Скачать или смотреть Deep learning algorithm to speed up the process of Chebyshev gridshell structural optimization

Parallel Session 15, Computational form-finding methods-2 – Deep learning algorithm to speed up the process of chebyshev gridshell structural optimization

Dai Kandil from @Institute for Advanced Architecture of Catalonia, Enrique Soriano & Dionís Boixader from Universitat Politècnica de Catalunya, Spain, present their work on gridshells.

The aim of this research is to develop an engine that uses a deep neural networks algorithm to speed up and simplify form finding and structural analysis of Chebyshev bending active gridshell. The systematic process generally involves many trials to reach a well performing structure. This algorithm lets users input freeform boundaries, cross sections and choose between materials as parameters to predict values of structural performance within a very short time. Also, it gives the ability to run optimization based on some desired parameters. A Generative Adversarial Networks (GANs) model is used additionally to predict an approximate 3D form. The automatic process of generating the dataset is discussed; as manual actions must be avoided due to large number of generated data. The algorithm’s architecture is explained through using different hyperparameters. After training and testing the model, a user interface is created to be handy for architects and designers to use.

Watch their presentation on "Deep learning algorithm to speed up the process of chebyshev gridshell structural optimization" from the Spatial Structures 2020/21 Conference.

#IASS #IASS2020_21 #SpatialStructures #Architecture #Structures #Engineering #LightweightStructures #CivilEngineering #StructuralEngineering #Conference #Construction #computationaldesign #nextgeneration #parallelsessions #gridshell #chebyshev #algorithm