergistic integration of quantum-enhanced processing and data-driven methodologies. This evolution has catalyzed a paradigm shift from conventional numerical approaches to innovative multi-framework solutions. Particularly in composite material analysis, modern homogenization techniques employing a two-phase strategy involving offline microscopic analysis and online macroscopic computation demonstrate superior computational efficiency through intelligent resource allocation. This study presents a novel hybrid computational framework that synergistically combines multi-scale homogenization principles with adaptive single-scale finite element modelling. The proposed methodology enhances traditional data-driven homogenization through three key innovations: (1) Implementation of Carrera’s Unified Formulation (CUF) for a dimensional reduction in complex structural modelling, (2) Development of a coupled data-model fusion architecture balancing physical principles with machine learning insights, and (3) Creation of an optimized computation workflow maintaining micron-scale accuracy while achieving microscale efficiency. The CUF-enhanced hybrid model was rigorously tested through deflection analysis of composite structures under cantilever and simply supported constraints. Validation against conventional FEM simulations and analytical benchmarks confirms the method’s accuracy, with comparative studies revealing <5% deviation in displacement predictions while achieving 40–60% reduction in computational resource requirements. This integrated approach demonstrates particular efficacy in layered composite analysis, successfully addressing the accuracy-efficiency trade-off inherent in multi-scale modelling through intelligent dimensionality management and physics-informed machine learning strategies. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.