基于应变信号时频分析与 CNN网络的车辆荷载识别方法

针对现有神经网络车辆荷载识别方法的识别精度不足且训练样本采集困难的问题,提出了一种基于应变信号时频分析与CNN网络的车辆荷载识别方法,对移动车辆总重进行荷载识别.首先,利用连续小波时频变换方法处理桥梁跨中应变信号,得到应变信号的时频特征,并利用双线性插值算法将时频信号矩阵变为大小为64×64的数值矩阵,作为CNN网络的输入数据;其次,利用CNN网络的回归学习算法,在训练少量数值矩阵后直接建立应变响应与车辆荷载的映射关系,从而实现对未知车辆荷载的识别;最后,通过模拟试验发现虽然在不同路面粗糙度和噪声影响下,CNN网络的荷载识别结果会受到不同程度的影响,但在一定范围内的路面粗糙度和噪声影响下仍然能较精确地识别车辆荷载.

Vehicle Load Identification Method Based on Time Frequency Analysis of Strain Signal and Convolutional Neural Network

Aiming at the problems of insufficient identification accuracy and difficulty in collecting training samples in existing vehicle load identification method based on neural network, a vehicle load identification method based on time-frequency analysis of strain signal and Convolutional Neural Network (CNN) is proposed to identify the total weight of mobile vehicles. Firstly, the time-frequency characteristics of the strain signal are obtained by using the wavelet time-frequency transform, and the time-frequency matrix is changed into a 64×64 numerical matrix as the input data of CNN. Secondly, in order to realize the purpose of unknown vehicle load identification, the map? ping relationship between strain response and vehicle load is directly established after training a small number of nu? merical matrices by using the regression learning algorithm of CNN. Finally, through simulation tests, it is found that although the load recognition results of the CNN are affected to varying degrees under the influence of different road roughness and noise, the vehicle load can still be more accurately identified under the influence of certain road rough? ness and noise.