UNI-MB - logo
UMNIK - logo
 
Fakulteta za gradbeništvo in geodezijo, Ljubljana (FGGLJ)
  • Telescope enclosure optimization using machine learned generative design [Elektronski vir]
    Gradišar, Luka, gradbenik, 1995- ; Dolenc, Matevž
    A machine learned generative design approach aims to learn the behavior of the computational model using machine learning algorithms and apply them to generate design options. By using a generative ... design framework with machine learning surrogate models, we are able to explore large design spaces much faster because such models can compute more iterations and in a shorter time frame than traditional computationally intensive models, but at the cost of accuracy and training data. Such an approach allows us to combine multiple analyses into one overall model that we can use to learn the behavior and trends of the whole design problem and solve it. The MLGD method was used in the development of the enclosure for the New Robotic Telescope, which will be the first of its size class to use a clamshell enclosure design. The problem in optimizing such a design was the different operating states, specifically a closed state and an opening/closing state, each with different structural behavior, and the large search space of all possible design solutions. Using machine learning models, we were able to predict multiple responses for each state and combine them into a single optimization model, from which a set of feasible solutions was found and used to advance the project development. Compared to the initial design, in best case we were able to achieve a 28% mass reduction and 36% smaller forces in the hydraulic cylinders. Additionally, we were also able to present several trade-offs to the project team so that the implications of the trade study were well understood before engaging the design partner.
    Vir: AI in AEC 2023 [Elektronski vir] : Artificial Intelligence in Architecture, Engineering and Construction, virtual conference 2023, March 22-23, 2023 (1 spletni vir)
    Vrsta gradiva - prispevek na konferenci
    Leto - 2023
    Jezik - angleški
    COBISS.SI-ID - 155313411

    Povezava(-e):

    https://aiaec2023.exordo.com/programme/presentation/2

Vnos na polico