首页 | 官方网站   微博 | 高级检索  
     

基于主动学习神经网络的转向架构架疲劳可靠性分析
引用本文:陈鹏.基于主动学习神经网络的转向架构架疲劳可靠性分析[J].计算力学学报,2023,40(3):491-498.
作者姓名:陈鹏
作者单位:大连交通大学 车辆工程学院, 大连 116028
摘    要:为了提高转向架构架疲劳可靠性分析的精度与效率,提出一种主动学习BR-BP神经网络模型与Monte Carlo法相结合的可靠性分析方法。该方法针对BP神经网络的缺陷,使用贝叶斯正则BR(Bayesian regularization)算法作为训练算法,以提高神经网络的拟合精度与收敛速度,并考虑可靠性分析的固有特点,构造了一种适用于BP神经网络的主动学习函数,用于指导最佳样本点的选择。提出的学习函数不仅保证了样本点分布在极限状态函数附近,还考虑了样本点的预测误差以及样本点分布对失效概率计算的影响。转向架构架可靠性分析结果表明,本文方法在提高拟合精度的同时兼顾了计算效率,验证了所提方法的优越性与可行性。

关 键 词:结构可靠度  BP神经网络  序列采样  转向架构架  疲劳可靠性分析
收稿时间:2021/11/16 0:00:00
修稿时间:2022/1/27 0:00:00

Fatigue reliability analysis of bogie frame based on active learning neural network
CHEN Peng.Fatigue reliability analysis of bogie frame based on active learning neural network[J].Chinese Journal of Computational Mechanics,2023,40(3):491-498.
Authors:CHEN Peng
Affiliation:School of Locomotive and Rolling Stock Engineering, Dalian Jiaotong University, Dalian 116028, China
Abstract:An active learning BR-BP neural network combining Monte Carlo simulation method is proposed to improve the accuracy and efficiency of fatigue reliability analysis of a bogie frame.In the method, the prediction accuracy and convergence speed of the neural network are improved by using Bayesian Regularization training algorithm.Considering the inherent characteristics of reliability analysis, an active learning function that can be suitable for BP neural network is constructed to guide the selection of optimal sample points.Through the learning function, these samples are guaranteed to be distributed near the limit state function, and consider the contribution of prediction error and the distribution of sample points to the failure probability.The reliability analysis results of the bogie frame show that the prediction accuracy and calculation efficiency can be satisfied at the same time, which verifies the superiority and feasibility of the proposed method.
Keywords:structural reliability  BP neural network  sequence sampling  bogie frame  fatigue reliability analysis
点击此处可从《计算力学学报》浏览原始摘要信息
点击此处可从《计算力学学报》下载全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司    京ICP备09084417号-23

京公网安备 11010802026262号