中密砂的弹塑性神经网络本构模型

Elasto-plastic neural network constitutive model of medium sand

基于建立岩土本构模型的数值方法和建模的4个基本步骤,建立了中密砂的弹塑性神经网络本构模型,绘出了屈服面轨迹,并通过试样分析验证了此模型的可行性与有效性.研究结果表明:一方面,该模型排除了确定塑性势函数的困扰;另一方面,用神经网络替代模型解析表达式,既提高了精度,又简化了确定参数的过程;同时,能够反映出应力路径对本构关系的影响,从而为工程实践中对不同应力路径进行数值模拟提供了可能.

Based on the numerical method in modeling the constitutive relations of rock and soil and its four basic steps, the elastoplastic neural network constitutive model of medium sand is set up, the yield surfaces are drawn from test data, and their feasibility and effectiveness are shown through the computed example. The research results show that the model has many advantages. On the one hand, the difficulty for determining the plastic potential in elastoplastic model can be overcome; on the other hand, the accuracy of model is enhanced and the process of determining parameters of model is simplified by using neural network instead of analytical expressions. Besides, the model can reflect the influence of stress paths, which makes it possible for numerical simulation of actual engineering under different stress paths.