Automated nonplanar slicing for robotic additive manufacturing via voxelized simulation and data-driven voting

Robotic arms enable curved-layer deposition beyond conventional gantry limitations, but selecting a suitable layer field remains challenging. Existing slicers often focus on surface conformity or require part-specific finite element constraints. This study selects layer fields from a bounded library of analytical height fields using voxelized toolpath–mesh agreement and load–layer-plane alignment. Weighted voting selects a field family under a user-defined objective weight, followed by scaling-factor estimation. Random-forest surrogates approximate evaluator-generated labels for repeated queries, with direct evaluation retained as the reference. Across eleven objective weights for one dogbone geometry, surrogate and direct evaluation selected the same field family but differed in scaling factor. Leave-one-geometry-out testing on four meshes showed limited, geometry-dependent surrogate accuracy, supporting direct evaluation for unseen parts. A robotic printing trial used an inclined-layer field selected under an earlier model configuration. Under the tested process, mean global geometric deviation was 2.32 × the planar mean, and mean ultimate tensile strength was 64.7% lower. Fracture morphology shifted from predominantly interlayer separation in planar specimens to failure through deposited roads in nonplanar specimens, consistent with the change in deposition orientation. The framework provides a traceable procedure for multiobjective layer-field selection and identifies limitations in surrogate generalization and physical realization.