不同应力路径下砂土的神经网络弹塑性本构模型研究

基于多种应力路径下砂土的加载、卸载和再加载过程的三轴排水试验数据，研究了三种应力路径对屈服面的影响，其结果表明，不同应力路径对砂土的本构模型有着不可忽视的作用。在此基础上，建立了砂土的弹塑性神经网络模型，对相应路径下的本构关系进行了学习。实例分析表明，该模型能够很好地描述具体应力路径下的本构关系，神经网络建模方法具有方便、容错性强的特点，对岩土力学快速、高效数值方法的进一步发展具有重要的参考价值。同时，在这种弹塑性应力-应变关系中，笔者并没有使用塑性势函数，这在目前的弹塑性本构关系的建模中是一种新的尝试。

Study of elastoplastic neural network constitutive models for sand under different stress paths

Based on the triaxial experiments data of sand on loading, unloading and reloading under different stress paths, the effects of three kinds of stress paths on the yielding surfaces have been studied. It is demonstrated that the effect of stress paths on the constitutive models of sand are obvious and cannot be neglected. On that basis, the elastoplastic neural network constitutive model of sand is established. It is shown that this model can approach the constitutive relations according to certain stress path correctly, it indicates that, the neural network method is convenient and has better fault-tolerance. In short, these methods are valuable for developing the numerical soil mechanics. At the same time, we abandon the theory of plastic potential function when setting up the elastoplastic stress-strain relations, which is a real new attempt in establishing the elastoplastic constitutive model.