基于神经网络的数据挖掘模型在吸能装置上的应用  

Data Mining Model Based on Neural Network and Its Application on Anti-Climber Device

作　　者：车全伟 雷成[3] 李玉如 朱涛[4] 唐兆[4] 姚曙光[1] CHE Quanwei;LEI Cheng;LI Yuru;ZHU Tao;TANG Zhao;YAO Shuguang(Traffic&Transportation Engineering,Central South University,Changsha 410075,China;CRRC Qingdao Sifang Co.,Ltd.,Qingdao 266111,China;Engineering&Technology Research Center of Rail Transit Intelligent Safety of Henan Province,Zhengzhou Railway Vocational and Technical College,Zhengzhou 451460,China;Traction Power National Key Laboratory,Southwest Jiaotong University,Chengdu 610031,China)

机构地区：[1]中南大学交通运输工程学院,湖南长沙410075 [2]中车青岛四方机车车辆股份有限公司,山东青岛266111 [3]郑州铁路职业技术学院河南省轨道交通智能安全工程技术研究中心,河南郑州451460 [4]西南交通大学牵引动力国家重点实验室,四川成都610031

出　　处：《西南交通大学学报》2021年第5期995-1001,共7页Journal of Southwest Jiaotong University

基　　金：国家重点研发计划课题(2016YFB1200404);四川省科技计划(2019YJ0216)。

摘　　要：针对传统有限元分析方法对机车车辆结构耐撞性计算效率低的问题,在已有仿真分析数据基础上,引入机器学习方法,对车辆关键结构的耐撞性以及碰撞安全性进行分析预测.首先,建立基于神经网络的数据挖掘模型,在此基础上构建车辆关键结构的碰撞响应预测方法;其次,通过试验验证了防爬吸能装置有限元模型的正确性,以此模型为基础获得不同壁厚防爬吸能装置的碰撞响应仿真数据;然后,以吸能装置壁厚作为模型输入,不同壁厚所对应的位移、速度、界面力和内能等碰撞响应作为模型输出,将有限元仿真数据用于模型训练,优化后的数据挖掘模型的拟合优度在0.922以上;最后,为验证模型预测的准确性,将碰撞数学模型的预测结果与有限元仿真结果进行对比,速度、位移、界面力和内能的平均相对误差分别为7.10%、4.51%、6.20%和2.50%.研究结果表明:基于神经网络构建的数据挖掘模型在保证精度的情况下,能很好地反映防爬吸能装置的碰撞特性,大幅降低了计算时间,提高了计算效率.Given the low efficiency of traditional finite element analysis method in calculating the crashworthiness of locomotive and vehicle structure,machine learning method is introduced to analyze and predict the crashworthiness and crash safety of vehicle structure on the basis of the existing simulation analysis data.Firstly,the neural network data mining model is established,and according to this model the prediction method for thecollision response of vehicle key structure.Secondly,tests are conducted to validate the finite element model of the anti-climber device,and the collision response data of the finite element model are obtained under different wall thicknesses.Then,the wall thickness of the anti-climber device is used as the model input,and the corresponding displacement,velocity,interfacial force and internal energy are used as the model output.The simulation data are used to train the model,the goodness fit of which is above 0.922.Finally,in order to test and verify the model,the predicted results of the energy absorption device are compared with the finite element simulation results,showing that the average relative errors of velocity,displacement,interfacial force,and internal energy are 7.10%,4.51%,6.20%,and 2.50%,respectively.The results indicate that the data mining model based on neural network can well reflect the collision characteristics of the anti-climber device with the precision;meanwhile,its computation time is greatly reduced and thecomputational efficiency is significantly improved.

关 键 词：机器学习 神经网络 碰撞响应 数据挖掘模型 吸能装置 计算效率 

分 类 号：U270.1[机械工程—车辆工程]