Finite element analysis & design optimization in a distributed computing environment

S. D. Rajan, Ashok D. Belegundu, D. Lee, Amol S. Damle, J. St Ville

    Research output: Chapter in Book/Report/Conference proceedingConference contribution

    2 Citations (Scopus)

    Abstract

    As product designs become more sophisticated, both the finite element (FE) and the design optimization (DO) models have grown bigger. At the same time there is increasing evidence that computing clusters created with commodity chips are capable of outperforming traditional supercomputers. In this paper, the HYI-3D design optimization software system is discussed. HYI-3D has been developed to work in sequential and distributed processing environments. The two major objectives are as follows - (a) implement the FE methodology using the well-known domain decomposition technique where the original FE model is split into a number of smaller subdomains, and (b) implement a design optimization methodology for sizing, shape and topology optimization using coarse and fine-grain parallelism. For the FE engine, a direct sparse solver is used at the subdomain level and preconditioned conjugate gradient (PCG) is used at the interface system level. The finite element equations are then generated and assembled at the individual domain level in parallel. Matrix and vector operations involving sparse matrices form the bulk of the computations in this step. Once these equations are assembled, the condensed system level equations are formed. These condensed system level equations are usually much smaller (but denser) than the original system equations, and hence can be computationally expensive. With respect to design optimization, multi-level parallelism is employed. Not only can the finite element analysis be carried out in parallel but also other steps in the design optimization algorithm can be computed in parallel - gradients, line search and direction-finding problem. Numerical examples show the gains obtained from coarse and fine grain parallelism.

    Original languageEnglish (US)
    Title of host publicationCollection of Technical Papers - 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference
    Pages1716-1726
    Number of pages11
    StatePublished - Dec 1 2004
    EventCollection of Technical Papers - 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference - Albany, NY, United States
    Duration: Aug 30 2004Sep 1 2004

    Publication series

    NameCollection of Technical Papers - 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference
    Volume3

    Other

    OtherCollection of Technical Papers - 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference
    CountryUnited States
    CityAlbany, NY
    Period8/30/049/1/04

    Fingerprint

    Distributed computer systems
    Finite element method
    Shape optimization
    Cluster computing
    Supercomputers
    Product design
    Design optimization
    Engines
    Decomposition
    Processing

    All Science Journal Classification (ASJC) codes

    • Engineering(all)

    Cite this

    Rajan, S. D., Belegundu, A. D., Lee, D., Damle, A. S., & St Ville, J. (2004). Finite element analysis & design optimization in a distributed computing environment. In Collection of Technical Papers - 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference (pp. 1716-1726). (Collection of Technical Papers - 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference; Vol. 3).
    Rajan, S. D. ; Belegundu, Ashok D. ; Lee, D. ; Damle, Amol S. ; St Ville, J. / Finite element analysis & design optimization in a distributed computing environment. Collection of Technical Papers - 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference. 2004. pp. 1716-1726 (Collection of Technical Papers - 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference).
    @inproceedings{38135b0a19174b75b18eae6dcc2a074f,
    title = "Finite element analysis & design optimization in a distributed computing environment",
    abstract = "As product designs become more sophisticated, both the finite element (FE) and the design optimization (DO) models have grown bigger. At the same time there is increasing evidence that computing clusters created with commodity chips are capable of outperforming traditional supercomputers. In this paper, the HYI-3D design optimization software system is discussed. HYI-3D has been developed to work in sequential and distributed processing environments. The two major objectives are as follows - (a) implement the FE methodology using the well-known domain decomposition technique where the original FE model is split into a number of smaller subdomains, and (b) implement a design optimization methodology for sizing, shape and topology optimization using coarse and fine-grain parallelism. For the FE engine, a direct sparse solver is used at the subdomain level and preconditioned conjugate gradient (PCG) is used at the interface system level. The finite element equations are then generated and assembled at the individual domain level in parallel. Matrix and vector operations involving sparse matrices form the bulk of the computations in this step. Once these equations are assembled, the condensed system level equations are formed. These condensed system level equations are usually much smaller (but denser) than the original system equations, and hence can be computationally expensive. With respect to design optimization, multi-level parallelism is employed. Not only can the finite element analysis be carried out in parallel but also other steps in the design optimization algorithm can be computed in parallel - gradients, line search and direction-finding problem. Numerical examples show the gains obtained from coarse and fine grain parallelism.",
    author = "Rajan, {S. D.} and Belegundu, {Ashok D.} and D. Lee and Damle, {Amol S.} and {St Ville}, J.",
    year = "2004",
    month = "12",
    day = "1",
    language = "English (US)",
    isbn = "1563477165",
    series = "Collection of Technical Papers - 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference",
    pages = "1716--1726",
    booktitle = "Collection of Technical Papers - 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference",

    }

    Rajan, SD, Belegundu, AD, Lee, D, Damle, AS & St Ville, J 2004, Finite element analysis & design optimization in a distributed computing environment. in Collection of Technical Papers - 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference. Collection of Technical Papers - 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference, vol. 3, pp. 1716-1726, Collection of Technical Papers - 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference, Albany, NY, United States, 8/30/04.

