High-performance numerical methods for modelling of granular flows and sediment dynamics

颗粒材料是由宏观的小颗粒组成的，泥沙材料是其中一个重要的例子。这些材料在自然界中无处不在，是工业中第二大被操纵的材料（水是第一）。颗粒物料的流动在工程、地球物理和环境过程中起着至关重要的作用。在当今科学进步的世界里，这种最熟悉的物质形式的流动在很大程度上仍然是不可预测的，这似乎令人困惑。这种知识差距源于这些材料在不同情况下的复杂力学行为，可能类似于固体、液体（非牛顿流体）甚至气体。当颗粒材料与水等环境流体相互作用时，情况就更加复杂了。预测这些所谓的多相颗粒流的行为对于进一步了解今天有限的河流和海岸沉积物动力学、海底滑坡或采矿作业尾矿池中的泥浆流动至关重要。随着计算能力和数值算法的进步，对颗粒流动系统进行数值模拟已经成为可能，特别是在物理模型受到限制的情况下。然而，处理多相颗粒流的复杂性仍然超出了许多现有数值方法的能力。这是由于颗粒材料的复杂行为以及存在于环境流体和颗粒材料界面上的大变形和破碎。此外，为了处理多尺度问题的深入分析，需要集群的“pb级”计算。革命性数值技术的发展，无网格拉格朗日（粒子）方法，为我们提供了第一次克服颗粒流动复杂性的机会。已知这些方法能够处理具有复杂边界和界面的多相连续体。因此，该计划旨在(1)阐述多相颗粒流的理论基础，描述多相颗粒流的力学，并开发新的算法，主要基于无网格拉格朗日方法，用于数值实现；(2)提高数值技术的鲁棒性和准确性；(3)开发大规模并行、精确和多尺度的算法，能够对这些流系统进行PetaFLOP的计算。重点将放在模型的开发上，这些模型允许精确地表示粒度运动，然后利用现代计算机的全部能力来实现大规模问题的可扩展性能。该项目还旨在(4)对现实生活中多相颗粒流的机制提供新的理解，特别是在河流环境、采矿尾矿浆和滑坡中的泥沙动力学分析。

Granular materials are made up of macroscopic small particles, of which sediment material is an important example. These materials are ubiquitous in nature and are the second-most manipulated material in industry (water being the first). Flow of granular materials plays a critical role in engineering, geophysical and environmental processes. It may seem confounding that in the today's world of scientific advancements, the flow of this most familiar form of matter remains largely unpredictable. This knowledge gap stems from the complex mechanical behaviour of these materials which may resemble those of solid, liquid (a non-Newtonian fluid) or even gas in different circumstances. The situation is still more complex when the granular material interacts with an ambient fluid like water. Predicting the behaviour of these so-called multiphase granular flows is critical to furthering today's limited understanding of fluvial and coastal sediment dynamics, submarine landslides, or slurry flow in tailing ponds of mining operations. With advances in computing power and numerical algorithms, it has become possible to numerically simulate granular flow systems, especially where physical models are restricted. Nevertheless, dealing with the complexities of multiphase granular flows is still beyond the capabilities of the many existing numerical methods. This is due to the complicated behaviour of granular material and the large deformations and fragmentations that exist at the interface of the ambient fluid and the granular material. Furthermore, to deal with the in-depth analysis of multi-scale problems, the cluster “peta-scale” computing is required. The development of a revolutionary generation of numerical techniques, the mesh-free Lagrangian (particle) methods, has provided us the first ever opportunity to overcome the granular flows complexities. These methods are known to be capable of handling the multiphase continuum with complex boundaries and interfaces. The proposed program, therefore, aims to (1) elaborate the theoretical foundation, describing the mechanics of multiphase granular flows, and develop novel algorithms, primarily based on the mesh-free Lagrangian methods, for numerical implementations; (2) improve the robustness and accuracy of these numerical techniques; and (3) develop massively parallel, accurate, and multi-scale algorithms, capable of PetaFLOP computation of these flow systems. The focus will be on development of models that permit accurate representation of the grain-scale motions and then harnessing the full power of modern computers to achieve scalable performance on large-scale problems. This program also aims to (4) provide new understanding of mechanisms involved in real-life multiphase granular flows, particularly for the case of sediment dynamics analysis in fluvial environments, mining tailing slurries and landslides.