Real-time design of architectural structures with differentiable mechanics and neural networks

Designing mechanically efficient geometry for architectural structures like shells, towers, and bridges is an expensive iterative process. Existing techniques for solving such inverse mechanical problems rely on traditional direct optimization methods, which are slow and computationally expensive, limiting iteration speed and design exploration. Neural networks would seem to offer a solution, via data-driven amortized optimization for specific design tasks, but they often require extensive fine-tuning and cannot ensure that important design criteria, such as mechanical integrity, are met. In this work, we combine neural networks with a differentiable mechanics simulator to develop a model that accelerates the solution of shape approximation problems for architectural structures modeled as bar systems. As a result, our model offers explicit guarantees to satisfy mechanical constraints while generating designs that match target geometries. We validate our model in two tasks, the design of masonry shells and cable-net towers. Our model achieves better accuracy and generalization than fully neural alternatives, and comparable accuracy to direct optimization but in real time, enabling fast and sound design exploration. We further demonstrate the real-world potential of our trained model by deploying it in 3D modeling software and by fabricating a physical prototype. Our work opens up new opportunities for accelerated physical design enhanced by neural networks for the built environment.