基于NNBoost的卫星用复合材料层合板结构不确定性固有频率分析方法研究  

Research on NNBoost-based Uncertain Natural Frequency of Composite Laminates for Satellite Structures

作　　者：赵琳 刘源[1] 曹喜滨[1] 侯耀东 张俊杰[2] ZHAO Lin;LIU Yuan;CAO Xibin;HOU Yaodong;ZHANG Junjie(Research Center of Satellite Technology,Harbin Institute of Technology,Harbin 150001;Center for Precision Engineering,Harbin Institute of Technology,Harbin 150001)

机构地区：[1]哈尔滨工业大学卫星技术研究所,哈尔滨150001 [2]哈尔滨工业大学精密工程研究所,哈尔滨150001

出　　处：《机械工程学报》2023年第24期242-250,共9页Journal of Mechanical Engineering

基　　金：国家自然科学基金资助项目(52372351,51875119)。

摘　　要：为实现卫星用复合材料层合板固有频率的精确分析,综合考虑结构的加工误差、材料随机偏差等不确定性因素,研究并提出一种采用神经网络增强(Neural network boosting,NNBoost)模型对正交各向异性复合材料层合板固有频率进行不确定性分析的方法。将NNBoost模型作为固有频率求解与预测的代理模型,其目标函数设定为损失函数与正则项之和,求解过程中采用一种基于泰勒展开式的梯度下降方法更新权重和阈值以加速收敛。采用该方法,分析考虑输入参数随机性时正交各向异性复合材料层合板固有频率的统计特性。仿真试验结果表明,与直接蒙特卡洛模拟(Monte Carlo simulation,MCS)相比,采用NNBoost方法在保证预测精度的同时显著提高了求解效率;与传统反向传播(Back propagation,BP)神经网络方法相比,采用NNBoost方法预测结果的均方误差小于BP神经网络预测结果的均方误差且误差收敛更加平稳。In order to realize accurate analysis of the natural frequency of composite laminates for satellite structures,a method for analyzing the uncertainty of the natural frequency orthotropic composite laminates by using neural network boosting(NNBoost)model is proposed by considering the factors of uncertainty such as machining errors and material random deviations.In this paper,the NNBoost model is used as a surrogate model for solving and predicting the natural frequency.The objective function is set as sum of the loss function and the regularization term.In the solving process,a gradient descent method based on Taylor expansion is used to update the weights and thresholds to accelerate the convergence.Using this method,statistical characteristics of the natural frequencies of an orthotropic composite laminate are analyzed with the randomness of input parameters considered.The simulation results show that compared with the direct Monte Carlo simulation(MCS),the proposed method significantly improves the solution efficiency while ensuring the prediction accuracy.Compared with the traditional back propagation(BP)neural network method,the mean square error of the prediction results with this method is smaller than that of the BP neural network and the error convergence is more stable.

关 键 词：卫星结构 复合材料层合板 神经网络增强模型 不确定性固有频率 

分 类 号：V19[航空宇航科学与技术—人机与环境工程]