This paper describes how parallel processing may improve the computational efficiency of the Sub-domain Generation Method for the decomposition of finite element domains. This mesh partitioning approach uses a genetic algorithm-based optimisation and a neural network-based predictive module. The genetic algorithm is applied recursively to a coarse background mesh using a neural network predictor to estimate the number of elements generated after adaptive remeshing of the coarse mesh. Two alternative schemes for the parallelisation of this procedure are described. The first, is a more simple approach to parallelisation and represents the parallelisation of the genetic algorithm itself. This is achieved by the concept of having separate sub-populations on each processor. The second scheme is concerned with rec. cursively using the processors to split their sub-domains in two using a local SGM procedure and then passing one of the resulting partitions to an idle processor. The virtue of this scheme is to deliver ready to analyse partitions to every processor immediately following the sub-domain generation. This saves the expense of distributing the sub-domain information to the processing units where they will be analysed. These methods are discussed. The implementation of the second scheme is described in detail and its efficiency reviewed.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £78 +P&P)