Font Size: a A A

Mesh Partition In Parallel Finite Element Computing And Its Application To Electric Field Of Aluminum Reduction Cell

Posted on:2015-09-08Degree:MasterType:Thesis
Country:ChinaCandidate:C W SunFull Text:PDF
GTID:2181330434453139Subject:Control Science and Engineering
Abstract/Summary:PDF Full Text Request
Parallel finite element computing can satisfy the speed of computing in huge complicated structure, so it’s definitely a promising direction for numerical simulation in huge complicated structure. Mesh partition is the key part in parallel finite element computing, and it plays a very important role in the efficiency of parallel computing. This paper focused on the mesh partition in parallel finite element computing, then designed an easily realizable, widely applicable and well functional mesh partition method of finite element, and further discussed its application in Aluminum Reduction Cell simulation.First of all, the paper introduced the rationale and process of finite element analyzing. Besides, it showed a hardware platform, as well as software designing, of finite element parallel computing. Secondly, to deal with the mesh partition in parallel finite element computing, it gave the mathematical description of mesh partition in three-divisional finite element model. Then considering that it’s not accurate to use cutting criterion to evaluate communication traffic of large-scale model, we use the total node number of partition result as communication criterion, and defined equilibrium coefficient δ and communication coefficient ε as evaluation index, because they can quantitatively reflect the effect of mesh partition on parallel computing and set a criterion for quantitatively analyzing mesh partition. Thirdly, in this paper we reviewed the Aluminum Reduction Cell simulation and showed the process of finite element analyzing. On the basis of engineering practice, it realized the common ANP (Al-NASRA and NGUYEN) algorithm and multilevel partition, and applied huge Aluminum Reduction Cell model. Both the qualitative and quantitative compared analysis proved that ANP algorithm was better than multilevel partition. When the results of mesh partition algorithms was brought into parallel finite element computing of Aluminum Reduction Cell, the short analyzing time proved that quantitative evaluation criterion was effective. Finally, in response to the time-consuming problem of ANP algorithm on the large-scale model partition, algorithm analysis was conducted to confirm the slow node--addition disadvantage which existed in the sub-domain decision making stage. Multipoint ANP algorithm was therefore proposed. In order to optimize the algorithm, the paper presented a dynamic self-adaptive ANP algorithm. An algorithm which self-adaptively determines the number of nodes and dynamically adds them on the basis of model scale and sub-domain partition stage, enables balanced speed precision of algorithm and expands its applied range in large and even ultra-large finite element simulation. The application in Aluminum Reduction Cell simulation verified the validity of the raised algorithm and the computational efficiency of the mesh partition was apparently superior to that of traditional ANP algorithm. Besides, dynamic self-adaptive ANP algorithm was far more functional than multipoint ANP algorithm with larger finite element model.35figures,10tables,69references.
Keywords/Search Tags:finite element analysis, parallel computing, mesh partition, Aluminum Reduction Cell, dynamic conditional ANP algorithm
PDF Full Text Request
Related items