30.03.2022

Bend, don't break.

Who isn't familiar with the magnificent facades of buildings designed by imaginative architects? Behind these spectacular glass facades, however, lurk a number of challenges: How do you make curved glass, and how do you calculate the bending process without shattering it into a thousand pieces? Computer scientists at ISTA (Institute of Science and Technology Austria) are developing a design tool that enables the use of cost-effective technology for curved glass panels. Based on a deep neural network, the tool allows for the free design of attractive glass facades.

Stockfoto*

How do you build a skyscraper out of glass?

Curved glass facades can be breathtakingly beautiful, but traditional construction methods are extremely expensive. Panes are typically manufactured by "hot bending," a process in which glass is heated and then shaped. This works, but it is energy-intensive and costly.

Cold-bent glass is a cheaper alternative, where flat sheets of glass are bent and attached directly to frames on-site. However, given the material's fragility, it is extremely difficult to find a shape that is both aesthetically pleasing and manufacturable.

When will the glass break? – a complicated calculation

Designing cold-bent glass facades presents a huge computational challenge. Ruslan Guseinov, a postdoctoral researcher at IST Austria, explains: "While it's possible to calculate when a single panel will break, the entire facade, which often comprises thousands of panels, is simply too complex for conventional design tools used by architects."

With an interactive, data-driven design tool, architects can now do just that.

The new design tool: Making beauty possible!

Developed by scientists (IST Austria, TU Wien, UJRC, KAUST), the software allows users to easily modify building facades with a computer mouse and receive immediate feedback on the building's appearance and its feasibility.

How does it work?

The software is based on a deep neural network trained with specialized physical simulations to predict the shapes and manufacturability of glass panels. It enables not only the interactive adaptation of a design but also the automatic optimization of a given design.

Tried and tested thousands of times

The team's goal was to develop software that allows even non-experts to interactively manipulate a glass surface. This means designing the building while simultaneously receiving real-time information about the curved shape and the associated stresses for each individual panel.

The idea:

The researchers opted for a data-driven approach: Every common glass shape was entered into a directory. The team ran more than a million simulations to build a database of possible curved glass shapes, represented in a CAD (Computer-Aided Design) format commonly used in architecture.

Design suggestions from the network

A deep neural network was then trained on this data. As a result, the network suggests one or two possible glass shapes based on the inputs. These suggested designs can then be used in a facade designed by architects.

The fact that this network predicts multiple shapes is "one of the most surprising aspects of deep neural networks," adds Konstantinos Gavriil, co-first author and researcher at TU Wien. “We knew that a specific frame wouldn’t definitively define the panel, but we didn’t expect the network to be able to find multiple solutions.

Ultimately, users of this new technology can modify, experiment, and adapt; they can develop variations for cold or hot bending or choose a ‘best fit’ solution suggested by the system.

“We believe we have created a novel, practical system that combines geometric and manufacturable design, enabling designers to efficiently strike a balance between economic, aesthetic, and technical criteria,” concludes Bernd Bickel, Professor at IST Austria.

Information & Sources:

Publikation K. Gavriil, R. Guseinov, J. Pérez, D. Pellis, P. Henderson, F. Rist, H. Pottmann, B. Bickel. 2020. Computational Design of Cold Bent Glass Façades. ACM Transactions on Graphics. DOI: https://doi.org/10.1145/3414685.3417843

http://visualcomputing.ist.ac.at/publications/2020/CDoCBGF/

This project was funded by the European Union’s Horizon 2020 research and innovation program under grant number 675789 – Algebraic Representations in Computer-Aided Design for complEx Shapes (ARCADES), by the European Research Council (ERC) under grant number 715767 – MATERIALIZABLE: Intelligent fabrication-oriented Computational Design and Modeling, and by the Austrian Science Fund (FWF) within the Collaborative Research Centre (SFB) Transregio “Discretization in Geometry and Dynamics” through grant I 2978. F. Rist and K. Gavriil received partial support through KAUST basic funding.