多项式混沌
人工神经网络
混沌(操作系统)
算法
多项式的
计算机科学
数学
应用数学
人工智能
深层神经网络
理论(学习稳定性)
随机共振
随机过程
噪音(视频)
降噪
信号处理
深度学习
控制理论(社会学)
作者
Van-Thien Tran,Trung-Kien Nguyen,Nuttawit Wattanasakulpong,Thuc P. Vo
标识
DOI:10.1080/15397734.2026.2695183
摘要
This study introduces PCE–BiDNN, a novel hybrid computational framework that integrates polynomial chaos expansion (PCE) with bidirectional deep neural networks (BiDNNs), to perform stochastic reliability assessment of graphene platelets reinforced (GPLR)-functionally graded (FG)-triply periodic minimal surfaces (TPMS) microplates subjected to material property scatter modeled through lognormal distributions with a coefficient of variation of 10%. A third-order shear deformation theory (TSDT) combined with modified couple stress theory (MCT) and a Chebyshev–Ritz formulation employing hybrid second-kind Chebyshev polynomials captures size-dependent bending, free vibration, mechanical buckling, and thermal buckling responses across SSSS, CSCS, and CCCC boundary conditions. Effective mechanical properties are evaluated via Halpin–Tsai micromechanics across six porous architectures, encompassing sheet-based TPMS types (Primitive, Gyroid, I-graph, and Wrapped Package-graph) and cellular topologies (closed-cell and open-cell structures). Results validate rapid five-term convergence and identify Primitive-GPL III as optimal across free vibration, mechanical buckling, and static bending, while GPL I yields superior thermal buckling resistance due to its uniform through-thickness distribution reducing thermally induced compressive stresses. Structural responses consistently intensify with decreasing thickness-to-length scale ratio, confirming strong size-dependent behavior. PCE–BiDNN delivers over 97% reduction in computational cost relative to direct MCS while maintaining errors well below 1%, with BiDNN augmentation providing improved PDF smoothness and tail robustness over baseline PCE surrogates. Sobol sensitivity analysis confirms mass densities of GPL and metal yield marginally higher stochastic influence than Young’s moduli, with predominantly additive parameter contributions that simplify reliability-based optimization of microscale GPLR-FG-TPMS architectures.
科研通智能强力驱动
Strongly Powered by AbleSci AI