摘要
为提高装夹布局优化计算的效率,同时考虑加工过程中振动对变形的影响,提出了融合改进的反向传播神经网络(back propagation neural network,BPNN)与快速非支配排序遗传算法(nondominated sorting genetic algorithms,NSGA-Ⅱ)的优化模型。首先,基于“N-2-1”定位原理,以定位点坐标为设计变量,薄壁件装夹变形和主振型位移为优化目标,通过有限元仿真建立了神经网络训练样本集;其次,引入粒子群算法(particle swarm optimization,PSO)改进神经网络,通过对有限样本集训练,构建了定位布局与装夹变形和振动位移之间的代理模型。实例结果表明,改进后的神经网络对装夹变形的预测精度提高了93%,对振动变形的预测最大误差仅为1.8%;最后,通过遗传算法求解预测模型得到了定位布局帕累托解集,进一步提高了优化效率。
In order to improve the efficiency of the optimization calculation of the clamping layout,and consider the influence of vibration on the deformation during the machining process,a combinate model of improved back propagation neural network(BPNN)and nondominated sorting genetic algorithms(NSGA)optimization was proposed.First,based on the"N-2-1"positioning principle,with the coordinates of the positioning point as the design variables,and the clamping deformation of the thin-walled parts and the displacement of the main mode shape as the optimization goals,a neural network training sample set was established through finite element simulation;Second,the particle swarm optimization(PSO)is introduced to improve the neural network,and the surrogate model between the positioning layout and the clamping deformation and vibration displacement is constructed by training the limited sample set.The example results show that the improved neural network can improve the prediction accuracy of clamping deformation by 93%,and the maximum error of prediction of vibration deformation is only 1.8%;Finally,a Pareto solution set of positioning and layout is obtained by solving the prediction model by genetic algorithm,which further improves the optimization efficiency.
作者
胡凯鑫
马嵩华
HU Kai-xin;MA Song-hua(School of Mechanical Engineering,Shandong University,Jinan 250061,China)
出处
《组合机床与自动化加工技术》
北大核心
2023年第3期1-4,共4页
Modular Machine Tool & Automatic Manufacturing Technique
基金
国家自然科学基金(51505254,51975326)
中国博士后基金(2021M691704)
山东省博士后创新基金(202101017)。