    Finite element analysis & design optimization in a distributed computing environment. / Rajan, S. D.; Belegundu, Ashok D.; Lee, D.; Damle, Amol S.; St Ville, J.

    Collection of Technical Papers - 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference. 2004. p. 1716-1726 (Collection of Technical Papers - 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference; Vol. 3).

    Research output: Chapter in Book/Report/Conference proceedingConference contribution

    TY - GEN

    T1 - Finite element analysis & design optimization in a distributed computing environment

    AU - Rajan, S. D.

    AU - Belegundu, Ashok D.

    AU - Lee, D.

    AU - Damle, Amol S.

    AU - St Ville, J.

    PY - 2004/12/1

    Y1 - 2004/12/1

    N2 - As product designs become more sophisticated, both the finite element (FE) and the design optimization (DO) models have grown bigger. At the same time there is increasing evidence that computing clusters created with commodity chips are capable of outperforming traditional supercomputers. In this paper, the HYI-3D design optimization software system is discussed. HYI-3D has been developed to work in sequential and distributed processing environments. The two major objectives are as follows - (a) implement the FE methodology using the well-known domain decomposition technique where the original FE model is split into a number of smaller subdomains, and (b) implement a design optimization methodology for sizing, shape and topology optimization using coarse and fine-grain parallelism. For the FE engine, a direct sparse solver is used at the subdomain level and preconditioned conjugate gradient (PCG) is used at the interface system level. The finite element equations are then generated and assembled at the individual domain level in parallel. Matrix and vector operations involving sparse matrices form the bulk of the computations in this step. Once these equations are assembled, the condensed system level equations are formed. These condensed system level equations are usually much smaller (but denser) than the original system equations, and hence can be computationally expensive. With respect to design optimization, multi-level parallelism is employed. Not only can the finite element analysis be carried out in parallel but also other steps in the design optimization algorithm can be computed in parallel - gradients, line search and direction-finding problem. Numerical examples show the gains obtained from coarse and fine grain parallelism.

    AB - As product designs become more sophisticated, both the finite element (FE) and the design optimization (DO) models have grown bigger. At the same time there is increasing evidence that computing clusters created with commodity chips are capable of outperforming traditional supercomputers. In this paper, the HYI-3D design optimization software system is discussed. HYI-3D has been developed to work in sequential and distributed processing environments. The two major objectives are as follows - (a) implement the FE methodology using the well-known domain decomposition technique where the original FE model is split into a number of smaller subdomains, and (b) implement a design optimization methodology for sizing, shape and topology optimization using coarse and fine-grain parallelism. For the FE engine, a direct sparse solver is used at the subdomain level and preconditioned conjugate gradient (PCG) is used at the interface system level. The finite element equations are then generated and assembled at the individual domain level in parallel. Matrix and vector operations involving sparse matrices form the bulk of the computations in this step. Once these equations are assembled, the condensed system level equations are formed. These condensed system level equations are usually much smaller (but denser) than the original system equations, and hence can be computationally expensive. With respect to design optimization, multi-level parallelism is employed. Not only can the finite element analysis be carried out in parallel but also other steps in the design optimization algorithm can be computed in parallel - gradients, line search and direction-finding problem. Numerical examples show the gains obtained from coarse and fine grain parallelism.

    UR - http://www.scopus.com/inward/record.url?scp=20344396865&partnerID=8YFLogxK

    UR - http://www.scopus.com/inward/citedby.url?scp=20344396865&partnerID=8YFLogxK

    M3 - Conference contribution

    AN - SCOPUS:20344396865

    SN - 1563477165

    SN - 9781563477164

    T3 - Collection of Technical Papers - 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference

    SP - 1716

    EP - 1726

    BT - Collection of Technical Papers - 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference

    ER -

    Rajan SD, Belegundu AD, Lee D, Damle AS, St Ville J. Finite element analysis & design optimization in a distributed computing environment. In Collection of Technical Papers - 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference. 2004. p. 1716-1726. (Collection of Technical Papers - 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference).