Multiscale Shakedown Capacity Prediction of Parameterized Lattices Under Biaxial Loading Using Ensemble Learning.

The structural lightweighting of next-generation containerized energy storage systems requires reliable fatigue design methodologies for architected lattice materials subjected to complex multiaxial service loading. Despite extensive studies on static performance, fatigue capacity prediction of parameterized lattices remains computationally demanding and experimentally fragmented, particularly when cyclic characteristics cannot be idealized as simple periodic histories. This work develops a unified multiscale evaluation platform grounded in shakedown theory to directly predict multiaxial fatigue capacity for lattice structures without explicit cycle counting. A topology-agnostic nodal-coupling periodic boundary formulation ensures consistent homogenized response evaluation across diverse unit-cell geometries. Numerical robustness in stress computation is achieved through full-integration tetrahedral discretization (FITD), enabling stable treatment of bending-dominated lattice members. To facilitate rapid exploration of high-dimensional design spaces, an ensemble-learning surrogate is trained on multiscale shakedown datasets for capacity prediction and parameter sensitivity analysis. The framework is demonstrated on body-centered cubic and peanut-like auxetic lattices relevant to lightweight container structures. Validation studies confirm accurate FITD scheme-based shakedown fatigue loading prediction. The surrogate model achieves high predictive precision, and parametric analysis reveals topology-dependent fatigue drivers, establishing quantitative linkages between mesoscale geometric variables and shakedown-based fatigue capacity. The proposed methodology provides an efficient and scalable route for fatigue-oriented lattice design and optimization in energy storage container applications.