2023 Annual International Solid Freeform Fabrication Symposium (SFF Symp 2023): Process Development: Powder Bed Fusion Monitoring and Imaging I
Program Organizers: Joseph Beaman, University of Texas at Austin
Tuesday 8:15 AM
August 15, 2023
Room: 410
Location: Hilton Austin
Session Chair: David Hoelzle, Ohio State University
8:15 AM
Convergence Studies of Ensemble Kalman Filter Accuracy at Estimating Powder Bed Fusion Temperatures as a Function of Measurement Resolution: Nathaniel Wood1; Edwin Schwalbach2; Sean Donegan2; Andrew Gillman2; David Hoelzle1; 1Ohio State University; 2Air Force Research Laboratory
Achieving in-situ quality control of the Laser Powder Bed Fusion (PBF) process is an ongoing challenge, and resolving process features near the melt pool is an area of active research. High-resolution infrared (IR) cameras can image these features directly, but the data is impractical to process in-situ, and using less detailed measurements typically requires large quantities of training data. An algorithm called the Ensemble Kalman Filter (EnKF) avoids the tradeoff between accurate feature identification and measurement resolution by self-tuning the correlation between features and measurement values. Here, our feature is the PBF temperature field, and our measurement is a combination of two IR cameras. We perform two tests of EnKF accuracy as measurement resolution, and the data processing burden, decreases: First, decreasing the source IR camera resolution, where performance asymptotes after the resolution exceeds a threshold. Second, the resolution of the combined measurement decreases, where no asymptote is present.
8:35 AM
Experimental Validations of an Ensemble Kalman Filter Method for Powder Bed Fusion Temperature Estimation: Nathaniel Wood1; Edwin Schwalbach2; Sean Donegan2; Andrew Gillman2; David Hoelzle1; 1Ohio State University; 2Air Force Research Laboratory
Standard methods for estimating Laser Powder Bed Fusion (PBF) process variables rely on costly and time-consuming training data. The Ensemble Kalman Filter (EnKF) avoids this burden by using in-situ measurements to apply self-tuned corrections to physics-based model predictions of the process. Here, the process variable is the PBF temperature field, and the model is a Finite Element Method (FEM) description of heat conduction. In this work, we describe implementing the EnKF with this model and two PBF measurement architectures. Using data from previous experiments, we demonstrate EnKF effectiveness under three subsets of PBF process physics, which tests how it well it corrects increasing modeling error: solely heat conduction, melting a metal surface, and fusing layers of powder. The EnKF accurately estimates heat affected zone temperatures in every test, which is critical for PBF quality control, but incorrect estimates at isolated FEM nodes become more frequent as modeling error increases.
8:55 AM
An Ensemble Kalman Filter Method for Laser Powder Bed Fusion Temperature Estimation, Augmented with Adaptive Meshing and Joint Estimation of the Absorptivity: Nathaniel Wood1; Edwin Schwalbach2; Sean Donegan2; Andrew Gillman2; David Hoelzle1; 1Ohio State University; 2Air Force Research Laboratory
Methods for in-situ process monitoring of Laser Powder Bed Fusion (PBF) typically use large quantities of training data, since accurate predictive models are too computationally expensive. The Ensemble Kalman Filter (EnKF) overcomes this limitation by using the available measurements to apply self-tuned corrections to naïve model predictions. Here, we estimate the PBF temperature field, and the naïve model is Finite Element Method (FEM) heat conduction. The laser absorptivity is a modeling hyperparameter. We test three implementations of the EnKF using data from previous experiments: Implementation 1 uses a time-varying FEM mesh that is only dense nearby the laser (adaptive meshing), with fixed absorptivity. Implementation 2 uses a time-invariant mesh while jointly estimating temperature and absorptivity. Finally, Implementation 3 combines joint estimation and adaptive meshing. Implementations 1 and 2 show good accuracy in the heat affected zone, which enables accurate identification of several defect types, but performance suffers in Implementation 3.
9:15 AM
AI-driven In Situ Detection of Keyhole Pore Generation in Laser Powder Bed Fusion: Zhongshu Ren1; Tao Sun1; 1University of Virginia
Laser powder bed fusion (LPBF) process is a metal 3D printing technology, where the laser selectively melts powder and fuses it with the underneath substrate based on computer design. Certain defects such as porosity hinders the widespread adoption of LPBF into applications, which have strict quality requirements. One type of porosity defects occurs under some conditions of high laser power and slow scan speed, known as keyhole porosity. We developed an artificial intelligence (AI)-driven approach to detect the pore generation in situ with near-perfect prediction. We used the synchrotron high-speed x-ray imaging as ground truth and acquired simultaneous thermal imaging of the sample surface as training data. We also performed multiphysics simulation to reveal the physical meaning of the features used in the training process. This approach shows a practical way of detecting porosity defects and potential of improving the build parts quality.
9:35 AM Cancelled
In-situ Monitoring of Laser Powder Bed Fusion for Production Environments: Jesse Adamczyk1; David Saiz1; Dan Bolintineanu1; Anthony Garland1; Ana Love1; Hyein Choi1; David Moore1; Catherine Appleby1; Michael Heiden1; 1Sandia National Laboratories
Additive manufacturing (AM) has shown major growth across the energy, aerospace, and automotive sectors. However, there is a critical need for identification of process-induced defects and porosity. Such determinations typically require expensive and time-consuming techniques that are not amenable to production environments. In-situ monitoring can reduce the need for post-build inspection by leveraging correlated multi-modal data streams to identify off-nominal build events and potential defects. Acoustic signals during builds can be correlated to laser energy density, along with the part surface finish and density. Commanded laser positions linked with layer-wise optical images of the actual laser position enables external validation of correct machine behavior. Additionally, relative heat inputs and spatter behavior can be determined by long-wave infrared imaging. Ultimately, this work highlights how in-situ monitoring can benefit AM production by automatically tracking process deviations and anomalies.
9:55 AM Break