Connect and transact with thousands of top North American companies today
Welcome to Thomas Insights — every day, we publish the latest news and analysis to keep our readers up to date on what’s happening in industry. Sign up here to get the day’s top stories delivered straight to your inbox.
As the use of artificial intelligence (AI) grows in industry, so does the number of possible applications. AI’s ability to manipulate datasets and reveal interesting correlations between data patterns — which is impossible to achieve with human intelligence — allows practitioners to utilize AI beyond novelty tools.
But innovation comes in waves: not only are we seeing the emergence of AI, but the electric vehicle (EV) market is booming, renewable energy adoption is expanding, and other major advancements are transforming industry.
The commonality between these advancements is a strong reliance on material science. Metals with precise physical, chemical, and scientific properties are in demand, yet our current material selection needs to keep up with the unique blends of characteristics needed for modern applications.
This is where AI comes in. Datasets of material information can now be leveraged to augment our existing supply and variety of engineering material options. When using traditional trial-and-error, it takes a lot of time to formulate new metal blends and even longer to find economically-viable mine locations.
AI can do this work much faster than any human could. This creates more opportunities to discover more alloy blends and bolster the supply of key base elements like nickel, cobalt, and lithium.
Though still evolving, AI has already proven successful in accomplishing this task. Below, we explore some of the essential applications of AI when hunting for new material blends and sources.
One of the most interesting applications of AI is in geology. By collecting data points about various geologic features and compositions, scientists may be able to pinpoint vast ore deposits without picking up a shovel.
KoBold Metals, an AI company that aims to discover mineral veins using data systems and predictive modeling, is already putting this to the test.
In an effort to keep to international environmental benchmarks, KoBold estimates that a $12 trillion supply gap exists in cobalt, lithium, copper, and nickel reserves. To meet the estimated demand for battery technology, which is vital for both EV and renewable energy applications, KoBold has implemented AI-powered geologic discovery in its TerraShedSM data system and Machine Prospector® predictive model library.
KoBold explorers first collect locational information such as magnetic, gravitational, radiometric, electromagnetic, acoustic, particle physics, and spectral data, before adding it to its TerraShedSM data system, which has over 100 years’ worth of global geoscience data. The company’s Machine Prospector® then scours the dataset and tags potential areas of high ore concentration, giving explorers a more targeted approach for surveying, drilling, and deposit discovery.
Currently, KoBold is working on over 30 projects. They bring AI experts, mine owners, and small exploration companies together to accelerate the discovery of key manufacturing elements essential in the search for out-of-sight resources.
While the list of engineering materials is growing steadily, there is still room for improvement. AI is helping close production gaps by shortening the discovery time of new material blends. In an effort to keep up with technological innovation, scientists are pushing the boundaries of metallic alloys using AI and have already found some success.
In a true AI approach, scientists use databases of inter-atomic data to compile a theoretical library of material blends.
M3GNet, the algorithm created by nanoengineers at the University of California San Deigo, predicts both existing and new materials based on structural information, such as the energies and forces throughout a collection of atoms. M3GNet was trained on a decade’s worth of material data and then used to create matterverse.ai, a digital library of over 31 million potential materials with accompanying material properties. These digital materials can be filtered based on desired properties such as yield strength, elastic modulus, and any other user specification, and M3GNet will provide a variety of potential formulations that may meet those criteria. Currently, one million of these materials are expected to be stable. However, more information will be needed before these computational estimations prove their worth.
Much like Google’s AI-powered AlphaFold, which predicts protein structure, M3GNet forecasts material structure by implementing graph neural networks and a deep learning architecture to mix and match all elements of the periodic table. M3GNet is currently open source, allowing the larger AI community to refine and troubleshoot the program.
Most alloys are made with a principal metal in mind, such as aluminum alloys that are primarily aluminum (>85% by mass). However, there are theoretically infinite alloys if we consider elements with more than one principal metal or those that have relatively equal proportions of five or more different metals. These multi-metal alloys are sometimes known as “high-entropy alloys,” or HEAs. With much more to learn about these alloys, there is the opportunity for additional experiments to discover precision-engineered properties.
The current process of alloy discovery is quite slow, typically involving a tedious research process, a lot of trial-and-error, and advanced metallurgical methods. This process will change with the advent of AI tools. With the right data, AI can perform bulk analyses of a large range of compositions and offer a condensed list of potential alloy blends targeted to a set of specifications.
Initiatives are already using AI in this way: the National Institute of Standards and Technology (NIST) has developed the Closed-Loop Autonomous system for Materials Exploration and Optimization (CAMEO). CAMEO helps skip redundant experiments through its self-learning algorithms.
As its name suggests, CAMEO uses closed-loop machine learning to run a virtual experiment, collect the data, analyze the results, and feed information into the loop to devise more targeted experiments. By connecting this system to a library of prior experiments and materials, researchers sped up the discovery of a new germanium-antimony-tellurium alloy from an estimated 90 hours to just under 10.
