Additive Manufacturing Modeling and Simulation: AM Materials, Processes, and Mechanics: Additive Manufacturing Modeling and Simulation - Machine Learning, Design and Optimization
Sponsored by: TMS Additive Manufacturing Committee
Program Organizers: Jing Zhang, Purdue University in Indianapolis; Brandon McWilliams, US Army Research Laboratory; Li Ma, Johns Hopkins University Applied Physics Laboratory; Yeon-Gil Jung, Korea Institute of Ceramic Engineering & Technology
Tuesday 2:00 PM
November 3, 2020
Room: Virtual Meeting Room 2
Location: MS&T Virtual
Session Chair: Jing Zhang, Indiana University – Purdue University Indianapolis; Brandon McWilliams, CCDC Army Research Laboratory ; Li Ma, Johns Hopkins University Applied Physics Laboratory; Yeon-Gil Jung, Changwon National University
2:00 PM
Reduced-order Process-structure Linkages during Post-Process Annealing of an Additively Manufactured Ni-base Alloy: Andrew Marshall1; Surya Kalidindi1; Bala Radhakrishnan2; John Turner2; 1Georgia Institute of Technology; 2Oak Ridge National Laboratory
Physics-based multiscale simulations of additive manufacturing are able to quantify the effect of processing conditions (i.e. thermal history) on the evolution of the material structure. However, these simulations are computationally costly, limiting the ability to accurately upscale the model predictions to benefit process optimization. We present physics-aware, reduced-order process-structure linkages for a materials dataset generated from phase field simulations of solid-state transformations during post-process annealing of additively manufactured 625 alloy using a surrogate Ni-Al-Nb ternary alloy. The linkages are developed via Gaussian process regression, a Bayesian machine learning approach that provides prediction uncertainty. Exploiting this property, protocols are developed to provide objective guidance for the selection of new simulations. These protocols allow for rapid identification of the governing physics for the materials phenomena of interest. Research funded by the Department of Energy’s Exascale Computing Project. Phase field simulations were performed using the Oak Ridge Leadership Class Computing Facilties at ORNL.
2:20 PM
Phase Field Modeling of AM Solidification Microstructure with Algorithmic Feature Extraction to Facilitate Reduced Order Model Development: Stephen DeWitt1; Bala Radhakrishnan1; Yuanxun Bao2; George Biros2; Jean-Luc Fattebert1; John Turner1; 1Oak Ridge National Laboratory; 2University of Texas at Austin
The solidification microstructure that forms during additive manufacturing strongly influences the material’s mechanical properties. Improved understanding of the process-microstructure-property relationship can enable the optimization of process conditions for desired localized properties. Here, we discuss the use phase field simulations to predict microstructural features during the additive manufacturing of aluminum alloys in 2D and 3D. We compare two common approaches: the Kim-Kim-Suzuki (KKS) model and a simplified dilute binary model. The effect of solidification conditions on traditional microstructural features (e.g. primary/secondary dendrite arm spacing, solute distribution) and non-traditional features related to hot cracking (e.g. the morphology of late-solidifying regions between the dendrites) is discussed. These quantities of interest are collected algorithmically from the phase field simulations to facilitate the creation of reduced order models for use in uncertainty quantification and optimal control problems. Sensitivity analysis results are presented as an example of this workflow.
