Multi-Objective Build-Orientation Optimisation for Support-Structure Reduction in Additive Manufacturing
Support structures in additive manufacturing consume additional material and build time, and damage the contact surface after removal. The build orientation determines which surfaces become overhangs before slicing, and is the lowest-cost intervention point, but the orientation space is continuous and the objective is highly non-convex, so neither the default orientation nor grid trial placements can reliably locate the optimal region. This paper expresses build-orientation selection as a bi-objective optimisation problem of support volume and build height, solved by a multi-start NSGA-II with axis-seeded initialisation and compared against area-uniform spherical grid enumeration at a matched evaluation budget on 120 real printable models from Thingi10K. Among the 84 models that require support under the as-uploaded orientation, optimisation reduced the median support volume by 69.0% (interquartile range 3.5% to 97.4%, p<0.001), while 16 models did not improve, so the benefit depends strongly on the specific geometry. With the grid given slightly more evaluations, the normalised hypervolume of NSGA-II was no less than that of the grid on 112 models (p<0.001). The two objectives were negatively correlated within all fronts, so the trade-off genuinely exists. The reduction is sensitive to the critical overhang angle: at 60° the median reduction is only 1.4%. This paper gives a budget-matched comparison protocol and quantifies the applicable boundary of the method's benefit.