Differentiable graph neural fields for manufacturable topology optimization under stress constraints

This paper presents a machine learning-based framework for topology optimization of self-supporting structures, specifically tailored for additive manufacturing (AM). By employing a graph neural network (GNN) that acts as a neural field over the finite element mesh, the framework effectively learns and predicts continuous material distributions. An integrated AM filter ensures printability by eliminating unsupported overhangs, while the optimization process minimizes structural compliance under volume and stress constraints. The stress constraint is enforced using a differentiable p -norm aggregation of von Mises stress, promoting mechanical reliability in the optimized designs. A key advantage of the approach lies in its fully differentiable architecture, which leverages automatic differentiation throughout the optimization loop, eliminating the need for explicit sensitivity derivation for both the filter and the stress constraint. Numerical experiments demonstrate the ability of the framework to generate stress-constrained manufacturable topologies under various loading and boundary conditions, offering a practical pathway toward AM-ready high-performance designs with reduced post-processing requirements.