DL-SCALE: a novel deep learning-based model order upscaling scheme for solving topology optimization problems
- PDF / 1,752,445 Bytes
- 20 Pages / 595.276 x 790.866 pts Page_size
- 105 Downloads / 206 Views
(0123456789().,-volV)(0123456789(). ,- volV)
ORIGINAL ARTICLE
DL-SCALE: a novel deep learning-based model order upscaling scheme for solving topology optimization problems Nikos Ath. Kallioras1 • Nikos D. Lagaros1 Received: 27 March 2020 / Accepted: 26 October 2020 Ó Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract The main scope of this study is to propose a novel methodology aiming at enhancing the computational efficiency of the approaches used for solving structural topology optimization (STO) problems. The methodology is based on machine learning combined with the idea of using multiple finite element (FE) models of reduced order. The capability of deep belief networks (DBNs) in discovering multiple representational levels of data nonlinearity in pattern recognition problems recently triggered the development of the DLTOP methodology by the authors Kallioras et al. (Struct Multidiscip Optim, 2020, https://doi.org/10.1007/s00158-020-02545-z), that is based on DBNs and the solid isotropic material with penalization (SIMP) approach. In this study, a FE model order upgrading scheme integrated with the DLTOP methodology is proposed for accelerating further the SIMP-based solution procedure of the STO problems with no scalability limitations, labeled as DL-SCALE. The framework of DL-SCALE is based on a combined implementation of DBNs and SIMP into a sequentially implemented ‘‘model-optimize-and-order-upgrade’’ scheme. DL-SCALE efficiency is validated over several benchmark topology optimization test-examples. The results obtained for the test-examples clearly prove its computational advantages; the computing time is reduced by almost one order of magnitude while the corresponding reduction in terms of iterations is more than one order of magnitude compared to the ones originally required by SIMP, without any loss with respect to objective function value. It is also concluded from the results obtained that the proposed methodology can escalate to various finite element mesh discretizations, while optimized layout information transfer is possible, contributing also in accelerating further the STO procedure. Keywords Topology optimization Order upgrading Deep learning Computational efficiency SIMP approach Deep belief networks
1 Introduction Structural optimization has been a key research topic in engineering for several decades resulting to the development of various methodologies and formulations for dealing with such problems [2–7]. Applied structural optimization can also be considered as an added value to structural engineering design, since the advantages of
& Nikos D. Lagaros [email protected] 1
Institute of Structural Analysis and Antiseismic Research, Department of Structural Engineering, School of Civil Engineering, National Technical University of Athens, 9, Heroon Polytechniou Str., Zografou Campus, GR-15780 Athens, Greece
performing structural optimization are rather significant in terms of performance, cost, performance-to-cost ratio and of course environmental re
Data Loading...
