Page 41 - IJAMD-1-3

International Journal of AI for
Materials and Design
Metal AM porosity prediction using ML

Table 1. Overview of proposed machine learning algorithms Table 1. (Continued)
to predict defects during the L‑PBF process
No. Insights of research References
No. Insights of research References 7 • Deep learning predicts porosity in L‑PBF 25
1 • The paper presents a deep learning approach 19 additively manufactured parts.
for defect detection. • ANN model trained with X‑ray CT images for
• It focuses on porosity and melt pool geometry accuracy.
segmentation. • Synthetic CT data enhances model
• Utilizes encoder‑decoder networks with performance and predictions.
optimization techniques for accuracy. • The study addresses porosity challenges in
• Addresses challenges in noisy microscopy data AM.
for multi-layer structures. 8 • The paper predicts micropore defects in L‑PBF 26
• Demonstrates superior performance in using thermal imaging.
identifying multiple features simultaneously. • ML models analyze in situ thermographic data
2 • The paper proposes a neural network for in situ 20 for predictions.
defect prediction. • Key features include time for achieving
• It focuses on microporosity localization in L‑PBF. melting threshold and maximum
• Utilizes within hatch stripe sensory data for radiance.
improved accuracy. 9 • The paper presents DSPMs. 27
• Achieves classification accuracy of 73.13% for • DSPMs quantify porosity in metal additive
porosity detection. manufacturing.
• Demonstrates a significant improvement in • Synchrotron‑based micro‑computed
detecting small porosities tomography identifies defect trends.
• Aims to enhance process control and defect • Ti‑6Al‑4 V test blocks were fabricated using
mitigation. varied parameters.
3 • The paper addresses porosity detection in AM. 21 • Keyhole and lack‑of‑fusion defects were
• ML methods are compared for porosity analyzed and mitigated.
classification. • Processing parameters significantly
• DCNNs outperform traditional methods in affect defect formation in L-PBF
accuracy. materials.
• DCNN achieved 95% accuracy. 10 • The paper studies porosity in L‑PBF. 28
4 • The paper develops a model for predicting 22 • X‑ray tomography reveals pore formation
keyhole porosity. mechanisms and characteristics.
• It uses a closed‑form analytical approach • Pore size and shape vary with process
without numerical calculations. parameters.
• The model predicts porosity based on molten • Increased power leads to more keyhole mode
pool characteristics. porosity.
• The proposed model shows good predictive • Lack of fusion occurs with poor hatch
accuracy and computational efficiency. overlap.
• Optimized parameters improve part density
5 • The paper presents a physics‑informed ML 23 and quality.
model for porosity analysis.
• It addresses the limitations of • Insights assist in quality control and process
improvement.
machine-dependent porosity prediction models.
• The model interprets machine settings into 11 • The study investigates AlSi10Mg alloy 29
physical effects. properties via L-PBF.
• It predicts porosity levels using “pass,” “flag,” • Sub‑optimal parameters affect density and
and “fail” categories. microstructure.
• The model achieved a prediction • Defect orientation impacts mechanical
error of 10 – 26%. properties significantly.
6 • The paper presents a ML approach for defect 24 • Findings are relevant for various
detection. L-PBF-fabricated AlSi10Mg alloys.
• It integrates fuzzy logic and self‑organizing 12 • The paper investigates pore formation in L‑PBF 30
maps for analysis. AM.
• The model predicts a lack of fusion and • Six pore formation mechanisms were identified
keyhole defects. during the L-PBF process.
• Experimental validation shows strong • Pores significantly affect mechanical
performance across various parameters. performance and fatigue life.
• Customizable fuzzy rules enhance defect • Understanding mechanisms aids in developing
detection accuracy. pore mitigation strategies.
(Cont'd...) (Cont'd...)

Volume 1 Issue 3 (2024) 35 doi: 10.36922/ijamd.4812