Study of Ultra High Temperature Fracture of Functionally Graded Material by Neural Network

复合材料具有复杂的微观材料组成，材料的不均匀性在界面处产生高应力集中。因此，在微机械区域中可能发生非线性现象，例如塑性流动、分层、开裂、损伤等。当夹杂物的体积分数较高时，夹杂物之间的相互作用促进了这种现象的发生，这将反映到复合材料的宏观力学性能上。在这项研究中，目标是深入调查的细观力学现象，最可靠的本构模型是开发的复合材料，包括功能梯度材料（FGM）的非线性加载范围。本文的工作主要集中在三个方面：用有限元法和神经网络方法进行细观力学分析;均匀化方法在复合材料细观领域的应用; ...更多信息 f本构关系。(1)开发了一种神经网络程序，用于计算建立非线性复合材料的本构关系。采用有限元法模拟了颗粒弥散复合材料的拉伸变形，得到了以温度和体积分数为参数的离散非线性拉伸数据。将数据送入神经网络，以优化复合材料热机械本构行为的网络参数。虽然这种方法需要准备的指导性数据，它提供了一个快速的过程，即使是非线性复合材料的变化的体积分数，如在FGM.An应用程序进行了热冲击FGM板的断裂评估与边缘裂纹，动态传播的梯度介质。该程序使用模糊推理技术在中间级平均神经推导的陶瓷基体和金属基体的材料特性。(2)均匀化方法是最适合于计算描述的本构张量的任意组合模式。目前，本构关系的发展是试图为颗粒分散复合材料的基体局部经历塑性变形或损伤，这是由颗粒之间的应力相互作用引起的。这种方法的优点是在获得宏观力学特性的同时，还可以监测微观力学现象。然而，计算过程花费太多的时间，特别是对于非线性问题，对于实际应用。(3)发展上述两种方法的原因是为了克服非线性复合材料本构模型难以完全理论化的困难。然而，试图建立一个本构模型的复合材料与椭球夹杂物。现有的模型大多采用Mori-Tanaka的平均场理论，但该模型的适用范围仅限于基体与夹杂物的材料刚度比较窄的范围和较低的体积分数。另一方面，本模型是有效的，并具有良好的精度在整个范围内的刚度比和体积分数。此外，所开发的模型扩展到弹塑性政权通过使用随机方法的分布塑性。将该模型应用于功能梯度材料的热冲击断裂问题。最后将混合律应用于相变问题，说明了混合律在形状记忆合金本构模型和塑性诱发相变现象中的重要性。少

Composite material has a complicated microscopic material constituent, where the material inhomogeneity generates a high stress concentration at the interface. Nonlinear phenomena, therefore, such as plastic flow, delamination, cracking, damage, etc. might occur in the micromechanical area. When the volume fraction of the inclusions is high, the interaction between the inclusions promotes the phenomena, which would reflect to the macromechanical performance of the composite. In this research, the objective is set to the thorough investigation of the micromechanical phenomenon, upon which the most reliable constitutive mode is developed over the nonlinear loading range of composite materials including a functionally graded material (FGM). The effort of this work is focused on the three subjects : a mesomechanics analysis by a finite element method and neural network generalization ; application of homogenization method to the meso-domain of the composite, and a theoretical development o … More f the constitutive law by using equivalent inclusion method.(1) A neural network procedure was developed to computationally build a constitutive law for nonlinear composite materials. A particle dispersed composite pattern was modeled by the finite element method, generating discrete nonlinear tensile data with a temperature and a volume fraction as parameters. The data were fed to the neural network to optimize the network parameters for the thermo-mechanical constitutive behavior of the composite. Although this method requires the preparation of the instructive data, it provides a fast process for nonlinear composites even for the varying volume fraction such as in the FGM.An application was made to the fracture evaluation of the thermally shocked FGM plate with an edge crack that dynamically propagates in the graded medium. The procedure uses fuzzy inference technique at the intermediate grade to average the neuro-deduced material characteristics of the ceramics base and the metal base.(2) The homogenization method is most suited to computationally describe the constitutive tensor for arbitrary composite patterns. In the present, the development of the constitutive law was attempted for the particle dispersed composite where the matrix locally undergoes a plastic deformation or a damage, which is caused by the stress interaction between particles. An advatage of this method is in that the micromechanical phenomena can be monitored as well as the macromechanical characteristic is obtained. However, the computational process takes too much time, especially for the nonlinear problem, for practical use.(3) The reason for the development of the above two method is to circumvent the difficulty of totally theoretical development of the constitutive model for the nonlinear composite. However, an attempt was made to build a constitutive model for a composite with ellipsoidal inclusions. The most of the existing models use Mori-Tanaka's mean field theory, but the applicability of the model is limited to a narrow band of the material rigidity ratio of the matrix and the inclusion, and to the low volume fraction. On the other hand, the present model is valid and has a good accuracy over the entire range of the rigidity ratio and the volume fraction. Also the developed model was expanded to the elastoplastic regime by using a stochastic approach to the distributed plastictiy. The model was applied to a thermal shock fracture problem of the functionally graded material. Another application was made to the phase transforming problem, and showed the significance of the mixing law in the constitutive modeling for the shape memory alloy and the plasticity induced phase transformation phenomena. Less