Neural field-based shape optimization for manufacturability-aware support structure minimisation

Support structure requirement for overhang regions is perhaps the most critical constraint in material extrusion-based additive manufacturing (AM), which increases material waste and print time, leading to high production cost. Direct shape optimisation eliminates the need to manually refine or redesign complex geometries to meet manufacturability requirements, including minimising support structures. However, direct mesh-based shape manipulation (e.g., rigid rotations) methods are not only limited in terms of part complexity and mesh resolution, but also inherent to large deviation and surface feature distortions. In this paper, we present a novel AM-oriented, end-to-end neural shape optimisation framework to minimise (if not eliminate) overhang regions of a range of complex geometric models with non-trivial topology and intricate surface features. Our method learns the optimal geometric deformation using a neural field which is governed by a set of manufacturability-oriented loss functions. This mesh-free approach realises overhang minimisation under negligible surface feature distortion, and minimal deviation for all the tested geometries. By leveraging the smoothness and continuity of the neural field, we then introduce a coarse-to-fine optimisation workflow to realise direct and efficient optimisation of high resolution meshes. The proposed approach is validated through extensive computational and physical printing experiments. Our results show an average reduction of 46% in support-structure print time, including a case in which support-structure print time is reduced by 100% for a complex geometry, which clearly demonstrate the effectiveness of the proposed computational framework and its potential as a strong foundation towards AI-driven design to achieve support-free extrusion-based additive manufacturing.