【摘要】 神经网络被称为行为模型技术,与传统的现象模型和统计方法不同,它不需要对要模拟的物理现象做出任何假定,它的本质是给出输入空间到输出空间的映射。基于多种应力路径下的砂土的加荷、卸荷和再加荷过程的三轴排水试验数据,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逼近。实例分析表明,该模型对具体应力路径下的本构关系能够很好地逼近和预测。同时,将建立的神经网络模型纳入到试样的有限元程序中去。结果表明,神经网络建模方法比传统的建模方法更方便、准确且容错性强,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。更多还原

【Abstract】 Neural network are called a behavior-modeling technique. It represent a different approach from phenomenological-modeling and the traditional statistical methods, since they need not make any hypothesis about the physical characteristics of the phenomenon being modeled. In essence, they map from one space-input patterns to a space of output patterns. Based on the triaxial experiments of loading, unloading and reloading of sand under different stress paths, we establish the elasto-plastic neural network constitutive model for sand. It is shown that this model can approach the constitutive relations according to certain stress path correctly We introduced the established neural network model into the finite element program for a test sample. It has been demonstrated that the neural network method is more convenient, accurate and has better fault-tolerance than the traditional methods. In short, it is valuable for developing the numerical soil mechanics.更多还原