ICME 2023: AI/ML: Properties II
Program Organizers: Charles Ward, AFRL/RXM; Heather Murdoch, U.S. Army Research Laboratory
Monday 3:20 PM
May 22, 2023
Room: Boca I-III
Location: Caribe Royale
3:20 PM Cancelled
Quantitative Precipitate Analysis of an Age-hardenable Aluminium Alloy Using a Deep Learning Approach: Ghezal Ahmad Jan Zia1; 1BAM
Mechanical properties of metals and their alloys are strongly governed by their microstructure. The nanometer-sized precipitates in hardenable wrought aluminium alloys, which can be controlled by heat treatment, act as obstacles to dislocation movement within the material and are critical to the mechanical performance of the component, in this case, a radial compressor wheel of a ship’s engine. Deep Learning based TEM image analysis is essential for the study to investigate the microstructural changes (precipitation coarsening) that occur as a result of aging at elevated temperatures.
3:40 PM
Machine Intuitive Development of Army Steels - MIDAS: Heather Murdoch1; Levi McClenny1; Benjamin Szajewski1; Daniel Field1; Berend Rinderspacher1; Mulugeta Haile1; Krista Limmer1; Andrew Garza2; 1U.S. Army Research Laboratory; 2UC Merced
Recent advances in data science and high-throughput materials simulations are being evaluated to accelerate advanced steel alloy development. Here we use machine learning (ML) to take advantage of the large amounts of historic data available for martensitic steels. A series of methods using varying amounts of data are used to develop predictive ML models. Our hypothesis is that the composition and processing variables are insufficient to inform a robust machine learning model, and the incorporation of CALPHAD inputs will provide the additional information required to inform the models and make a more robust framework that is generalizable to other systems. These intermediate variables are synthetic microstructures and thermodynamic properties generated using high-throughput CALPHAD simulations. We present a preliminary integrated machine learning and alloy design workflow based on steel hardness and toughness response during tempering. Model uncertainty and databasing challenges will also be discussed.
4:00 PM
Analysis of AA6061 Cladding Diffusion Bonding Quality for the U-10Mo Monolithic Fuel Using Multi-fidelity Machine Learning Surrogate: Yucheng Fu1; Rajib Kalsar1; Taylor Mason1; Zhijie Xu1; Kriston Brooks1; Ayoub Soulami1; Vineet Joshi1; 1Pacific Northwest National Laboratory
To reduce nuclear proliferation, low-enriched U-10Mo alloy has been identified as a promising fuel candidate for United States high-performance research reactors. During fabrication, the fuel will be encapsulated in the aluminum alloy 6061 (AA6061) cladding, to prevent fuel corrosion and fission product release. The cladding was diffusion bonded using the hot isostatic pressing (HIP), which promotes a homogeneous AA6061/AA6061 bonding interface. To reduce the high experimental cost and efficiently optimize the diffusion bonding process, a multi-fidelity Gaussian process surrogate was developed to predict the aluminum cladding bond strength. This machine learning surrogate leverages the high-fidelity experimental data with the low-fidelity numerical model to maximize the bond strength prediction accuracy. Sensitivity analysis was followed to identify the influential HIP parameters. It was found that the interface Mg2Al2O5 particles were closely related to the bond strength and the temperature was suggested as the most dominant factor in determining the bonding quality.
4:20 PM Cancelled
Generative Alloy Design Based Framework for In-silico Design of HSLA Steels: Akash Bhattacharjee1; KV Vamsi1; Bilal Muhammed1; Amol Joshi1; Gerald Tennyson1; BP Gautham1; 1TCS Research, Tata Consultancy Services Limited
Advanced high strength steels are generally designed to meet target mechanical properties either by modifying the alloy composition or engineering the microstructure via processing. Despite the advances in computational tools available to model the process chain, exploring the vast design space is extremely challenging for achieving the target properties. Recently, generative design approaches that are used in product design are presently being envisaged for alloy design. A generative design based framework is developed for active exploration and optimization of alloy chemistry and process parameters for an integrated computational process chain consisting of casting - reheating - hot rolling - cooling (ROT+coiling) - properties in the design of hot rolled HSLA steel for desired properties through thin slab as well as conventional casting and rolling routes. The details of the novel approach, constraints, and limitations will be presented.