For an engineer, getting the simulation results as soon as the finite element model is ready would be the perfect scenario. Today, crash analysis still needs large computation time. In the RADIOSS team we understand the need to speed up the analysis and as a result, we work on several methods such as the HPC scalability, Advanced Mass Scaling method (AMS) and Multi-Domains method.
The Multi-Domains method consists of splitting the model into several models, or domains, according to the minimum time step value and the number of elements. Each domain will be computed with its own and adequate time step. The total number of cycles needed for the computation will be reduced and elapse time will be saved.
The different domains will be computed sequentially on the cluster with the same number of cores. When the simulation time reached in a domain is bigger than the other one(s), output such as node velocity, forces are exchanged between domains through a master process named rad2rad.
The model domain creation can be created by the user or it be done automatically by RADIOSS. Keywords are set in the engine and the master process control file to define and manage the interfaces between the domains.
Applications like aircraft ditching or any structure impacting fluid domain has a real interest to use multi-domain. The time step for fluid elements are usually 10 times bigger than structural mesh and the fluid domain may have 10 times more elements. The boundary between the domains, structure and fluid, is easily defined with the fluid-Structure contact interface (FSI).
Dihedral shape impact on water
This method can also be used for automotive applications where some component needs to be meshed with fine shell or solid element in order to study the material failure. The part with small mesh size will be set in the subdomain. Because many connection and interface may exist between the refined component and the other ones, the automatic option to create the multi-domains model is really helpful.
From bumper beam impact with refined area
To be efficient, this method requires specific conditions about the number of elements in the different domains and the time step ratio. Simple equations defined in the documentation helps to compute the expected speed-up before creating the multi-domains model.
The multi-domain method can also benefit from the RADIOSS HPC scalability presented by Eric Lequiniou.