New Method of Experimentation and Constitutive Modelling Using Neural Networks

本研究的目的是：1)通过MRI（磁共振成像）对适当质量的颗粒材料的内部和行为进行无创可视化，2)验证通过DEM（离散元法）模拟获得的微观观察结果。现有的椭球随机阵列宏观DEM结果是很有前景的。它们为大范围的应变、排水和不排水加载、静态和循环加载提供了定性正确的结果和一些良好的定量预测。然而，需要三维观测来验证。随着MRI技术的发展，本研究是可行的。研究结果将为颗粒土的本构关系建模提供一个一般性的理论框架。本研究的目的是：1)通过MRI（磁共振成像）对合适质量的颗粒材料的内部和行为进行无创可视化，2)验证DEM（离散元法）模拟获得的微观观察结果。现有的椭球随机阵列宏观DEM结果是很有前景的。它们为大范围的应变、排水和不排水加载、静态和循环加载提供了定性正确的结果和一些良好的定量预测。然而，需要三维观测来验证。随着MRI技术的发展，本研究是可行的。研究结果将为颗粒土的本构关系建模提供一个一般性的理论框架。*** 9503462 Ghaboussi将开发一种新的方法，直接从土试验结果确定土的本构模型。主要目标是发展和实验验证这种测定方法，并将其结果应用于根据现场试验和新型实验室试验的结果确定土壤的本构特性。在开发适当的计算方面，利用神经网络来模拟行为，并使用有限元法，进行一系列的实验来模拟现场试验。与传统的土壤试验不同，在提出的方法中，必须使样品处于非均匀应变状态，以便样品内的不同点可以承受尽可能多的不同应力路径。将开发新的土壤试验，包括一种相对简单的打孔试验，并在实验室进行试验。这些结果将用于训练神经网络。方法和计算方面的发展将在伊利诺伊大学进行。拟议研究的实验部分将在约翰霍普金斯大学进行。*** 9503697 Lade将开发一种新的方法，直接从土试验结果确定土的本构模型。主要目标是发展和实验验证这种测定方法，并将其结果应用于根据现场试验和新型实验室试验的结果确定土壤的本构特性。在开发适当的计算方面，利用神经网络来模拟行为，并使用有限元法，进行一系列的实验来模拟现场试验。与传统的土壤试验不同，在提出的方法中，必须使样品处于非均匀应变状态，以便样品内的不同点可以承受尽可能多的不同应力路径。将开发新的土壤试验，包括一种相对简单的打孔试验，并在实验室进行试验。这些结果将用于训练神经网络。方法和计算方面的发展将在伊利诺伊大学进行。拟议研究拱门的实验部分将在约翰霍普金斯大学进行。* * *

9505338 Ng The objective of this research is: 1) to non-invasively visualize the interior and the behavior of a suitable mass of granular material by MRI (Magnetic Resonance Imaging), and 2) to verify the microscopic observations obtained by DEM (Discrete Element Method) simulations. Available macroscopic DEM results on random arrays of ellipsoids produced by others are very promising. They provide qualitatively correct results and some good quantitative predictions for a wide range of strains, drained and undrained loading, and static as well as cyclic loading. However, 3-D observations are needed for verification. This research is feasible with the availability of MRI. The results will provide a general theoretical framework for modeling the constitutive relations of granular soils. *** 9505340 Nakagawa The objective of this research is: 1) to non-invasively visualize the interior and the behavior of a suitable mass of granular material by MRI (Magnetic Resonance Imaging), and 2) to verify the microscopic observations obtained by DEM (Discrete Element Method) simulations. Available macroscopic DEM results on random arrays of ellipsoids produced by others are very promising. They provide qualitatively correct results and some good quantitative predictions for a wide range of strains, drained and undrained loading, and static as well as cyclic loading. However, 3-D observations are needed for verification. This research is feasible with the availability of MRI. The results will provide a general theoretical framework for modeling the constitutive relations of granular soils. *** 9503462 Ghaboussi A new method will be developed to determine constitutive models of soils directly from the results of soil tests. The primary objective is to develop and experimentally verify this method of determination, with the results applied to the determination of the constitutive properties of soils from the results of in situ tests and n ew types of laboratory tests. After developing appropriate computational aspects utilizing neural networks to model behavior and using the FEM, a series of experiments will be conducted to simulate the in situ tests. Unlike conventional soils tests, in the proposed method it is essential that the sample be subjected to a non uniform state of strain so that different points within the sample can be subjected to as many different stress paths as possible. New soils tests, including a relatively simple punched test, will be developed and tested in the laboratory. These results will be used to train the neural network. The development of methodology and the computational aspects will be conducted at the University of Illinois. The experimental part of the proposed research will be conducted at the Johns Hopkins University. *** 9503697 Lade A new method will be developed to determine constitutive models of soils directly from the results of soil tests. The primary objective is to develop and experimentally verify this method of determination, with the results applied to the determination of the constitutive properties of soils from the results of in situ tests and new types of laboratory tests. After developing appropriate computational aspects utilizing neural networks to model behavior and using the FEM, a series of experiments will be conducted to simulate the in situ tests. Unlike conventional soils tests, in the proposed method it is essential that the sample be subjected to a non uniform state of strain so that different points within the sample can be subjected to as many different stress paths as possible. New soils tests, including a relatively simple punched test, will be developed and tested in the laboratory. These results will be used to train the neural network. The development of methodology and the computational aspects will be conducted at the University of Illinois. The experimental part of the proposed rese arch will be conducted at the Johns Hopkins University. ***