Curvature-Aware Feedrate Control for Non-Planar Additive Manufacturing

Recent advances in non-planar deposition have shown that conformal layers suppress stair-stepping on curved shells, yet feedrates remain constant per layer—a mismatch that concentrates defects where surfaces bend and confines reliable multi-axis printing to industrial platforms. We present a pipeline that mitigates this constraint on accessible hardware by modulating deposition velocity point-wise along conformal trajectories. Layers generated in Grasshopper are exported as point–normal sequences; a Python stage computes discrete mean curvature at each point through a cotangent-Laplacian operator, and a Random Forest regressor—trained on labels derived exclusively from instrumented calibration prints, with PrusaSlicer, Cura, and OrcaSlicer profiles as baselines—predicts a locally adapted feedrate (mm/s) before G-code generation, achieving a cross-validated R² of 0.85 on held-out trajectories. Factorial trials on a Wlkata Haro 380 with Bowden extruder, spanning shells of positive, zero, and negative Gaussian curvature, show that curvature-modulated feedrate improves mean surface roughness by 16% and reduces its standard deviation by 36%, stabilizes extrusion width (line-width coefficient of variation reduced by 30%), and increases the maximum structural load sustained before failure by 10% under quasi-static compression, with ANOVA interaction effects significant at p = 0.009 across roughness, line width, and load at failure. Print success climbs from 40% to 62% of attempts, rendering geometries printable that consistently failed under constant feedrate on the same low-cost Bowden and lightweight-arm configuration. Deterministic offline execution positions curvature-aware process planning as a deployable upgrade path for laboratories and small manufacturers operating entry-level robotic arms.