An official website of the United States government
Understanding the fatigue behavior and accurately predicting the fatigue life of laser powder bed fusion (L-PBF) parts remain a pressing challenge due to complex failure mechanisms, time-consuming tests, and limited fatigue data. This study proposes a physics-informed data-driven framework, a multimodal transfer learning (MMTL) framework, to understand process-defect-fatigue relationships in L-PBF by integrating various modalities of fatigue performance, including process parameters, XCT-inspected defects, and fatigue test conditions. It aims to leverage a pre-trained model with abundant process and defect data in the source task to predict fatigue life nondestructive with limited fatigue test data in the target task. MMTL employs a hierarchical graph convolutional network (HGCN) to classify defects in the source task by representing process parameters and defect features in graphs, thereby enhancing its interpretability. The feature embedding learned from HGCN is then transferred to fatigue life modeling in neural network layers, enabling fatigue life prediction for L-PBF parts with limited data. MMTL validation through a numerical simulation and real-case study demonstrates its effectiveness, achieving an F1-score of 0.9593 in defect classification and a mean absolute percentage log error of 0.0425 in fatigue life prediction. MMTL can be extended to other applications with multiple modalities and limited data.
more »
« less
Defects Classification via Hierarchical Graph Convolutional Network in L-PBF Additive Manufacturing
Three typical types of defects, i.e., keyholes, lack of fusion (LoF), and gas-entrapped pores (GEP), characterized by various features (e.g., volume, surface area, etc.), are generated under different process parameters of laser beam powder bed fusion (L-PBF) processes in additive manufacturing (AM). The different types of defects deteriorate the mechanical performance of L- PBF components, such as fatigue life, to a different extent. However, there is a lack of recognized approaches to classify the defects automatically and accurately in L-PBF components. This work presents a novel hierarchical graph convolutional network (H-GCN) to classify different types of defects by a cascading GCN structure with a low-level feature (e.g., defect features) layer and a high-level feature (e.g., process parameters) layer. Such an H-GCN not only leverages the multi- level information from process parameters and defect features to classify the defects but also explores the impact of process parameters on defect types and features. The H-GCN is evaluated through a case study with X-ray computed tomography (CT) L-PBF defect datasets and compared with several machine learning methods. H-GCN exhibits an outstanding classification performance with an F1-score of 1.000 and reveals the potential effect of process parameters on three types of defects.
more »
« less
- Award ID(s):
- 2134689
- PAR ID:
- 10438211
- Date Published:
- Journal Name:
- 2022 International Solid Freeform Fabrication Symposium
- Page Range / eLocation ID:
- 1568-1580
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Additive manufacturing (AM), particularly Laser Powder Bed Fusion (L-PBF), holds the potential for producing high-quality parts with intricate details. However, optimizing process parameters for arbitrary alloy chemistries to ensure printability remains challenging. This study evaluates machine learning (ML) models to predict a material’s amenability to L-PBF via the printability index, focusing on High Entropy Alloy (HEA) spaces. The printability index of a material is defined as the percentage of the defect-free L-PBF processing window. Our study revealed that CatBoost Regressors and Random Forest Regressors excel in predictive accuracy, consistently yielding predictions with competitive error metrics such as the mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE), and . In addition, competitive rank-order coefficients such as Spearman and Kendall-tau demonstrated that these models are not overfitting. Feature importance analysis using Shapley Additive Explanations (SHAP) highlighted key material properties influencing printability, such as kinetic viscosity, average Pauling electronegativity, and electric conductivity. While both models performed comparably in predictive accuracy, the Random Forest Regressor demonstrated superior computational efficiency, particularly with large datasets. Robustness tests confirmed its reliability across different test sizes. This research underscores the importance of considering factors like computational efficiency, interpretability, and robustness to noise when selecting ML models for L-PBF material printability prediction. Leveraging Integrated Computational Materials Engineering (ICME) methodologies and ML models can significantly optimize process parameters and material properties, paving the way for innovative solutions in L-PBF. This approach accelerates the assessment of new materials and optimizes existing ones for L-PBF processes, contributing significantly to the field of AM.more » « less
-
ABSTRACTElectro-chemical polishing (ECP) was utilized to produce sub-micron surface finish on Inconel 718 parts manufactured by Laser Powder-Bed-Fusion (L-PBF) and extrusion methods. The L-PBF parts had very rough surfaces due to semi-welded powder particles, surface defects, and difference layer steps that were generally not found on surfaces of extruded and machined components. This study compared the results of electro-polishing of these differently manufactured parts under the same conditions. Titanium electrode was used with an acid-based electrolyte to polish both the specimens at different combinations of pulsed current density, duty cycle, and polishing time. Digital 3D optical profiler was used to assess the surface finish, while optical and scanning electron microscopy was utilized to observe the microstructure of polished specimens. At optimal condition, the ECP successfully reduced the surface of L-PBF part from 17 µm to 0.25 µm; further polishing did not improve the surface finish due to different removal rates of micro-leveled pores, cracks, nonconductive phases, and carbide particles in 3D-printed Inconel 718. The microstructure of extruded materials was uniform and free of processing defects, therefore can be polished consistently to 0.20 µm. Over-polishing of extruded material could improve its surface finish, but not for the L-PBF material due to defects and the surrounding micro-strain.more » « less
-
Laser powder bed fusion (LPBF) is an additive manufacturing process that has gained interest for its material fabrication due to multiple advantages, such as the ability to print parts with small feature sizes, good mechanical properties, reduced material waste, etc. However, variations in the key process parameters in LPBF may result in the instantiation of porosity defects and variation in build rate. Particularly, volumetric energy density (VED) is a variable that encapsulates a number of those parameters and represents the amount of energy input from the laser source to the feedstock. VED has been traditionally used to inform the quality of the printed part but different values of VED are presented as optimal values for certain material systems. An optimal VED value can be maintained by changing the key process parameters so that various combinations yield a constant value. In this study, an optimal constant VED value is maintained while printing SS316L with variable key processing parameters. Porosity analysis is performed using optical microscopy, as well as X-ray computed tomography, to reveal the volume density and distribution of those pores. Two primary defect categories are identified, namely lack of fusion and porosity induced by balling defects. The findings indicate that, even at optimal VED, variations in process parameters can significantly influence defect type, underscoring the sensitivity of defect formation to the variation of these parameters. Furthermore, a minor change in the build rate, driven by adjustments in process parameters, was found to influence defect categories. These findings emphasize that fine tuning the process parameters and build rate is essential to minimize defects. Finally, fiducial marks have been identified as a source of unintentional porosity defects. These results enable the refinement of process parameters, ultimately optimizing LPBF to achieve enhanced material density and expedite the printing.more » « less
-
Abstract Laser powder-bed fusion (L-PBF) additive manufacturing presents ample opportunities to produce net-shape parts. The complex laser-powder interactions result in high cooling rates that often lead to unique microstructures and excellent mechanical properties. Refractory high-entropy alloys show great potential for high-temperature applications but are notoriously difficult to process by additive processes due to their sensitivity to cracking and defects, such as un-melted powders and keyholes. Here, we present a method based on a normalized model-based processing diagram to achieve a nearly defect-free TiZrNbTa alloy via in-situ alloying of elemental powders during L-PBF. Compared to its as-cast counterpart, the as-printed TiZrNbTa exhibits comparable mechanical properties but with enhanced elastic isotropy. This method has good potential for other refractory alloy systems based on in-situ alloying of elemental powders, thereby creating new opportunities to rapidly expand the collection of processable refractory materials via L-PBF.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
