Overcoming small sample challenges in predicting mechanical properties of 3D printed parts via feature engineering and model interpretation

Name: M. Saedan

Organization : Chiang Mai University. Department of Mechanical Engineering

Email : mana.saedan@cmu.ac.th

Name: W. Tachajapong

Organization : Chiang Mai University. Department of Mechanical Engineering

Name: P. Suttakul

Organization : Chiang Mai University. Department of Mechanical Engineering

keyword: XGBoost

LCSH: Three-dimensional printing

; Tensile strength

LCSH: Additive manufacturing

; Feature engineering (Computer science)

LCSH: Machine learning

LCSH: Experimental design

Abstract: Additive manufacturing enables rapid prototyping, but optimizing mechanical properties like ultimate tensile strength (UTS) remains challenging due to complex parameter interactions and limited experimental data. This study addresses small-sample constraints by integrating an augmented experimental design with machine learning to predict UTS in polyethylene terephthalate glycol (PETG) 3D-printed parts. A primary Taguchi L27 orthogonal array investigated twelve printing parameters. To expand dataset coverage, 24 additional parameter combinations were generated using computer-simulated designs that emulate Taguchi principles of balance and orthogonality, resulting in a total of 51 unique experimental runs. After removing high-variance data points, a final dataset of 40 samples was used for modeling. XGBoost regression models were trained using cross-validation to mitigate overfitting. SHapley Additive exPlanations (SHAP) analysis revealed wall loops and extrusion width as the most influential parameters. A feature selection method, SelectKBest, improved model robustness, achieving an evaluation R-squared of 0.782 with only 10 key features. The proposed framework demonstrates that combining strategic experimental design, interpretable machine learning, and automated feature engineering enables accurate UTS prediction under data scarcity, offering a practical tool for parameter optimization in resource-constrained settings.

King Mongkut's University of Technology North Bangkok. Central Library

Address: BANGKOK

Email: library@kmutnb.ac.th

Created: 2025

Modified: 2026-06-25

Issued: 2026-06-25

บทความ/Article

application/pdf

BibliograpyCitation : In Thai Society of Mechanical Engineer (TSME) and Mahanakorn University of Technology (MUT). The 15th TSME International Conference on Mechanical Engineering (TSME-ICoME 2025) (AMM0003). Bangkok : Mahanakorn University of Technology

eng

©copyrights King Mongkut's University of Technology North Bangkok

ใช้เวลา

0.033473 วินาที