ADVANCED MATERIALS & PROCESSES | MAY/JUNE 2024 19 simulations from Materials Project and Google DeepMind and also used natural- language models trained on literature data to extract the synthesis recipes. Then active learning was used to optimize the recipes based on thermodynamics. They proposed 58 recipes including a variety of oxides and phosphates and could realize 41 of them over 17 days of continuous operation of robots. The autonomous lab for materials discovery via artificial intelligence platform provided a 75% success rate in discovery and thus presents a new efficient pathway to accelerate and automate the materials design process. Case Study 2. As the second successful recent example for auto- mating materials discovery, Boiko et al. developed a transformer-based large language model, Coscientist, to design, plan, and conduct complicated experiments in an automatic manner[17]. Their artificial intelligence system, which is driven by GPT-4, incorporates large language models empowered by searching the documents on the internet, executing code, and performing experiments automatically. Coscientist showed promising acceleration in research in six tasks: They could (1) use available public data and design the chemical synthesis of known compounds; (2) efficiently search and explore significant hardware documentation; (3) execute high-level commands in a cloud laboratory by using the documentation; (4) control precisely the instruments that handle liquids giving low-level instructions; (5) handle multi-field, multi-modal data sources and perform complicated scientific tasks; and (6) finally solve optimization problems dynamically while they were analyzing collected experimental data. Development of these intelligent multiagent models has potential to greatly accelerate new material discoveries. CONCLUSION AI and ML technologies are transforming our daily lives and countless industries around the world. As shown from this small sampling of the innumerable case studies in the open literature, materials science is being transformed as well. Tedious and time-intensive tasks such as metallographic studies and failure analysis are being automated. The design of novel materials and manufacturing processes are being dramatically accelerated. Robotic labs for completely handsfree synthesis and characterization of materials are being developed. These examples are the current work, and every year brings significant advancements. New algorithms bring new capabilities, new data digitization efforts enable new materials classes and properties to be analyzed, and new hardware automates new lab tasks. One recent prominent example of an AI/ML technology suddenly appearing and transforming industries is ChatGPT and other large language models (LLMs), which, in only a year, has brought significant advancements in materials science digital lab assistants and data digitization. Also, significant progress has been made over the past years to digitize historical materials data in organizations around the world, including ASM International, which will add to the momentum of AI/ML adoption in materials industries. There are challenges to this adoption. Materials science is a wide and diverse field with many unique problems, data types, and interests. Metal alloys, in particular, are complex physical systems that are difficult to study at scale, with relatively small datasets and difficult-to-automate experiments. However, many scientists, engineers, and entrepreneurs are chipping away at solutions to these challenges. So, it is only a matter of time until AI/ML completely transforms the materials and metals industries. As a professional society, as an industry, and as a field, we should embrace the opportunities that AI/ML offers and work to build the next generation of materials science and engineering. ~AM&P For more information: Joshua Stuckner, materials informatics scientist, NASA Glenn Research Center, 21000 Brookpark Rd., Cleveland, OH 44135, 216.433.4000, joshua.stuckner@nasa.gov. References 1. J. Stuckner, B. Harder, and T.M. Smith, Microstructure Segmentation with Deep Learning Encoders Pretrained on a Large Microscopy Dataset, npj Comput. Mater., 8(1), 2022, doi: 10.1038/s41524-022-00878-5. 2. T.M. Smith, et al., Utilizing Local Phase Transformation Strengthening for Nickel-base Superalloys, Commun. Mater., 2(1), p 1–9, 2021. 3. B.J. Harder, et al., Steam Oxidation Performance of Yb2Si2O7 Environmental Barrier Coatings Exposed to CMAS, J. Eur. Ceram. Soc., 44(4), p 2486–2498, 2024. 4. H. Baumgartl, et al., A Deep Learning-based Model for Defect Detection in Laser-powder Bed Fusion using in-situ Thermographic Monitoring,” Prog. Addit. Manuf., 5(3), p 277–285, 2020. 5. H. Elwarfalli, et al., In situ Process Monitoring for Laser-powder Bed Fusion using Convolutional Neural Networks and Infrared Tomography, in 2019 IEEE National Aerospace and Electronics Conference (NAECON), p 323– 327, 2019. 6. A. Caggiano, et al., Machine Learning-based Image Processing for GET ENGAGED, GET INVOLVED, GET CONNECTED The ASM Materials Informatics Committee is an active group of ASM members working together to drive industry-wide efforts in implementing artificial intelligence and machine learning methods to accelerate the design, discovery, and deployment of new materials, processes, and analysis techniques. Committee projects include developing a new ASM Handbook volume and organizing IMAT conference programming. Members with similar interests are welcome to join. For more information, contact committee chair Joshua Stuckner at josh.stuckner@gmail.com.
RkJQdWJsaXNoZXIy MTYyMzk3NQ==