A New Algorithmic Approach For Solving Frictional Contact Problems Using An Optimization Procedure

Author : Matias Ignacio Rivero
Publisher :
ISBN 13 :
Total Pages : 172 pages
Book Rating : 4.:/5 (112 download)

DOWNLOAD NOW!

Download or read book A Parallel Algorithm for Deformable Contact Problems written by Matias Ignacio Rivero and published by . This book was released on 2018 with total page 172 pages. Available in PDF, EPUB and Kindle. Book excerpt: In the field of nonlinear computational solid mechanics, contact problems deal with the deformation of separate bodies which interact when they come in touch. Usually, these problems are formulated as constrained minimization problems which may be solved using optimization techniques such as penalty method, Lagrange multipliers, Augmented Lagrangian method, etc. This classical approach is based on node connectivities between the contacting bodies. These connectivities are created through the construction of contact elements introduced for the discretization of the contact interface, which incorporate the contact constraints in the global weak form. These methods are well known and widely used in the resolution of contact problems in engineering and science. As parallel computing platforms are nowadays widely available, solving large engineering problems on high performance computers is a real possibility for any engineer or researcher. Due to the memory and compute power that contact problems require and consume, they are good candidates for parallel computation. Industrial and scientific realistic contact problems involve different physical domains and a large number of degrees of freedom, so algorithms designed to run efficiently in high performance computers are needed. Nevertheless, the parallelization of the numerical solution methods that arises from the classical optimization techniques and discretization approaches presents some drawbacks which must be considered. Mainly, for general contact cases where sliding occurs, the introduction of contact elements requires the update of the mesh graph in a fixed number of time steps. From the point of view of the domain decomposition method for parallel resolution of numerical problems this is a major drawback due to its computational expensiveness, since dynamic repartitioning must be done to redistribute the updated mesh graph to the different processors. On the other hand, some of the optimization techniques modify dynamically the number of degrees of freedom in the problem, by introducing Lagrange multipliers as unknowns. In this work we introduce a Dirichlet-Neumann type parallel algorithm for the numerical solution of nonlinear frictional contact problems, putting a strong focus on its computational implementation. Among its main characteristics it can be highlighted that there is no need to update the mesh graph during the simulation, as no contact elements are used. Also, no additional degrees of freedom are introduced into the system, since no Lagrange multipliers are required. In this algorithm the bodies in contact are treated separately, in a segregated way. The coupling between the contacting bodies is performed through boundary conditions transfer at the contact zone. From a computational point of view, this feature allows to use a multi-code approach. Furthermore, the algorithm can be interpreted as a black-box method as it solves each body separately even with different computational codes. In addition, the contact algorithm proposed in this thesis can also be formulated as a general fixed-point solver for the solution of interface problems. This generalization gives us the theoretical basis to extrapolate and implement numerical techniques that were already developed and widely tested in the field of fluid-structure interaction (FSI) problems, especially those related to convergence ensurance and acceleration. We describe the parallel implementation of the proposed algorithm and analyze its parallel behaviour and performance in both validation and realistic test cases executed in HPC machines using several processors.