From metrics to decisions: a robust evidential methodology for dimensional accuracy and manufacturability in DMLS SS316L

Abstract Dimensional accuracy remains a persistent barrier to qualified deployment of direct metal laser sintering metal parts because multiple deviation indicators and manufacturability constraints must be optimized simultaneously rather than in isolation. This study addresses the gap by proposing a Robust Evidential Multi-Criteria Geometry Optimizer (REMGO) that ranks parameter sets while explicitly propagating uncertainty in both criterion utilities and decision scores. A Taguchi L9 design (laser power 300-360 W, scan speed 800-1000 mm/s, layer thickness 40-60 μm; three replicates, 27 samples) was fabricated in SS316L on an EOS M290. Dimensional deviations were quantified using a CNC coordinate measuring machine under controlled metrology conditions. Five geometry/manufacturability criteria capturing central tendency, tail behavior, and spread of dimensional error were aggregated using median absolute deviation scaling, evidential weighting derived from grouped 5-fold cross-validated performance, and interval aware TOPSIS closeness based on bootstrap resampling. REMGO identified the combination of 300 W, 900 mm/s, 40 μm as optimal and Sample 4 is reported as a representative replicate, achieving a median closeness of 0.839 with a tight 90% interval of (0.747, 0.837). Benchmarking confirmed the same leading solution under the weighted sum model (score 0.8937, Rank 1) and showed near tie behavior under VIšekriterijumska Optimizacija I Kompromisno Rešenje (VIKOR) (Sample 4: Index = 0.038, Rank 2; Sample 17: Index = 0.035, Rank 1), with a preference shift toward Sample 4 for compromise parameter ≥ 0.6. The results indicate an operating window that balances geometric accuracy with a cost proxy of manufacturability, while providing a conservative and quantified uncertainty recommendation.