基于辛叠加方法与迁移学习的功能梯度任意四边形开孔板自由振动问题求解OA
Free vibration solution of functionally graded plates with arbitrary quadrilateral cutouts based on symplectic superposition method and transfer learning
功能梯度开孔板是工程中常见的承力结构,研究其动力行为具有重要意义.针对规则形状开孔板,通过辛叠加方法结合子域分解技术已能系统解析求解,但是当研究对象扩展至应用更加广泛的任意四边形开孔板时,解析求解因边界更加复杂而面临巨大挑战.尽管可以通过近似/数值方法进行求解,但存在计算耗时长、网格依赖性强等问题,导致结果精度不足.建立结合辛叠加方法与迁移学习技术的求解框架,利用可解析求解的功能梯度矩形开孔板自振频率进行知识迁移,从而高效高精度求解功能梯度任意四边形开孔板自由振动问题.首先,基于多层感知机神经网络对大样本解析解数据集进行预训练,提取不同几何形状间几何参数的复杂映射关系;其次,基于小样本有限元仿真数据对预训练模型进行迁移,将矩形开孔板自振频率数据集中的知识有效迁移至任意四边形开孔板自由振动问题;最后,利用均方误差和决定系数等指标,对所提求解框架用于功能梯度任意四边形开孔板自振频率预测的准确性及适用性进行验证.利用迁移学习技术,基于小样本数据与已有解析解,高效高精度求解了功能梯度任意四边形开孔板的自由振动问题,并且通过调整模型参数可适用于不同工况,为复杂形状板的力学分析提供了新思路.
Functionally graded plates with cutouts are common load-bearing structures in engineering,making the study of their dynamic behavior crucial.For plates with regularly shaped cutouts,the symplectic superposition method combined with subdomain decomposition technique can be used to establish a comprehensive analytical solution framework.However,when the research object is extended to plates with arbitrarily shaped cutouts,which have broader applications,the analytical solution faces significant challenges due to the increased complexity of the bound-ary conditions.While approximate or numerical methods exist,they may suffer from time-consuming computations and strong mesh dependency,often leading to insufficient accuracy in results.A novel solution framework that inte-grates the symplectic superposition method with the transfer learning technique is established.It leverages the analyti-cally solvable natural frequencies of functionally graded rectangular plates with rectangular cutouts for knowledge trans-fer,thereby achieving efficient and highly accurate solutions for the free vibration problems of functionally graded plates with arbitrary quadrilateral cutouts.First,a multi-layer perceptron neural network is pre-trained on large-scale analytical solution dataset to extract the complex mapping relationships of geometric parameters between different shapes.Sec-ond,the pre-trained model is transferred based on few-shot finite element simulation data,effectively transferring knowledge from the natural frequency dataset of the rectangular plates with rectangular cutouts to the free vibration of plates with arbitrary quadrilateral cutouts.Finally,the accuracy and applicability of the proposed solution framework for predicting the natural frequencies of functionally graded plates with arbitrary quadrilateral cutouts are validated,using metrics such as mean squared error and coefficient of determination.The transfer learning technique is used to lever-age small-sample data and existing analytical solutions to efficiently and accurately solve free vibration problems of functionally graded with arbitrary quadrilateral plates cutouts.The proposed framework,adaptable to various working conditions through model parameter adjustments,offers a novel strategy for the mechanical analysis of complex-shaped plates.
程超宇;徐典;郭程洁;李进宝;李锐
大连理工大学 力学与航空航天学院 工业装备结构分析优化与CAE软件全国重点实验室,大连 116024大连理工大学 力学与航空航天学院 工业装备结构分析优化与CAE软件全国重点实验室,大连 116024大连理工大学 力学与航空航天学院 工业装备结构分析优化与CAE软件全国重点实验室,大连 116024大连理工大学 力学与航空航天学院 工业装备结构分析优化与CAE软件全国重点实验室,大连 116024大连理工大学 力学与航空航天学院 工业装备结构分析优化与CAE软件全国重点实验室,大连 116024
航空航天
四边形开孔板自由振动辛叠加方法迁移学习
quadrilateral plates with cutoutsfree vibrationsymplectic superposition methodtransfer learning
《航空学报》 2026 (5)
171-182,12
国家自然科学基金(12372067,12502071)国防基础科研计划(JCKY2021205B003)大连市杰出青年科技人才项目(2024RJ005)辽宁省自然科学基金杰出青年基金(2025JH6/101100005) National Natural Science Foundation of China(12372067,12502071)National Defense Basic Scientific Research Program of China(JCKY2021205B003)Dalian Science Fund for Distinguished Young Scholars(2024RJ005)Liaoning Science Fund for Distinguished Young Scholars(2025JH6/101100005)
评论