AI/Data informatics: Design of Structural Materials: AI/ML for Design of Structural Alloys & Additively Manufactured Materials
Sponsored by: TMS Materials Processing and Manufacturing Division, TMS Structural Materials Division, TMS: Mechanical Behavior of Materials Committee, TMS: Computational Materials Science and Engineering Committee, TMS: Integrated Computational Materials Engineering Committee
Program Organizers: Jennifer Carter, Case Western Reserve University; Amit Verma, Lawrence Livermore National Laboratory; Natasha Vermaak, Lehigh University; Jonathan Zimmerman, Sandia National Laboratories; Darren Pagan, Pennsylvania State University; Chris Haines, Ccdc Army Research Laboratory; Judith Brown, Sandia National Laboratories
Tuesday 2:00 PM
March 16, 2021
Room: RM 32
Location: TMS2021 Virtual
2:00 PM Invited
Zoning Processing Spaces for Additive Manufacturing: Applications for Inverse Design: Sean Donegan1; Edwin Schwalbach1; Matthew Krug1; 1Air Force Research Laboratory
Metals additive manufacturing technologies, such as powder bed fusion, allow for the production of geometrically complicated components, offering significant freedom to a component designer. Although this geometric freedom can lead to novel capabilities, it also causes significant difficulty in understanding the local processing state of the material. This complexity is driven by dependence between local geometric features and the path of the energy source, leading to variations in thermal state across a component. Given a target for a kind of local thermal profile, determining “optimal” paths for an energy source remains challenging due to the overwhelming size of the geometry space. We demonstrate that zoning can be used to make this problem tractable by hierarchically grouping regions of similar desired processing state together. We showcase how this approach can be used to construct targets for scan path optimization.
2:30 PM
High-throughput Alloy Design via Additive Manufacturing: Olivia Dippo1; Kevin Kaufmann1; Grant Schrader1; Kenneth Vecchio1; 1University of California San Diego
Establishing a commercially viable alloy from lab-scale alloy development research typically takes decades, in part due to the typical one-sample-at-a-time approach of sample synthesis, preparation, characterization, and analysis. To accelerate alloy development, we have designed an integrated, high-throughput method focused on parallelizing, miniaturizing, and automating each of these steps. In this method, alloys are selected using a CALPHAD-based alloy design algorithm. Then, test samples are built using laser metal deposition in 15-sample libraries with a unique geometry that facilitates rapid characterization. Samples can then be automatically prepared and characterized by XRD, EDS, and EBSD. These analyses, combined with various material property assessments, can be coupled with machine learning techniques to accelerate future materials analysis and guide subsequent composition and processing decisions. This method is applied to functionally graded metallic materials as well as discrete libraries of materials to better understand alloy development and improve predictive capabilities.
2:50 PM
Alloy Design for Additive Manufacturing: Mariam Assi1; Julien Favre1; Anna Fraczkiewicz1; Franck Tancret2; 1Mines Saint-Etienne, Univ Lyon, LGF - UMR 5307 CNRS/ Centre SMS; 2Université de Nantes, Institut des Matériaux Jean Rouxel (IMN), Polytech Nantes, BP 50609
In recent years, the development of innovatory processes like “additive manufacturing” (AM) has opened new fields in modern metallurgy. These new processes are also at the origin of development of new processable or printable alloys. The aim of this study is to propose a computational method based on Bayesian machine learning (ML) algorithms combined with a thermodynamic approach (CALPHAD) and integrated in a multi-objective genetic algorithm (GA) to design AM-optimized alloys. In this context, several material criteria influencing the defects commonly observed in AM-fabricated parts, such as solidification cracking, porosity, residual stresses and distortions, were taken into account. The proposed model uses data sets constructed from published literature and industrial material datasheets. Its application to design improved grades of austenitic stainless steels will be shown and discussed.
3:10 PM Invited
Multi-objective Lattice Optimization Using an Efficient Neural Network Approach: Anthony Garland1; Ben White1; Brad Boyce1; Ryan Alberdi1; 1Sandia National Labs
Additively manufactured lattice metamaterials offer the ability to design effective material properties tailored to meet specific engineering requirements. However, the optimal design of such lattices is currently limited in a number of ways: (1) most topology optimization methods are restricted to linear phenomena such as elasticity, (2) explicit finite element representation and contribution toward the objective of every feature is computationally expensive or intractable, (3) intermediate densities between material and void create computational instabilities, and (4) the optimization rarely takes into consideration manufacturing limits. To address these challenges, a convolutional neural network (NN) was trained on thousands of simulations of random and rational unit cell designs. The NN is differentiable and serves as an efficient surrogate model in place of expensive finite element simulations. In this presentation, we explore gradient optimization of multiple competing objectives on a pareto front, including explicit dynamic shock response, stiffness, and manufacturability constraints.
