人工神经网络在无缝线路应力检测中的研究  

Research on artificial neural networks in continuous welded rail stress test

作　　者：舒迪[1] 祁欣[1] 

机构地区：[1]北京化工大学信息科学与技术学院控制理论与控制工程系,北京100029

出　　处：《电子测量与仪器学报》2013年第6期515-520,共6页Journal of Electronic Measurement and Instrumentation

基　　金：国家自然科学基金(60971019;11004005)资助项目

摘　　要：人工神经网络通过借助数学和物理的方法从信息的角度对人脑神经系统进行模拟后建立数学简化模型,对复杂数学模型的非线性映射进行模拟,有相当高的精度。利用人工神经网络对铁路温度应力检测中的众多非线性参数进行人工神经网络建模,从而得出铁路温度应力的精确检测方法,并与X射线应力检测仪做对比实验,进行误差分析证明人工神经网络的可行性。分别对2个模型建模,第1个模型的特征向量选取的是巴克豪森噪声(MBN)信号峰-峰值、峰值、均方根、半峰宽、峰宽比。第2个模型的特征向量选取的是MBN信号峰-峰值、钢轨洛氏硬度、钢轨自有漏磁场梯度。通过对比实验以及误差分析判断各个应力相关因数对应力检测的贡献度,进一步提高检测精度。Simulating the human brain system from the perspective of information by means of mathematical and physical methods and establishing the mathematical simplified model to simulate the nonlinear mapping of complex mathematical models,artificial neural network has quite high accuracy.In this paper,artificial neural network is used for modeling some nonlinear parameters in the rail temperature stress detection,thereby a precise detection method of the rail temperature stress is gotten.And contrast experiment by X-ray stress detector proves the feasibility of artificial neural networks for error analysis.Two group parameters is modeled,the eigenvector of the first group is Barkhausen noise（MBN） peak to peak,peak,root mean square,half band width and peak width ratio,and the eigenvector of the second group is MBN peak to peak,rail hardness of the rail and the leakage magnetic field gradient.By comparing the experiment and analyzing the error to estimate the contribution of the various stress-related parameters in stress detecting,the detection accuracy is improved further.

关 键 词：人工神经网络 温度应力 巴克豪森 

分 类 号：U238[交通运输工程—道路与铁道工程]