An Insight into Defect Mechanisms and Process Optimization in Laser Powder Bed Fusion of 18Ni300 Using Hybrid Models

Abstract

The laser powder bed fusion (LPBF) process used for 18Ni300 (MS300) parts often faces challenges such as porosity and defects, impacting relative density (RD) and surface roughness (SR) of final product. This study presents an artificial intelligence-based prediction model, validated through both experimental and analytical methods. Specifically, artificial neural networks (ANN), genetic algorithms (GA), and response surface methodology (RSM) are employed to predict RD and SR outcomes effectively. Additionally, an analytical model based on Gaussian distribution patterns and molten pool dimensions was developed to estimate energy density necessary for complete melting. The results showed a variation in energy density requirements: the Gaussian model predicted 344 J/m, RSM estimated 334 J/m, and the GA model suggested a lower requirement of 286 J/m. Focusing on RD and SR predictions, the ANN model demonstrated prediction errors ranging from 0.01 to 2.75% for RD and 0.06% to 31.05% for SR. In comparison, the central composite design (CCD) model, a subset of RSM, exhibited errors from 0.01 to 0.56% for RD and 0.25 to 29.49% for SR. The novelty of this work lies in its comprehensive use of advanced predictive modeling to not only determine optimal energy density but also to forecast surface characteristics accurately. The study provides deeper insights into defect mechanisms in low-density (LD) samples by investigating defects such as balling, keyhole effects, unfused particles, and porosity and details microstructural analysis of high-density (HD) samples, paving the way for enhancing LPBF process outcomes. © ASM International 2025.