论文标题：《Thermodynamics-Informed Attention Networks for Concurrent Multiscale Modeling of Nonlinear Composites》

论文作者：Xuyang Zhang, Qiang Chen*, Xuefeng Chen

论文摘要：We present a novel multiscale thermodynamics-informed neural network framework (MulTIAN) for concurrent modeling of the inelastic response of unidirectional composites and structures under arbitrary loading paths. The core innovation lies in the development of the dual network architecture, where the primary attention network predicts internal state variables that encode the composite’s deformation history, while the auxiliary network predicts the Helmholtz free energy potential, from which constitutive relations are derived. The loss function is formulated to explicitly enforce fundamental thermodynamic constraints. Moreover, the developed framework is seamlessly integrated into ABAQUS for structural applications, enabling direct coupling between the composite structural model and the microscopic unit cell. Comparative studies with classical multiscale methods, conducted at both the material-point level and on the Meuwissen structure, central-hole laminate, elbow pipe, and L-shaped perforated plate under complex loading-unloading conditions, demonstrate the high predictive accuracy of the proposed framework. The MulTIAN achieves significant computational savings, reducing memory usage by up to a factor of 208 and computation time by up to a factor of 21 compared with conventional multiscale techniques. Finally, dog-bone tensile and V-notched rail shear tests were performed to validate the predictions against experimental data.

这是张旭阳同学发表IJSS之后，第二篇发表的论文，恭喜