A Data-Driven Multiscale Framework for Nonlinear Rheo-Structural Dynamics in Non-Newtonian Fluid-Structure Interactions

Non-Newtonian fluid-structure interactions (NNF-FSI) present critical challenges in diverse engineering domains, including the analysis of sediment erosion at hydraulic structures and the design of viscoelastic dampers for seismic protection. Conventional computational models, which often rely on simplified Newtonian or basic non-Newtonian constitutive laws, can exhibit predictive errors exceeding 30% in complex, nonlinear scenarios. This paper introduces the Multiscale Nonlinear Rheo-Structural Dynamics (MNRSD) framework, a novel, data-driven methodology that synergistically integrates advanced rheological modeling via dimensionless groups, principles of nonequilibrium thermodynamics, adaptive finite element methods (AFEM), and machine learning (ML). The MNRSD framework demonstrates a substantial reduction in predictive errors for key physical quantities, achieving discrepancies of less than 10% relative to experimental benchmarks. This constitutes a 15–25% accuracy improvement over established models. The framework's predictive fidelity is rigorously validated through canonical analytical problems, high-fidelity computational simulations, and direct experimental measurements. By enhancing the precision of FSI predictions, MNRSD facilitates improvements in infrastructure resilience and operational efficiency, thereby supporting sustainable engineering objectives. This study establishes a robust, computationally efficient, and data-informed paradigm with broad applicability in civil, mechanical, and environmental engineering.