MIT is incorporating AI to accelerate the discovery of new materials for 3D printing
MIT is incorporating AI to accelerate the discovery of new materials for 3D printing
Increased demand for new 3D printing materials designed for extremely specific applications has been driven by the growing popularity of 3D printing for the manufacture of a variety of items, ranging from customized medical devices to affordable housing.
In order to reduce the amount of time it takes to discover new materials, MIT researchers developed a data-driven process that uses machine learning to optimize new 3D printing materials with a variety of properties, such as toughness and compression strength, before they are commercialized.
By streamlining material development, the system lowers costs while also having a positive impact on the environment through the reduction of chemical waste generated. Additionally, the machine learning algorithm has the potential to stimulate innovation by suggesting novel chemical formulations that would otherwise be overlooked by human intuition.
It is still a manual process for the majority of materials development. Upon entering a laboratory, the chemist manually mixes ingredients, prepares samples, and tests them until he or she arrives at a final formula. In contrast, instead of having a chemist perform a few iterations over a few days, our system can perform hundreds of iterations in the same amount of time," says Mike Foshey, a mechanical engineer and project manager in the Computer Science and Artificial Intelligence Laboratory's (CSAIL) Computational Design and Fabrication Group (CDFG) and co-lead author of the paper.
Timothy Erps, a technical associate with the CDFG; Mina Konakovic Lukovic, a CSAIL postdoc; Wan Shou, a former MIT postdoc who is now an assistant professor at the University of Arkansas; Wojciech Matusik, a professor of electrical engineering and computer science at MIT; and Hanns Hagen Geotzke, Herve Dietsch, and Klaus Stoll of BASF are among the other authors on this paper. On October 15, 2021, the findings of the study were published in the journal Science Advances.
Increasing the likelihood of discovery
The researchers created a system in which an optimization algorithm takes care of the majority of the trial-and-error discovery process, rather than the researchers themselves.
A material developer chooses a few ingredients and feeds information about their chemical compositions into an algorithm, after which he or she specifies the mechanical properties of the newly created material. Once the proportions of those components have been determined (similar to how knobs on an amplifier are adjusted), the algorithm evaluates how each formula affects the material's properties before arriving at the optimal combination.
The sample is then mixed, processed, and tested by the developer in order to determine the true performance of the material. The results of the experiment are communicated to the algorithm, which automatically learns from the experience and uses the newly acquired knowledge to select another formulation to test.
Because it relies more heavily on the optimization algorithm to find the optimal solution, we believe that this method will outperform the conventional method in a number of applications. "You wouldn't need the presence of an expert chemist to pre-select the material formulations," Foshey says.
A freely available open-source materials optimization platform, AutoOED, has been developed by the researchers, which incorporates the same optimization algorithm as the original AutoOED. Researchers can perform their own optimization with AutoOED, which is a comprehensive software suite that allows them to do so.
The researchers used the system to optimize formulations for a new UV-curable 3D printing ink, which served as a validation for the system.
Their algorithm was given six chemicals to work with in the formulations, and they instructed it to look for the material that had the highest toughness, compression modulus (stiffness), and strength possible.
Because these three properties can be in conflict with one another, manual optimization of these three characteristics would be particularly difficult. For example, the strongest material may not be the stiffest. A chemist would typically attempt to optimize a single property at a time using a manual process, resulting in a large number of experiments and significant waste.
A computer algorithm was used to identify 12 materials that had the best trade-offs between the three different properties after 120 samples were tested.
As a result of the algorithm's versatility, Foshey and his collaborators were taken aback by the range of materials produced, noting that the results were far more varied than they had anticipated based on the six ingredients. When using the system, exploration is encouraged, which may be particularly beneficial in situations where specific material properties are not readily apparent intuitively.
In the future, things will move faster
Additional automation may be able to accelerate the process even further. Currently, the researchers are manually mixing and testing each sample, but in future versions of the system, robots could operate the dispensing and mixing systems, according to Foshey.
Additionally, the researchers hope to use this data-driven discovery process for other purposes in the future, such as the development of new 3D printing inks.
In general, this has a broad range of applications across the field of materials science. For example, if you wanted to research and develop new types of batteries that were both more efficient and less expensive, you could do so with the assistance of this system. Alternately, he says, "if you wanted to optimize paint for a car that was both efficient and environmentally friendly, this system could also do that."
In the opinion of Keith A. Brown, assistant professor in the Department of Mechanical Engineering at Boston University, this work could be a significant step forward in the development of high-performance structures because it presents a systematic approach for identifying optimal materials. Brown is a leading expert on high-performance structures.
In particular, the emphasis on novel material formulations is heartening because it is an aspect that is frequently overlooked by researchers working with commercially available materials. Furthermore, the team's efficient identification of materials is made possible by the combination of data-driven methods and experimental science, which they have developed. Given that "experimental efficiency" is a concept that is shared by all experimenters, the methods presented here have the potential to motivate the scientific community to adopt more data-driven practices, according to the author.
