EU Regional School - De Lorenzis Seminar
Prof. Dr. Laura De Lorenzis - Phase-Field Modeling and Computation of Fracture
Institute of Applied Mechanics, Technische Universität Braunschweig, Germany
The phase-field modeling approach to fracture, after the pioneering investigations of the early 2000 in the mathematics community, has recently attracted a tremendous interest also in the engineering community due to its theoretical soundness, computational flexibility and demonstrated predictive power. In the talk, we review the basic ingredients of the classical phase-field model of brittle fracture and highlight the consequences of different modeling choices on the predicted behavior. We then illustrate the basic features of phase-field models for ductile fracture and finally overview the latest extensions and applications of the framework.
SSD - Berre Seminar
Prof. Dr. Inga Berre - Three-Dimensional Numerical Modelling of Hydraulic Stimulation of Geothermal Reservoirs: Permeability Enhancement and Induced Seismicity
Department of Mathematics, University of Bergen, Norway
Understanding the controlling mechanisms underlying injection‐induced seismicity is important for optimizing reservoir productivity and addressing seismicity‐related concerns related to hydraulic stimulation in Enhanced Geothermal Systems as well as other sub-surface engineering applications. Hydraulic stimulation enhances permeability through elevated pressures, which cause normal deformations and the shear slip of preexisting fractures.
The process involves strongly coupled physical processes, involving reactivation and deformation of fractures, deformation of surrounding rock, and fluid flow in the fractures and their surroundings. The talk presents an approach for modelling of the governing flow and mechanics, where fractures are modelled as surfaces with associated apertures in a three-dimensional domain. Considering both flow and deformation, processes in the fractures are coupled with processes in the surrounding rock. While flow is assumed to be governed by Darcy’s law both in the fractures and the matrix, the model for deformation is inherently different for the fractured and non-fractured parts of the domain. Fracture reactivation is based on a Mohr-Coulomb criterion, and the corresponding irreversible deformation is based on an empirical model for friction.
Furthermore, fractures may continuously deform in the normal direction according to a non-linear model accounting for the normal loading. For the rock surrounding the matrix, we assume a continuous elastic deformation. Numerical results are presented to show how the methodology can be applied to understand important mechanisms affecting permeability and induced seismicity. In particular, we show how normal closure of fractures enhances pressure propagation away from the injection region and significantly increases the potential for postinjection seismicity.
SSD - Kollmannsberger Seminar
Dr. Stefan Kollmannsberger - Simulation in Additive Manufacturing with Modern Discretizational Techniques
Chair of Computation in Engineer, Technical University of Munich, Germany
The talk will present a general framework for the simulation of (initial) boundary value problems which may be defined on almost any type of geometric model. This framework is the finite cell method, a high order embedded domain method which the presenter has helped to develop in the recent decade.Geometric models which form the basis of computational mechanics often stem from Computer Aided Design (CAD). Two variants dominate in this setting: Constructive Solid Geometry (CSG) and boundary representation (B-Rep). The usual CAD to computational analysis process requires the generation of boundary conforming meshes. These, in turn, require the geometric models to be valid, i.e. water tight and flawless. To the contrary, industrial models are often flawed such that model healing must be carried out before boundary conforming meshes can be generated. This presentation will demonstrate how such a potentially expensive healing step may be avoided and how it is possible to directly compute on geometrically and topologically flawed models.Another type of geometric model are voxel models. They usually stem from computed tomography and are omnipresent, for example in medical applications. Yet, for example for the computation of implants, it is advantageous to augment voxel models by B-Rep models using CSG operations. The talk will discuss how computational analysis is possible on these combined models within the presented framework.For some applications, as for example in the computational analysis of historic structures, neither CAD nor voxel models are available. Moreover, the construction of an accurate, reverse-engineered CAD model is extremely complex and only possible in a very limited number of cases. As a remedy, this talk will present a new paradigm: to use pictures directly as geometric models for computational mechanics. The talk will close demonstrating the computational analysis of complex-shaped large historic structures from drone images.
EU Regional School - Hoefler Seminar
Prof. Dr. Torsten Hoefler - MPI Remote Memory Access Programming and Scientific Benchmarking of Parallel Codes
We will provide an overview of advanced MPI programming techniques. Specifically, we will focus on MPI-3's new Remote Memory Access (RMA) programming and an implementation thereof. We will discuss how to utilize MPI-3 RMA in modern applications. Furthermore, we will discuss issues in large-scale implementation and deployment. The lecture will then continue to a small number of other advanced MPI usage scenarios that every scientific computing researcher should know. Finally, we will discuss how to benchmark parallel applications in a scientifically rigorous way. This turns out to be surprisingly difficult and the state of the art is suboptimal. We will present twelve simple rules that can be used as guidelines for good scientific practice when it comes to measuring and reporting performance results.
SSD - Hoefler Seminar
Prof. Dr. Torsten Hoefler - High-Performance Communication in Machine Learning
One of the main drivers behind the rapid recent advances in machine learning has been the availability of efficient system support.
Despite existing progress, scaling compute-intensive machine learning workloads to a large number of compute nodes is still a challenging task. In this talk, we provide an overview of communication aspects in deep learning. We address the communication challenge, by proposing SparCML, a general, scalable communication layer for machine learning applications. SparCML is built on the observation that many distributed machine learning algorithms either have naturally sparse communication patterns, or have updates which can be sparsified in a structured way for improved performance, without loss of convergence or accuracy. To exploit this insight, we analyze, design, and implement a set of communication-efficient protocols for sparse input data, in conjunction with efficient machine learning algorithms which can leverage these primitives. Our communication protocols generalize standard collective operations, by allowing processes to contribute sparse input data vectors, of heterogeneous sizes. Our generic communication layer is enriched with additional features, such as support for non-blocking
(asynchronous) operations and support for low-precision data representations. We validate our algorithmic results experimentally on a range of large-scale machine learning applications and target architectures, showing that we can leverage sparsity for order-of-magnitude runtime savings, compared to existing methods and frameworks.