基于BP神经网络的场道脱空检测方法及实验

1)冻土工程国家重点实验室,甘肃兰州 730000; 2)中国民航大学机场学院,天津 300300

道路工程; 场道脱空; 小波变换; BP神经网络; 模型试验; 现场测试

Airport pavement void testing based on back propagation neural network
Liu Guoguang1,2, Wu Zhiwei2, Niu Fujun1, and Zhao Longfei2

1)State Key Laboratory of Frozen Soil Engineering, Lanzhou 730000, Gansu Province, P.R.China 2)Airport College, Civil Aviation University of China, Tianjin 300300, P.R.China

pavement engineering; pavement void; wavelet transform; back propagation neural network; model experiment; site test

DOI: 10.3724/SP.J.1249.2016.03309

备注

为研究场道脱空检测方法,进行室内模型试验,获得冲击荷载作用下道面板加速度响应时程曲线,利用Matlab小波变换工具箱提取加速度曲线特征值,分析脱空对振动信号的影响规律. 通过优化荷载级数、筛选输入向量,建立了场道脱空的BP(back propagation)神经网络预测方法. 为检验理论研究结果的正确性,利用重锤式弯沉仪在机场进行跑道脱空测试,通过场道取芯脱空观察评价BP神经网络预测结论的可靠性. 结果表明,荷载级数、输入向量、训练次数、训练强度和算法对BP神经网络预测准确性影响较大; 脱空影响下场道加速度信号可作为BP神经网络脱空预测的输入向量,取芯后场道脱空状况同BP神经网络预测结果一致.

In order to investigate the test method of airport pavement voids, a laboratory model test was conducted to achieve the time-history curve of acceleration response of airport pavement under impact loading. The characteristic value of the acceleration curve was obtained by the wavelet transform tool box of Matlab, by which the influence law of pavement voids on vibration signal was analyzed. By optimizing loading steps and screening input vectors, a pavement void prediction method based on back propagation neural network was established. To calibrate the prediction method, a runway field test was carried out by using heavy weight deflector, by which the reliability of back propagation neural network prediction was evaluated by plate coring. The results show that the loading steps, the input vector, training times, the training intensity and algorithm have significant influences on the prediction accuracy of the back propagation neural network. The pavement acceleration influenced by void could be used as the input vector of back propagation neural network, the prediction result of the back propagation neural network was proved by void observation of site coring.

·