Streamlining materials testing using machine learning

Machine learning redefines material testing. MAD3, an algorithm from Sandia National Laboratories, expedites evaluations, promising efficiency across the manufacturing and aerospace industries.


Manufacturing and production have long grappled with time-consuming quality assurance tests to ensure the reliability of materials. Today, we stand on the brink of a paradigm shift โ€“ one where machine learning could redefine how we approach testing and research in a plethora of industries, from auto manufacturing to aerospace.

The machine learning revolution in materials testing

Hailing from Sandia National Laboratories, a new machine-learning algorithm may offer several industries a significantly expedited and more economical avenue for assessing bulk materials. This innovative technique was recently showcased in the esteemed scientific journal, Materials Science and Engineering: A.

For manufacturers, halting production due to unforeseen issues with materials can incur substantial financial burdens. Traditionally, materials like sheet metal undergo rigorous screenings to determine their formability, ensuring they withstand various manufacturing processes without any compromises in integrity. While commercial simulation software assists in these evaluations, calibrating them through multiple mechanical tests can span several months.

Moreover, while existing high-fidelity simulations can somewhat hasten the process, they demand both a supercomputer and niche expertise to operate. In a significant development, Sandia National Laboratories’ recent discovery suggests machine learning could be a game-changer in this field.

The MAD3 algorithm

The novel algorithm, known as MAD3 (Material Data Driven Design), is poised to replace traditional mechanical tests. At its core, MAD3 functions on a simple principle: metal alloys comprise minuscule “crystallographic” grains, which, when combined, yield a texture, granting the metal varying strengths in different directions, termed “mechanical anisotropy.”

Dr. David Montes de Oca Zapiain, the leading author of the breakthrough study, remarked, “We’ve orchestrated the model to decipher the intricate nexus between crystallographic texture and anisotropic mechanical responses.” In simpler terms, with just an electron microscope to discern a metal’s texture, this data can be inputted into MAD3. Subsequently, the algorithm proffers the requisite data for the simulation software, bypassing any need for exhaustive mechanical tests.

In collaboration with Ohio State University, Sandia facilitated the training of MAD3 on the outcomes of a staggering 54,000 simulated materials tests, leveraging a method known as a feed-forward neural network. To determine its precision, 20,000 novel microstructures were then introduced to the algorithm, with its results being juxtaposed against data derived from experiments and supercomputer-based simulations.

MAD3: The new gold standard?

Hojun Lim, a scientist at Sandia and a contributor to this pioneering research, elaborated on the sheer efficiency of MAD3, stating, “The formulated algorithm operates at a speed roughly 1,000 times faster than high-fidelity simulations.” The team’s ambitions don’t halt here, as ongoing efforts aim to enhance the model by integrating advanced functionalities to capture anisotropy evolutions crucial for predicting material fracture limits.

With its roots in national security, Sandia’s research has now expanded its horizon to discern if MAD3 can expedite the quality assurance processes, especially for the U.S. nuclear stockpile. Given the stringent criteria these materials need to adhere to before being green-lit for production, the implications of a successful integration are profound.

Sharing the breakthrough

Sandia is not keeping this groundbreaking technology to itself. In a bid to empower other institutions, they’ve assembled a cross-disciplinary team dedicated to devising the Material Data Driven Design software. This user-centric, graphics-rich software was brought to life post extensive consultations, encompassing over 75 interviews with potential users, facilitated by the Department of Energy’s Energy I-Corps programme.

In essence, the union of machine learning and materials testing, as showcased by the MAD3 algorithm, could herald a new era in manufacturing and production, ensuring that industries no longer need to tread the fine line between time efficiency and quality assurance.ย 

The broader implications of MAD3 and generative AI for QA

The emergence of MAD3 as a transformative tool is not merely confined to the realms of manufacturing and aerospace. In an era where efficiency and precision are paramount, the introduction of machine learning algorithms, as well as generative AI for QA, signals vast potential across a myriad of sectors. Industries ranging from pharmaceuticals, where the quality assurance of compounds is critical, to electronics manufacturing, where the minutiae of materials can influence product longevity, could benefit immensely. As more sectors begin to harness the capabilities of such advanced algorithms, we could witness a significant paradigm shift, not just in quality testing, but in the very essence of research and development across diverse industries.