Machine learning-enhanced reverse engineering for dimensional accuracy optimization in DMLS-fabricated SS316L components
Year: 2025
Type: article
Abstract: The present study develops an integrated methodology that links design of experiments, reverse engineering, and data-driven prediction to minimize dimensional deviation in direct metal laser sintering of SS316L stainless steel. A full factorial process matrix was implemented in which laser power, scan speed, and layer thickness were systematically varied, and all builds were produced on an EOS M 290 system. Dimensional accuracy was quantified using high-precision coordinate measurement and reverse engineering, providing reference geometry for analysis. Dimensional deviation was found to be primarily controlled by the coupled effects of scan speed and layer thickness, with values ranging from -0.886 mm to 0.261 mm across the build set. The most favorable results were obtained at an intermediate laser power of 330 W, a moderate scan speed of 800 mm/s, and a layer thickness of 40 µm. When averaging across all samples, the average deviation (bias) was -0.057 mm which corresponds to a consistent, slight undersize trend in the samples with a normal deviation of ±0.157 mm. Supervised regression models (Random Forest and XGBoost) were trained on the reverse engineering dimensional data and reproduced the recorded deviations to a high degree of accuracy (R 2 > 0.999, mean absolute percentage deviation < 1%). Of the two models, Random Forest regression showed slightly better generalization, RMSE = 0.1154 mm and MAPE = 0.68% and remained within error limits of ±1% in the confirmation builds. By controlling metrology grade measurement with predictive modeling, dimensional quality control is changed from a retrospective inspection step to being predictive and exercise in variable selection. This enables proactive tuning of process conditions to reduce scrap and limit downstream finishing. Such a workflow supports the shift to intelligent closed loop control of laser powder bed fusion systems and has direct relevance to high precision applications for these processes in aerospace, biomedical device manufacture and tooling.
Source: Materials Today Communications
Authors: Bridjesh Pappula, Jagadesh Kumar Jatavallabhula, Venkata Phani Babu Vemuri, Ravikiran Chintalapudi, Geetha Narayanan Kannaiyan, Seshibe Makgato
Institution: University of South Africa
Language: English
Subfield: Mechanical Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $2,830
OpenAlex ID: W4416092164