The neural network was taught to build huge glass facades

Traditional methods of building curved glass road facades. Scientists have created software that makes the process cheaper and simplifies it at all stages of implementation.

Shaped glass is usually manufactured using the “hot bending” method. It is heated and shaped using a press or special machines. This is an energy-intensive process that generates unnecessary waste. Cold bent glass is a cheaper alternative in which flat glass is folded and attached to frames. But coming up with a shape that is aesthetically pleasing and easy to manufacture, given the material’s fragility, is a challenge. A new interactive design tool helps architects create these complex designs.

The software (software), created by a group of scientists from IST Austria, TU Wien, UJRC, and KAUST, allows users to interactively control the facade’s design and receive immediate feedback on the feasibility of its manufacture and the aesthetics of paneling. This is a convenient way to choose the best design available. The new software is based on a deep neural network trained to predict glass panels’ shape and manufacturability. In addition to allowing users to customize the design, the software automatically optimizes it for integration into the architect’s normal workflow.

The design of cold-formed glass facades is a huge computational challenge. It’s too complicated for humans, but processing thousands of data and variations is not a problem for machine learning software. The scientists sought to create software that would allow a user (not an expert) to edit the surface interactively, receiving real-time information about each panel’s glass bending shape and stress. They took a data-driven approach: the team ran over a million simulations to create a database of possible curved glass shapes represented in the architecture’s traditional CAD format. Then a deep neural network (DNN) was trained on this data.

Google News button