First World Congress on Artificial Intelligence in Materials and Manufacturing (AIM 2022): Artificial Intelligence in Specific Manufacturing Processes I
Program Organizers: Taylor Sparks, University of Utah; Michael Dawson-Haggerty, Kerfed, Inc.; Elizabeth Holm, University of Michigan; Jin Kocsis, Purdue University; Adam Kopper, Mercury Marine; Benji Maruyama; James Warren, National Institute of Standards and Technology
Monday 1:30 PM
April 4, 2022
Room: William Penn Ballroom
Location: Omni William Penn Hotel
Session Chair: Joshua Fody, NASA Langley Research Center
1:30 PM
Machine-Learning-Guided Automated Training for Adaptive Printing: Automating Intuition for Unintuitive Toolpaths and Geometries: Erick Braham1; Marshall Johnson2; James Hardin1; Surya Kaladindi2; 1Air Force Research Lab; 2Georgia Institute of Technology
Additive manufacturing has enabled advancement in high-mix low-volume manufacturing and rapid prototyping of materials and geometries. Selecting process parameters for new materials, geometries, applications, or hardware configurations in additive systems currently relies on intuition-based trial and error. However, the difference between machines and humans limit the power of this intuition. Can machines develop their own “intuition” and how much data does such a process require? To answer these questions, we are developing an automated rapid parameterization protocol. Specifically, this work proposes a machine-learning-driven automated parameterization scheme demonstrated with the direct ink write (DIW) printing of freestanding structures. This case study examines several factors crucial to the success of automated parameterization including efficiency, design space considerations, use of prior data, and physical limitations. Ideally, automated training of machine intuition will enable more agile manufacturing, high-throughput screening of materials and geometries, and more efficient collaboration across systems.
1:50 PM
Teaching Printers How to Print: from Closed-Loop Control to Integrating AI and Cloud Computing into Additive Manufacture: James Hardin1; J. Dan Berrigan1; 1Air Force Research Lab / RXMS
Additive manufacturing (AM) is a naturally digital manufacturing process driven by complex physics yielding massive process parameter spaces. We are using direct-ink write (DIW) AM to explore how the latest digital engineering tools can be brought to bear on a high-mix low-volume manufacturing process. This presentation will start with our work on printing high-precision multi-material AM of electronics using closed-loop feedback. In response to the challenges of this work, we developed an agile approach to automating AM and applied it to DIW by building AI tools into closed-loop control. Finally, we will demonstrate progress toward integrating AM with cloud computing resources to alleviate the computational and data burdens of digital engineering tools.
2:10 PM
Now On-Demand Only - Defect Minimization in Additive Manufacturing Through a Customized High-Throughput Experimental Methodology and Machine Learning Approach: Baldur Steingrimsson1; Benjamin Adam; Michael Gao2; Graham Tewksbury3; E-Wen Huang4; Peter Liaw5; 1Imagars LLC; 2National Energy Technology Laboratory; 3Oregon State University; 4National Yang Ming Chiao Tung University; 5University of Tennessee
This presentation addresses defect mitigation in additively manufactured (AM) parts, in particular hot cracking, through a customized approach combining high-throughput experimental methodology with machine learning (HTEML). Nickel-base superalloys behave quite different, when fabricated with AM compared to the manufacturing methods that they were originally designed for. Defects in AM superalloy parts can partially be attributed to the fact that these alloys were not specifically designed for AM. Cracking in AM parts represents particularly deleterious failure mode that can result in unexpected catastrophic failures. To address this problem, we target high-strength Ni-base superalloys with high γ’ fraction, but prone to hot cracking, such as CM247. One innovative aspect of the presentation relates to trade-off between high throughput (HT) and measurement accuracy in HTEM screening. As opposed to sole emphasis on HT, we will investigate HT in context with the associated measurement accuracy and what one is looking to measure or detect.
2:30 PM Break
3:00 PM
Machine Learning-Based Thermal Analysis for Laser-Based Powder-Bed Fusion Additive Manufacturing: Raeita Mehraban Teymouri1; Ardalan Sofi1; Chinnapat Panwisawas2; Bahram Ravani1; 1University of California, Davis; 2University of Leicester
Laser-based powder-bed fusion additive manufacturing (L-PBF AM) processes involve complex micro time scale and length scale phenomena. Computationally expensive high-fidelity thermo-fluid simulations and efficient yet over-simplified conduction-only models are two main categories of existing thermal modeling techniques for L-PBF AM processes. However, there is a pressing need to develop a Reduced-Order Model (ROM) that would account for lower length scale phenomena while allowing to scale up to larger domains for L-PBF AM applications. This ROM can then conveniently be used to develop a dataset to train a Machine Learning (ML) model capable of performing an accurate near-real-time evaluation of the temperature field of additively manufactured parts at any stage of the manufacturing process. Moreover, one of the major objectives of this work is developing process maps using key process variables to understand the composition-process relationships and determine the optimum process window for L-PBF AM of widely used commercial alloys.
3:20 PM
High-Throughput Development of MPEAs by Combined Approach of Additive Manufacturing and Machine Learning Methods: Phalgun Nelaturu1; Jason Hattrick-Simpers2; Michael Moorehead1; Santanu Chauduri3; Adrien Couet1; Dan Thoma1; 1University of Wisconsin; 2University of Toronto; 3Argonne National Laboratory
This work details a framework for material discovery by combining high-throughput experimental and computational techniques. Additive manufacturing via directed energy deposition was employed as a high-throughput technique to synthesize alloys in the Cr-Fe-Mn-Mo-Ni system. More than 200 discrete, bulk samples of unique alloy compositions were synthesized, exploring a vast compositional space. Tight compositional control within ±10 at%, and low porosity and unmelted powder fractions of <0.3% were achieved. The rapid synthesis combined with rapid heat treatment, characterization, and nano- and micro-hardness measurements enabled high-throughput evaluation of these materials. The large dataset of experimentally measured properties was used to develop a predictive hardening model using an active machine learning algorithm coupled with thermodynamic modeling. The outcome of this effort was the discovery of a learned parameter, deltaLP, that was representative of the lattice distortion in these alloys. deltaLP was trained using the alloy compositions and was highly predictive of hardness.