Computational Techniques for Multi-Scale Modeling in Advanced Manufacturing: Multiscale Computational Techniques
Sponsored by: TMS Materials Processing and Manufacturing Division, TMS Extraction and Processing Division, TMS: Computational Materials Science and Engineering Committee, TMS: Process Technology and Modeling Committee
Program Organizers: Adrian Sabau, Oak Ridge National Laboratory; Anthony Rollett, Carnegie Mellon University; Laurentiu Nastac, University of Alabama; Mei Li, Ford Motor Company; Alexandra Anderson, Gopher Resource; Srujan Rokkam, Advanced Cooling Technologies, Inc.

Wednesday 8:30 AM
March 17, 2021
Room: RM 1
Location: TMS2021 Virtual

Session Chair: Adrian Sabau, Oak Ridge National Laboratory


8:30 AM  
Machine-learning Informed Design of High-strength Gradient Metals for Additive Manufacturing: S. Mohadeseh Taheri-Mousavi1; A. John Hart1; 1MIT
    The advent of additive manufacturing creates a need for computationally-efficient design approaches to determine complex relationships between local composition, microstructure, and resulting mechanical properties. In particular, if these attributes can be controlled with the dimensional and compositional fidelity needed to activate tailored strain gradient plasticity, i.e., the incompatibility of plastic strain near interfaces of the adjacent gradient components, one can achieve mechanical properties exceeding rule of mixture estimates. Here, we present a comprehensive strain gradient theory which accounts for the contribution of this incompatibility to strength and strain hardening. By calibrating the model with a bi-layered material system of copper and bronze, and by deep learning optimization, we reveal design motifs that achieve enhanced strength tailored to specific boundary conditions and load cases. Moreover, the framework is broadly applicable to meso-scale gradient systems and offers new perspectives to the discovery of next-generation structural alloys.

8:55 AM  Cancelled
A Hybrid Approach to Connecting a Low Fidelity Model to a High Fidelity Model for Efficient and Accurate Prediction of Thermal History of Large Domains in Additive Manufacturing: Christopher Katinas1; Corbin Grohol1; Yung Shin1; 1Purdue University
    Accurate prediction of thermal history during additive manufacturing is important since it affects resultant microstructure, phases, distortion and residual stresses. Metal additive manufacturing processes involve very complex physical mechanisms including laser-powder interaction, powder motion, melting and solidification, fluid motion, denudation and keyhole formation in some cases. High fidelity models to account for these complex mechanisms have been developed, but their high computational cost prohibits them from being used for predicting the temperature history of large domains of actual parts. Contrarily, simpler conduction based models have been used owing to their computational efficiency despite the lack of accuracy. This presentation will describe a hybrid approach of combining a low fidelity model with a high fidelity model using intelligent sampling and a data-driven approach such that accurate temperature fields can be obtained for large domains with requisite computational efficiency. The corresponding experimental results are shown to demonstrate the efficacy of this approach.

9:20 AM  Invited
Predicting Mechanical Performance in Additive Manufacturing Components Using Deep Learning: Kyle Johnson1; John Emery1; Demitri Maestas1; Matthew Smith1; Carianne Martinez1; Mircea Grigoriu2; 1Sandia National Laboratories; 2Cornell University
    Across a range of disciplines, Deep Learning (DL) has shown tremendous success in discovering features and patterns within input data by detecting structures and hierarchies. This talk will present results of a recent effort to utilize DL algorithms to predict microstructure-dependent mechanical performance in synthetic metal coupons representative of additively manufactured material. To train the DL network, a large database of synthetic data was developed based on physical material measurements of AlSi10Mg. First, finite element meshes of tensile specimens containing voids were generated based on statistical distributions measured through high-resolution X-ray tomography. The samples were then loaded in tension, with different pore distributions leading to strain localization in different regions over a range of peak loads. The resulting data was used to train a 3D Convolutional Neural Network to predict mechanical properties in different stress states and geometries. Progress to date will be discussed, along with challenges and future work.

10:00 AM  
Smoothed Particle Hydrodynamics based approach for 3D Modeling of Linear Friction Welding Process: Srujan Rokkam1; Quang Truong1; 1Advanced Cooling Technologies Inc
    Linear friction welding (LFW) is a solid-state joining process in which a weld between two metals is formed by combined action of frictional heating and forming force that creates a weld interface. Due to large deformation phenomena, commercially available software tools are limited to modeling of LFW in 2D using Finite Element Method (FEM) with adaptive mesh controls. In this work, we developed a meshless approach that utilizes a combination of Smoothed Particle Hydrodynamics (SPH) and FEM to obtain a physics-based model capable of capturing the thermo-mechanical behavior LFW process in 3D. The developed model is employed to simulate and investigate flash formation and burn-off distance in surrogate welds. The simulation results agreed well with FE simulation and experimental data. This work was funded by an U.S. Air Force Phase II SBIR program, Contract FA8650-19-C-5050, awarded to ACT Inc.

10:25 AM  
Synchrotron Calibrated Lagrangian Particle Tracking of Melt-pool Ejections during Laser Powder Bed Fusion: Samuel Clark1; Gongyuan Zeng2; Juergen Jakumeit2; Chu Lun Alex Leung1; Yunhui Chen1; Sebastian Marussi1; Lorna Sinclair1; Margie Olbinado3; Alexander Rack4; Peter Lee1; 1University College London; 2Access e.V.; 3Paul Scherrer Institute; 4European Synchrotron Radiation Facility
    Laser Powder Bed Fusion(LPBF) is a turbulent process where a high-energy density laser beam induces a high-velocity metal vapour plume which can denude powder particles from the powder bed and induce the stochastic ejection of large spatter droplets. The vapour jet imparts ballistic trajectories of the large entrained spatter droplets, sending far across the powder bed. These large particles can disrupt subsequent laser tracks and cause uneven powder spreading of the next layer. Using fast synchrotron imaging we capture the initial spatter velocity, enabling the full particle trajectory to be calculated. We supplement these results with additional insights into the physics of the phenomena using a multi-physics model of the vapour plume coupled with Lagrangian Particle Tracking(LPT). The results reveal new understanding of spatter trajectories in relation to where they are generated in the keyhole and how this might be mitigated to minimise spatter contamination of the powder bed.