For years, the transportation industry has used 3D printers, but while the machines and materials have evolved,
the procedures for generating those materials have not.
However, a research group at MIT, with the help of BASF and Boston University, believes it has discovered a superior technique.
“Materials development is still very much a manual process: a chemist goes into a lab, mixes ingredients by hand, makes samples, tests them, and comes to a final formulation,”
says Mike Forshey, a mechanical engineer and project manager in the Computational Design and Fabrication Group (CDFG) of the Computer Science and Artificial Intelligence Laboratory (CSAIL).
However, rather than having a scientist who can only do a few iterations over a few days,
In the same amount of time, our system may perform hundreds of iterations.”
The goal is to improve material properties in a variety of areas, such as strength in one direction or toughness.
“We think this would outperform the current way for a variety of applications because you can rely more heavily on the optimisation process to identify the ideal solution,” Foshey explains.
To preselect the material formulas, you wouldn’t require an expert chemist on hand.”
Machine learning is at the centre of it all.
Rather than going through the tedious process described above,
the operator selects a few substances and enters their chemical compositions into the algorithm.
After the final attributes have been defined,
the algorithm changes the proportions to produce a satisfactory end product.
The results are supplied back into the algorithm after the ingredients have been blended and tested.
There’s also AutoOED, a free software spin-off that researchers might find beneficial.
The company thinks that this will open up new material alternatives that humans might ignore.
The best is being put to the test.
The researchers designed a test to see how well it worked: creating an ink that sets under ultraviolet light.
Toughness, compression modulus (stiffness), and strength were dialled in as final attributes since this particular combination has competing criteria, requiring one value at a time to be obtained manually.
A total of six materials were chosen.
After a 120-sample test,
12 optimal alternatives were identified.
“This has extensive applications across materials research in general,” Foshey explains.
For example, if you wanted to create new sorts of batteries that were more efficient and less expensive, you could utilize a system like this.
This method could also help automobile manufacturers optimize paint that performs better while being environmentally friendly.”
“The focus on novel material formulations is particularly encouraging,” says team member Keith A. Brown, assistant professor in the Department of Mechanical Engineering at Boston University.
“This is a factor that is often overlooked by researchers who are constrained by commercially available materials.”
The team can also discover materials quickly because of the mix of data-driven approaches and experimental science. The methods offered have a chance of persuading the community to embrace more data-driven practices because experimental efficiency is something that all experimenters can identify with.”