卸载应力路径下黄土强度的神经网络预测

Predicting strength of loess under unloading stress paths with neural networks

各种施工活动,将对应力状态与应力路径产生影响;在分析基坑开挖及隧道掘进等卸载作用对应力状态与应力路径影响的基础上,进行了不同应力路径的室内非常规三轴卸载试验,根据卸载试验结果,建立了以轴向与侧向卸(加)载增量比例、应力路径变化方向、卸(加)载前轴向与侧向初始固结应力为输入层参数、以轴向与侧向破坏应力为输出层参数的神经网络预测模型,并对卸载应力路径下黄土强度进行了神经网络预测及误差分析,其结果表明, 预测值与试验值较为接近。

Various construction activities result in the influences on stress states and stress paths. For example, the stress states and stress paths at sides and bottom of pit change during the excavation. In the process of shielding, earth surrounds the gaps of shield ends takes various unloading effect to alter the stress states and stress paths. There exist compressive disturbances in the earth right in the front of tunneling. The gaps between the end of shield and lining give rise to moving and loosened earth to form unloading disturbance areas. Based on the analysis of unloading effect in different types of construction, a series of laboratory tests for loess with non-regular triaxial shear under different unloading stress paths were carried out. The model for predicting loess strength under different stress paths with parameters (the ratio of stress increment of axial and confining unloading / loading, the change direction of stress paths and initial stresses of consolidation) at input layer and ones (ultimate axial and confining stresses) at output layer has been developed. The results show that the value of prediction with neural networks is in good agreement with that of unloading tests.