Nonlinear optimization of the composite laminate possesses a prohibitive computational cost due to the frequent and time-consuming nonlinear analysis. To address this issue, the geometrically nonlinear analysis is completely surrogated by the attention-based Sequence-to-Sequence (Seq2Seq) model in this work. Specifically, the model maps its inputs, consisting of lamination parameters (LP) and the linear analysis results, into a sequence of deformation states in the equilibrium path, replacing the nonlinear analysis for various designs of a composite structure with different LP. To evaluate the strength failure of the structure with large deformation, the strength failure criterion is formulated in terms of nonlinear strains for the isogeometric continuum shell element. Whereafter, the failure load is maximized by globally searching the feasible domain of LP with the constrained genetic-algorithm optimizer. The efficiency and reliability of the proposed approach are validated by comparing the existing deep learning method in the numerical tests on two laminated shell structures. In the two tests, the absolute percentage errors of the failure load obtained from the Seq2Seq prediction are 2.77% and 0.96% respectively, and the time costs of performing the Seq2Seq-based optimization are merely 3.5% and 8.3% of the existing deep learning method. After the strength optimization process, the failure loads of the optimal laminates fulfill a nearly 1.5-time improvement in contrast to their quasi-isotropic cases.

中文翻译：

---

使用基于注意力的 Seq2Seq 模型预测复合材料层合板强度优化的几何非线性行为


由于频繁且耗时的非线性分析，复合材料层压板的非线性优化具有高昂的计算成本。为了解决这个问题，在这项工作中，几何非线性分析完全由基于注意力的序列到序列（Seq2Seq）模型替代。具体来说，该模型将其输入（包括层压参数 (LP) 和线性分析结果）映射到平衡路径中的一系列变形状态，从而取代具有不同 LP 的复合材料结构的各种设计的非线性分析。为了评估大变形结构的强度破坏，根据等几何连续体壳单元的非线性应变制定了强度破坏准则。此后，通过使用约束遗传算法优化器全局搜索 LP 的可行域，使故障负载最大化。通过在两种层合壳结构的数值试验中比较现有的深度学习方法，验证了该方法的效率和可靠性。在两次测试中，Seq2Seq预测获得的故障负载的绝对百分比误差分别为2.77%和0.96%，而执行基于Seq2Seq的优化的时间成本仅为现有深度学习方法的3.5%和8.3% 。经过强度优化过程后，最佳层合板的失效载荷比准各向同性情况提高了近 1.5 倍。