Multiscale topology optimization using physics-augmented neural networks

In our latest publication, Jonathan Stollberg in collaboration with our former postdoc Tarun Gangwar (IIT Roorkee) and Prof. Oliver Weeger (Cyber-Physical Simulation) presents a novel optimization algorithm for functionally graded and additively manufacturable lattice structures. Among other things, the method is based on a parameterized material model, which was implemented as a neural network. Basic physical conditions for the model are already fulfilled by the architecture of the neural network. The article is available as open access in Computer Methods in Applied Mechanics and Engineering:

Stollberg, J., Gangwar, D., Weeger, O., Schillinger, D. Multiscale topology optimization of functionally graded lattice structures based on physics-augmented neural network material models. Comput. Methods Appl. Mech. Eng. 438, 117808 (2025). DOI: 10.1016/j.cma.2025.117808