Next: Time Integration Up: Top Previous: Top


Introduction

The engineering community is facing growing demands for state of the art modelling and realistic predictions in many fields. Such demands lead to models of a very complex geometry with many degrees of freedom. In many cases, a nonlinear behaviour has to be taken into account, including complex nonlinear constitutive models and geometrically nonlinear effects.

These requests for such complex analyses initiate demands for large scale computing, which must be feasible from the view of both time and available resources. There are several types of available computer hardware architectures. The recent progress in computer technology allows practical use of parallel computers. The parallel computer architectures are based on shared-memory or massively parallel (multiple instruction and data) concepts. Also small design offices can profit from parallel technology. Connecting the office workstations together into a computer cluster can provide a powerful parallel machine. Despite the variety of available platforms, the message passing is a common communication tool, available on most hardware platforms. The parallel processing allows to obtain results in acceptable time by significantly speeding up the analysis. It also allows large and complex analyses, that often do not fit into single, even well equipped, machine with one processor unit, to be performed (regardless of achieved speedup).

Explicit integration schemes are currently widely accepted and used. They lead to extremely efficient algorithm, which can be parallelized in a straightforward way. The node-cut and element-cut techniques can be used to formulate efficient parallel algorithms. Realistic constitutive models for quasi-brittle materials (such as concrete or mortar) must properly capture the localized solution in tension regime. A very promising concept, recently discovered, is the concept of nonlocal averaging. This approach is based on the averaging of a certain suitable local quantity over characteristic volume, which is considered to be a material property. The nonlocal quantity is then substituted into a local constitutive relation. The recently developed models for the prediction of concrete cracking are able to properly represent localized solution and are less sensitive to mesh orientation than previously used models. A generalization of node-cut and element-cut techniques is necessary to efficiently use the nonlocal constitutive models.



Next: Time Integration Up: Top Previous: Top

Daniel Rypl
2005-12-03