Scientists at the Ames National Laboratory in the U.S. have developed a systematic path to the discovery of new approaches to building permanent magnets without relying on rare-earth elements. The approach combines physics-based modeling and high-throughput simulations with reasoning-based artificial intelligence (AI) tools to guide the discovery process even before materials are made in the laboratory.\u00a0<\/p>\n
Permanent magnets are a crucial part of our everyday lives. From aiding data storage to helping us drive electric vehicles (EVs) to enabling high-quality medical imaging, permanent magnets are used everywhere.\u00a0<\/p>\n
High-performance permanent magnets made with rare-earth elements are used in defense applications and energy generation. However, the U.S. is highly dependent on other countries to refine its rare earth elements, which increases costs and security risks. Scientists at the Ames National Laboratory have therefore been tasked with developing permanent magnets that do not use rare-earth elements.\u00a0<\/p>\n
AI, AI everywhere<\/p>\n
At a time when AI adoption is soaring and no organization wants to remain behind, another story about AI workflows is not a major surprise. However, for an AI model to perform perfectly, it must be trained on the right data.\u00a0<\/p>\n
For a general use case, an AI model can be trained on general data. However, for an AI that needs to excel in materials science, a research team led by Prashant Singh, a scientist at Ames, ensured that models were trained on experimentally measured and scientifically calculated material properties. Only this can ensure that the predictions the AI model is making are grounded in real-world behavior.\u00a0<\/p>\n
\u201cUnderstanding the physics of materials is important to include in AI frameworks when you are trying to design new materials. If you just use the data to train your models, you are going to get only the predictions within the range of information you have,\u201d said Singh in a press release.\u00a0<\/p>\n
\u201cBut once you understand the physics of what controls specific properties, then you and your agentic tools<\/a> or AI frameworks can search arbitrary material space,\u201d added Singh.\u00a0<\/p>\n Addressing the full pipeline<\/p>\n The advantage of using such an AI-based approach is that it uses a material\u2019s atomic structure and electronic behavior to determine its properties. Magnetization strength, resistance to demagnetization, energy storage capacity, and the\u00a0permanent magnet\u2019s<\/a>\u00a0behavior at high temperatures are major parameters that determine its usefulness.\u00a0<\/p>\n Using an AI model that accounts for this allows researchers to identify the most promising material candidates and arrive at likely outcomes through computations rather than real-world iterations.\u00a0<\/p>\n \u201cAmes Lab\u2019s strength comes from its deep expertise and a long history of data in the magnet space that no other institution has,\u201d elaborated Singh in the press release.\u00a0<\/p>\n \u201cIn any material design problem, you need to know how combining two elements will change their performance before you ever run an experiment. We have been building both theoretical and analytical tools to answer that question, and we are now bringing AI into that process to make it faster and broader in scope.\u201d<\/p>\n However, the researchers did not just stop there. The AI tools also take into account the availability and cost of the material. Since the frailty of supply chains has been repeatedly exposed in the past decade, the AI models also factor in these conditions during the discovery process itself to ensure that the material developed is practically possible and can be scaled.\u00a0<\/p>\n In doing so, the researchers have addressed the complete pipeline from discovery to industrial availability as the U.S. looks to reduce its dependence on other countries for critical components.\u00a0<\/p>\n