Topology optimization is a branch of structural optimization which solves an optimal material distribution problem. The resulting structural topology, for a given set of boundary conditions and constraints, has an optimal performance (e.g. minimum compliance). Conventional 3D topology optimization algorithms achieve quality optimized results; however, it is an extremely computationally intensive task which is, in general, impractical and computationally unachievable for real-world structural optimal design processes. Therefore, the current development of rapid topology optimization technology is experiencing a major drawback. To address the issues, a new approach is presented to utilize the powerful abilities of large deep learning models to replicate this design process for 3D structures. Adversarial models, primarily Wasserstein Generative Adversarial Networks (WGAN), are constructed which consist of 2 deep convolutional neural networks (CNN) namely, a discriminator and a generator. A minimax game is conducted between the generator and the discriminator as part of training where the discriminator maximizes the loss function whereas the generator tries to minimize the loss function of the model. Once trained, the generator from GAN can produce 3D structures in a computationally inexpensive process instantaneously. The corresponding input variables of the new generated structures are evaluated using a trained convolutional neural network. The dataset needed for training is generated using the traditional 3D topology optimization algorithms. Results from the GANs are validated by comparing these optimal structures against the 3D structures generated from the traditional algorithms with the same design settings. The potential issues and future extension of this work are discussed in detail in the article. As illustrated, introducing deep learning into the field of design will remarkably reduce the work time of an iterative design process.


    Zugriff

    Zugriff prüfen

    Verfügbarkeit in meiner Bibliothek prüfen

    Bestellung bei Subito €


    Exportieren, teilen und zitieren



    Titel :

    Application of Adversarial Networks for 3D Structural Topology Optimization


    Weitere Titelangaben:

    Sae Technical Papers


    Beteiligte:
    Shen, MH Herman (Autor:in) / Rawat, Sharad (Autor:in)

    Kongress:

    WCX SAE World Congress Experience ; 2019



    Erscheinungsdatum :

    2019-04-02




    Medientyp :

    Aufsatz (Konferenz)


    Format :

    Print


    Sprache :

    Englisch




    Multi-Head Self-Attention Generative Adversarial Networks for Multiphysics Topology Optimization

    Parrott, Corey M. / Abueidda, Diab W. / James, Kai A. | AIAA | 2022


    Structural topology optimization for crashworthiness

    Soto,C.A. / Ford Motor,US | Kraftfahrwesen | 2004


    Modern Structural Optimization Concepts Applied to Topology Optimization

    Leiva, J. / Watson, B. / Kosaka, I. | British Library Conference Proceedings | 1999


    Modern structural optimization concepts applied to topology optimization

    Leiva, Juan / Watson, Brian / Kosaka, Iku | AIAA | 1999


    Multidisciplinary Structural Truss Topology Optimization for Reconfigurability

    Nadir, William / Kim, Il Yong / Hauser, Dominic et al. | AIAA | 2004