This paper aims to report the implementation and initial experimental evaluation of an adaptive infill method for fused deposition modeling that combines cross-section-driven variable-density assignment with an explicit transition-layer branch triggered at detected geometric discontinuities, all operating at the slicing stage without external geometry modification or load-path input.
A continuous exponential mapping converts the slice-wise cross-sectional area sequence into a layer-specific infill density bounded by user-defined limits. A dual-criterion detection module identifies abrupt geometric transitions using area-change rate and centroid displacement; a sliding-window slope-exclusion rule prevents false detection on continuous gradients; densified reinforcement layers are inserted only at flagged abrupt interfaces. The two algorithmic branches were exercised separately on two test geometries – compression specimens with circular necked profiles and tensile specimens with rectangular stepped profiles – printed in polylactic acid (PLA) at 20\% baseline infill density with five specimens per group.
The adaptive group showed a marginal trend toward higher first peak force on compression specimens (5.6∼kN to 5.9∼kN, +5.4\%; Welch’s t-test, p = 0.054) and a statistically significant 15.4\% higher peak load on tensile specimens (803.2∼N to 926.8∼N; p = 0.043), with mass increases of 0.5\% and printing time increases below 7\%. A directional check with glycol-modified polyethylene terephthalate (PETG) at three specimens per group gave +14.2\% tensile peak-load difference (p = 0.0045), consistent in direction with the PLA result. Sparse-infill toolpath length differed from the uniform baseline by less than 1.5\%. The algorithm transferred without modification to three geometries of greater complexity.
This work presents an integrated implementation that combines an existing class of cross-section-driven variable-density assignment with an explicit transition-layer detection and reinforcement branch within a single slicing-level pipeline, together with per-layer diagnostic data that make the algorithm’s geometric decisions inspectable.