3:40 PM
Design of Ti-Al-Cr-V Alloys for Maximum Thermodynamic Stability: Rajesh Jha; George Dulikravich1; 1Florida International University
Chemistries of Ti-Al-Cr-V alloys were computationally Pareto-optimized for simultaneously maximizing the Young’s modulus and minimizing density for a range of temperatures. Compositions at different temperatures of these alloys were then analyzed for phase stability in order to generate new data for compositions and volume fractions of stable phases at various temperatures. This resulted in a large dataset where a lot of data were still missing as all the phases are not stable at a given temperature for all the compositions. The concept of Self Organizing Maps was then applied to correlate alloy compositions, stabilities of desired phases at various temperatures, associated Young’s moduli and densities, and the effect of the composition of phases on these properties. This work should help alloy designers to determine the required chemical composition of a new alloy with reference to the temperature of application and see the effect on stable phases and properties of alloys.
4:00 PM
Prediction of the Mechanical Properties of Aluminum Alloy Using Bayesian Learning for Neural Networks: Shimpei Takemoto1; Kenji Nagata2; Takeshi Kaneshita1; Yoshishige Okuno1; Katsuki Okuno1; Masamichi Kitano1; Junya Inoue3; Manabu Enoki3; 1Showa Denko K.K.; 2National Institute for Materials Science; 3The University of Tokyo
Understanding the process-structure-property relationship is one of the goals of computational materials design. In this study, we analyze the strengthening mechanism of 2000 series aluminum alloy using neural networks. We have constructed a neural network for the simultaneous prediction of multiple mechanical properties, including ultimate tensile strength, tensile yield strength, and elongation. Replica-exchange Monte Carlo method, an extended Markov chain Monte Carlo (MCMC) method, has been applied for the Bayesian estimation of the optimal neural network architecture and hyperparameters. The obtained neural network, combined with the thermodynamic analysis using the CALPHAD method implemented in the Thermo-Calc software, enables us to identify a dominant combination of additive elements and thermal processing for strengthening alloys. We have also addressed an inverse problem for optimizing the process parameters for a set of desired properties. The approach we propose will accelerate the design of high strength alloys for high-temperature applications.
4:20 PM
Machine Learning Assisted Exploration of FeCoCrNi Based Nanocrystal-amorphous Dual-phase Alloys: Yi Yao1; Xiaobing Hu2; Xiaoxiang Yu2; Jiaqi Gong1; Feng Yan1; Lin Li1; 1The University of Alabama; 2Northwestern University
Alloys with nanocrystal-amorphous dual-phase structure are elaborately designed to promote strength and ductility synergy. The lack of a physical understanding of their formation mechanism, however, makes the alloy design exceedingly laborious. In this study, we utilize machine learning (ML) models with three algorithms (i.e. artificial neural network, logistic regression, and support vector machine) to search multicomponent nanocrystal-amorphous dual-phase alloys. The models are trained on electronic, atomic, thermodynamic features of 5527 alloys. The cross-validation of the three ML algorithms demonstrates the artificial neural network has the highest performance, the area under curve is 0.9889. The model is then used to explore FeCoCrNi based alloy system, and FeCoCrNi-Mo alloys have been predicted to form the dual-phase structure, and then validated experimentally. We further perform a detailed analysis of data features that dominate the dual-phase formation for different synthesis methods, gaining insights into the controlling features that will accelerate the dual-phase alloy discovery.
4:40 PM
Topology Optimization for Design of Stress-dependent Material Properties: Justin Unger1; Matthew Vaughn1; Andrew Gaynor2; Brandon McWilliams2; James Guest1; Kevin Hemker1; 1Johns Hopkins University; 2CCDC U.S. Army Research Laboratory
Existing automated design methods frequently assume isotropic material models that do not accurately reflect experimentally observed constituent properties of additively manufactured materials, including those that exhibit tension-compression asymmetry of their elastic stress-strain response. A design optimization approach is proposed that incorporates stress-dependent constitutive material models into a topology optimization framework to mitigate specific failure modes while leveraging more realistic material-dependent stress states. Inspired by an idea in [1], this methodology utilizes a bilinear tension-compression elastic formulation based on an orthotropic stress-dependent constitutive material model. This topology optimization scheme is demonstrated on the design of maximum stiffness structures resembling lattices that satisfy minimum feature size constraints and leverage experimentally measured material property data. [1] A. Gaynor, J.K. Guest and C. Moen. Reinforced concrete force visualization and design using bilinear truss-continuum topology optimization. Journal of Structural Engineering, 139(4): 607-618, 2013.