2:40 PM
A Process Parameter Prediction Framework for Metal Additive Manufacturing: Praveen Sreeramagiri1; Ankit Roy1; Ganesh Balasubramanian1; 1Lehigh University
Process parameter optimization plays a vital role in additive manufacturing, as the processing conditions can directly influence the quality of the synthesized materials and components. The current optimization methods involve a Design of Experiments (DoE) approach within a parameter window that emphasizes on three primary variables, viz., laser power, powder feed rate and scanning velocity. However, selection of the initial parameter window is itself a challenge and theory-guided methodologies, which can facilitate the baseline parameters, are sparse. We present results from a computational framework that is able to predict the process parameter window for advanced metals and alloys based on their solidification behavior, employing classical atomistic simulations of material properties (such as, diffusion coefficient, phase segregation) that are crucial during laser deposition. The predictive model correlates the atomic scale material features to the macroscopic manufacturing conditions. Keywords: Metal Additive Manufacturing, Processing Parameters, Molecular Dynamics
3:00 PM
Feature Engineering for Surrogate Models of Consolidation Degree in Additive Manufacturing: Mriganka Roy1; Olga Wodo1; 1University at Buffalo
The limited in-situ control, process optimization, and quality assurance are hindering AM’s widespread acceptance. A fast and accurate process evaluation could alleviate these challenges. In recent years there has been an effort to achieve this goal by developing surrogate models (SM). In this work, we utilize the knowledge of the underlying process to engineer features that efficiently parameterizes the geometry and printing pattern. We quantify the localized behavior of the process by defining a heat influence zone that limits the search area for the features and the size of the feature set. The engineered features enabled the training of the SM which was 1000 times faster than the numerical model and highly accurate (90% accuracy).
3:20 PM
Multi-Fidelity Surrogate Assisted Prediction of Melt Pool Geometry in Additive Manufacturing: Nandana Menon1; Sudeepta Mondal1; Daniel Gwynn1; Amrita Basak1; 1Pennsylvania State University
Melt pool geometry plays a critical role in controlling the microstructure and therefore the properties of additively manufactured components. There is a plethora of models available for predicting the steady state melt pool geometry in metal additive manufacturing. There models are either fast but inaccurate (e.g., analytical models) or slow and accurate (e.g., finite element-based models). The objective of this paper is to take advantage of the hierarchy of multi-scale multi-physics models to construct a multi-fidelity (MF) surrogate assisted framework that encapsulates the statistical information in the varied fidelity levels via MF Gaussian Processes with computational budget constraint so that the computationally inexpensive models are exploited more and the usage of expensive models are restricted. We demonstrate this framework for a nickel-base superalloy, CMSX-4® using three different fidelity models such as Eagar-Tsai, analytical directed energy deposition (DED) model, and Autodesk NetFabb DED model.
3:40 PM
Expanding Process Space of Laser Powder Bed Additive Manufacturing Using Alternative Scan Strategies: Elizabeth Chang-Davidson1; Nicholas Jones1; Jack Beuth1; 1Carnegie Mellon University
Metals additive manufacturing is an emerging field in manufacturing, in which one commonly used technology is laser powder bed fusion (L-PBF). Melt pool sizes in L-PBF are closely tied to printed part material properties, but are currently limited by keyholing porosity flaws or by machine limits on laser power and velocity. For carefully selected process parameters, much larger than typical melt pools were created by rapidly scanning the laser back and forth across a constant width. A systematic way to apply this technique was mapped across laser power and velocity using semi-analytical simulation software. Sample single stripes and sample cubes were printed using parameters selected to span power and velocity process space. These experimental results were used to calibrate the simulations and demonstrate viability of the technique. This systematically applicable technique increases the range of melt pool sizes and therefore range of material properties possible to print using L-PBF machines.
4:00 PM
Design Optimization for Residual Stress in Complex Low-density Support Regions: Kevin Glunt1; Shawn Hinnebusch1; Owen Hildreth2; Wen Dong1; Xuan Liang1; Florian Dugast1; Albert To1; 1University of Pittsburgh; 2Colorado School of Mines
Support structure can be built using different design parameters to ensure easier post processing removal in laser powder bed fusion (LPBF) metal components. The objective is to add additional features to a numerical design optimization method using a modified inherent strain method that quickly predicts the stress and deformation of the support and part. A projection scheme is used which takes the minimum support required and maps the domain of the support structure. The density field is used with support lattice topology optimization to minimize the mass subjected to a yield stress constraint. This will ensure the structural integrity of the reduced support region and reduce the overall chance of cracking by limiting the residual stress. This method is effective in predicting the optimized geometry for low density support regions. Numerical and experimental results prove the successful printing of multifaceted geometries with complex support structures using the proposed method.