航天器柔性伸展臂热致振动的高精度PINN预测方法OA
High-precision PINN prediction method for thermally induced vibration of flexible spacecraft booms
针对空间柔性伸展臂的热致振动响应预测问题,文章提出一种基于混合变量与硬约束的物理信息神经网络(PINN)预测框架.基于欧拉-伯努利梁理论和单向非耦合热传导假设,采用"热-力半解耦"策略预先拟合热弯矩并将其作为边界输入,以有效避免直接热-结构耦合导致的训练振荡.为解决梯度计算不稳定和神经网络"谱偏差"问题,引入混合变量降低控制方程阶数,并设计几何硬约束严格满足根部边界条件;结合傅里叶特征映射,提高网络对多频瞬态响应特征的捕捉能力.以高保真有限元仿真数据为基准进行了对比验证,结果表明,模型预测的末端位移与基准解高度一致,均方根误差在10-5 量级,相对误差不超过 1.27%.该方法在多物理场耦合动力学响应快速预测中表现出较高的准确性和鲁棒性.
To address the problem of predicting thermally induced vibrations in flexible spacecraft booms,a physics-informed neural network(PINN)prediction framework based on mixed variables and hard constraints was proposed.Based on the Euler-Bernoulli beam theory and the assumption of one-way uncoupled heat conduction,a thermo-mechanical semi-decoupling strategy was employed to precompute thermal bending moments as boundary inputs,effectively avoiding training oscillations caused by direct thermo-structural coupling.To mitigate gradient instability and the spectral bias problem,mixed variables were introduced to reduce the equation order,and geometric hard constraints were designed to satisfy the boundary conditions at the root.Fourier feature mapping was also integrated to enhance the network's ability to capture multi-frequency transient response features.Validations against high-fidelity finite element simulation data demonstrate that the predicted tip displacements agree well with the benchmark solutions,with the root mean square error reaching the order of 10-5 and the relative error remaining not exceeding 1.27%.The proposed method exhibits high accuracy and robustness in rapid prediction of multiphysics-coupled dynamic responses.
王钰超;张强;蔡守宇;刘磊;卫洪涛
郑州大学 力学与安全工程学院,郑州 450001郑州西亚斯学院 工学部,郑州 451150郑州大学 力学与安全工程学院,郑州 450001西北工业大学 航天学院,西安 710072郑州大学 力学与安全工程学院,郑州 450001
航空航天
热致振动预测物理信息神经网络航天器柔性伸展臂混合变量硬约束
thermally induced vibration predictionphysics-informed neural networkflexible spacecraft boommixed variableshard constraints
《航天器环境工程》 2026 (4)
257-266,10
国家自然科学基金项目(编号:51675430)
评论