An Interactive Structural Optimization of Space Frame Structures Using Machine Learning

  • Zeynep AksözEmail author
  • Clemens Preisinger
Conference paper


The conventional methods used for structural optimization are handled by iterative simulation and evaluation runs, that are developed specifically for each problem, these processes can become computationally expensive very quickly. This work describes an approach to optimization of free form space structures, using Finite Element Analysis and Machine Learning, to overcome long computation periods.

As Hajela and Berke also state, approaching a structural problem as a single optimization problem results in a large dimensionality problem that can be cumbersome to solve by heuristic methods that are commonly used for the optimization of complex problems. Subsampling the problem into smaller independent parts can ease the problem-solving process. Through the use of artificial neural networks, the solutions can be obtained in parallel [1].

Through their modularity, space frames can be decomposed into sub-structures that share the same topology, and the complex problem can be simplified into independent modules. A supervised learning algorithm (in this case Back Propagation) is trained on a set of optimized modules to execute an optimal geometry for each structural node in parallel on a given load.

The trained ANN can be applied to any spaceframe structure regardless of their topology, size or complexity. This strategy shifts the optimization procedure from being a problem-specific process to domain-specific process, where within the same domain different problems can be solved. Thus the procedure becomes more sustainable.


Machine learning structural optimization Finite element analysis 


  1. 1.
    Hajela, P., Berke, L.: Neural network-based decomposition in optimal structural synthesis. Comput. Syst. Eng. 2(5–6), 473–481 (1991)CrossRefGoogle Scholar
  2. 2.
    Holland, J.: Genetic algorithms. Sci. Am. 267(1), 66–72 (1992)CrossRefGoogle Scholar
  3. 3.
    Brown, N., Mueller, C.: Design variable analysis and generation for performance-based parametric modeling in architecture. Int. J. Archit. Comput. 17(1), 36–52 (2018)CrossRefGoogle Scholar
  4. 4.
    Tseranidis, S., Brown, N., Mueller, C.: Data-driven approximation algorithms for rapid performance evaluation and optimization of civil structures. Autom. Constr. 72, 279–293 (2016)CrossRefGoogle Scholar
  5. 5.
    Swift, R.A., Batil, S.M.: Application of neural networks to preliminary structural design. In: 32nd Structures, Structural Dynamics, and Materials Conference, Baltimore, MD, U.S.A. (1991)Google Scholar
  6. 6.
    Hanna, S.: Inductive machine learning of optimal modular structures: estimating solutions using support vector machines. AI EDAM 21(4), 351–366 (2007)Google Scholar
  7. 7.
    Preisinger, C.: Linking structure and parametric geometry. Archit. Des. 83, 110–113 (2013)Google Scholar
  8. 8.
    Mitchell, T.: The Discipline of Machine Learning. School of Computer Science, Carnegie Mellon University Pittsburgh, PA, (2006)Google Scholar
  9. 9.
    Dreyfus, S.: Artificial neural networks, back propagation and the Kelley-Bryson gradient procedure. J. Guidance Control Dyn. 13, 926–928 (1990)MathSciNetCrossRefGoogle Scholar
  10. 10.
  11. 11.
    Stanley, K.O.: Compositional pattern producing networks: a novel abstraction of development. Genet. Program. Evolvable Mach. 8(2), 131–162 (2007)CrossRefGoogle Scholar
  12. 12.
    Vierlinger, R.: Towards Ai Drawing Agents (2015)Google Scholar
  13. 13.
    Wagarachchi, M., Karunananda, A.: Optimization of artificial neural network architecture using neuroplasticity. Int. J. Artif. Intell. 15(1), 112–125 (2017)Google Scholar
  14. 14.
  15. 15.
    MIT Online Lecture Notes. Accessed 05 Apr 2019
  16. 16.
    Linear regression. Accessed 07 June 2019

Copyright information

© Springer Nature Switzerland AG 2020

Authors and Affiliations

  1. 1.University of Applied Arts ViennaViennaAustria

Personalised recommendations