This approach is not only useful for brand-new material discovery but also for refining existing alloys. For example, machine learning process optimization has successfully been used to create an optimized 7xxx series aluminum alloy. For reference, 7075 aluminum is a common 7xxx series alloy used in aerospace applications for its impressive fatigue strength and resistance to temperature changes, but can still be augmented further with additional alloying elements.
Researchers implemented an ML adaptive design loop to discover, model, fabricate, and evaluate a 7xxx aluminum alloy with a superior 952 MPa ultimate tensile strength and a 6.3% elongation at break, which showed after casting to follow the expected results closely. This alloy was created using traditional manufacturing methods, meaning that, when perfected, the blend can be easily deployed into industry for various applications.
Another benefit of this approach was that it unveiled a unique nanoscale structure between alloying elements. The structure, an unusual Al8Cu4Y lattice, provides insight into why certain aluminum alloys are stronger than others and will inform new high-strength alloy formulations.
At an atomic level, metals are crystalline or highly ordered in a geometrical lattice. Glass or plastic, on the other hand, is amorphous, meaning their atoms are disordered and do not form large-scale coherent structures. While much is understood about metals, glass, and plastic, researchers are now exploring a new class of alloys, sometimes found as “metallic glass,” “glassy alloys,” or, more generally, “amorphous alloys.”
In these blends, the metal atoms are disordered like glass and are technically a supercooled liquid that retains the good qualities of both material classes. Metallic glasses have been shown to be extremely strong at lower temperatures, highly flexible at higher temperatures, and possess impressive chemical and physical characteristics. Initial discoveries have been used to make corrosion-resistant coatings and injection-moldable metal parts, but AI is offering even greater insight into potential applications.
Researchers at Standford’s SLAC National Accelerator Laboratory, alongside NIST and Northwestern University, are implementing AI to reduce the time needed to explore the millions of potential metallic glasses. Using the Stanford Synchrotron Radiation Lightsource (SSRL), a particle accelerator that examines atomic and molecular structures, researchers are utilizing machine learning to predict new experiments and target useful metallic glass types.
These predictions are informed by an existing dataset of experimental results. In turn, it’s accelerating the discovery of new metallic glasses, allowing scientists to make and screen over 20,000 combinations in a single year.
The hope is that metallic glasses made from cheap, easily-sourced materials will contribute to the advancement of key battery technologies such as solid-state batteries and hyper-elastic materials needed for larger wind turbines, hydroelectric plants, solar arrays, and a variety of other electric applications currently restricted by available materials.
The current goal in material science is to develop a cheap and effective solid-state electrolyte for battery technology. Current batteries require aqueous electrolyte solutions to function, which requires caustic and flammable liquids that increase fragility and decrease maximum capacity. Creating a solid electrolyte could significantly reduce battery weight and increase capacity and safety, revolutionizing the capabilities of energy storage.
While progress is slow in the discovery of solid-state electrolytes, AI tools have already sped up the process.
Research at the University of Liverpool designed an unsupervised machine learning tool to scour chemical and physical traits of existing materials and then rank potential new blends for investigation. The machine learning system manages the tedious pattern recognition needed to predict new potential formulations, and human experts use these insights to perform targeted experiments. Using this ML-human expert workflow, researchers have already discovered Li3.3SnS3.3Cl0.7, a lithium solid electrolyte with advantageous material properties for solid-state battery applications.
The most promising aspect of this unsupervised system is that both good and bad results are fed back into the algorithm, further refining the prediction capabilities and informing new novel combinations of materials.
AI, machine learning, and other innovative tools will be increasingly vital for the discovery of metals for EVs, renewable energy, and the unforeseen applications of the future.
The ability to make informed, targeted decisions surrounding experimentation will speed up the discovery of geologic deposits as well as novel metal formulations. Though still being deployed, AI tools in metallurgy have already proven invaluable to researchers and may provide a whole new catalog of metals with multifaceted properties.
For now, we have to wait for systems to mature and be optimized for real-world applications.
Image Credit: Cat Us / Shutterstock.com
TechnologyAI-powered Smart Glasses Provide Subtitles for Those Who Are Deaf
TechnologyDrone Use in Agriculture Soaring to New Heights
TechnologyHow Deep Under the Ocean Have Humans Explored?
TechnologyHow AI Is Uncovering Essential Metals for EVs, Green Innovation
TechnologyFrom Ameca to Sophia, You Won't Believe How Lifelike Some Robots Are
Select From Over 500,000 Industrial Suppliers
Find and evaluate OEMs, Custom Manufacturers, Service Companies and Distributors.
Stay up to date on industry news and trends, product announcements and the latest innovations.
Find materials, components, equipment, MRO supplies and more.
Download 2D & 3D CAD Models
10+ million models from leading OEMs, compatible with all major CAD software systems.
Copyright© 2023 Thomas Publishing Company. All Rights Reserved. See Terms and Conditions, Privacy Statement and California Do Not Track Notice. Website Last Modified March 15, 2023. Thomas Register® and Thomas Regional® are part of Thomasnet.com. Thomasnet Is A Registered Trademark Of Thomas Publishing Company.