See the article below by QuesTek president Jason Sebastian and director of solutions architecture Keith Fritz as it originally ran in Industry Today, and stay tuned for a more detailed analysis of how QuesTek applies AI technology in materials engineering.
When artificial intelligence (AI) burst into the public conversation earlier this year, every industry was faced with a question: How will AI change your business? In the field of materials engineering, the specific question was, “How can recent advances in AI technology accelerate the development and qualification of materials?”
The answer is that it already has. For about a decade, AI and machine learning (ML) have been applied to materials engineering.
Across industries such as aerospace, automotive, medical devices and consumer electronics, an increasing number of manufacturers are discovering that making materials engineering part of their design process is needed to unlock greater performance. Companies that design novel materials — or that refine their materials selection and processing decisions using digital tools — are able to differentiate from the competition by shifting the bounds of what is possible in their industry.
AI has already helped push those boundaries, and recent advances in AI technology make it increasingly complementary to the leading approach to materials science: physics-based modeling.
Physics-based modeling of novel materials
Physics-based modeling is a materials engineering technique that relies on fundamental physical laws and equations that describe the performance of materials. This approach requires a deep knowledge of the underlying physics of the material system being studied. What it doesn’t require is a huge amount of data.
Physics-based modeling can be used, along with a few critical data points, to determine which variables have the greatest impact on a material’s behavior. From that, materials engineers can create a modeling framework to accurately predict the behavior of materials under a variety of conditions. In short, physics-based modeling can predict performance beyond the limits of previous observations.
The predictive capability of physics-based modeling using limited data is tremendously important for the development of novel materials because there is no comprehensive dataset for materials. The Materials Genome Initiative has made dramatic strides forward towards the development of a more-comprehensive materials dataset, but there are vast blind spots in currently available materials data.
How AI can effectively support physics-based modeling
The effectiveness of an AI tool is dependent on the size and quality of the dataset on which it is trained. ChatGPT, which launched the frenzy around AI with the release of its latest chatbot in the spring, was essentially trained on the whole of the internet so that it can generate new text. No comparable publicly available dataset for materials exists because there are dozens of variables involved in designing a material including composition, heat treatment and processing, as well as properties such as corrosion resistance, fatigue tolerance and weight.
Yet, AI does have a role to play. When there is sufficient data available to implement physics-based modeling, this approach is highly effective at designing a producible material that performs as predicted. But when there are gaps in the data or the potential areas to explore need to be narrowed, AI can highlight areas where modeling may result in breakthroughs.
AI technologies have four levels of advancement: descriptive, predictive, prescriptive and generative. For the past year, the world has been captivated by large language models (LLM) like ChatGPT and their ability to generate text that very closely resembles what a human would write. AI can also generate highly realistic images through services like Midjourney and Bing Image Creator.
But AI in materials engineering has not yet reached generative or predictive capabilities. AI can’t, on its own, design a new steel for use in an aircraft gearbox. But AI can describe an existing dataset to help engineers evaluate its potential to provide a solution to a given materials challenge. AI can’t predict what a dataset might look like if extended, so that physics-based modeling can prescribe what a given variable needs to look like to achieve a given outcome.
A winning combination
When used appropriately in combination, physics-based modeling supplemented with AI methods can dramatically speed up the process of developing and qualifying new materials. For example, AI played a crucial role in the development of a new niobium alloy that QuesTek recently completed for a space exploration OEM.
Based on materials data from previous projects, we were able to model all but one of the variables on the project using our ICMD® (Integrated Computational Materials Design) software. Our team leveraged AI for that one missing piece of the design, which tested out successfully and resulted in a viable new alloy that met the requirements of the OEM.
In the world of materials engineering, it’s a home run when you’re able to identify the right combination of variables to create a novel material that expands the bounds of performance. Physics-based modeling readily gets runners in scoring position. AI can provide the hit needed to bring runners home.
QuesTek is hosting an upcoming webinar titled “Physics-based modeling & machine learning to solve materials challenges.” Find more information and register here.