October 20, 2021 2 min read
Three-dimensional (3D) printing is being used in ground-breaking ways. Nonprofits like WaterScope are using 3D printing to create lightweight, inexpensive microscopes to detect bacteria in water. Scientists at Osake University have used 3D printers to create meat. And researchers at Carnegie Mellon are even using 3D printers to print human organs. In an article published today by Andrei Mihai, he writes about the work of Massachusetts Institute of Technology (MIT) researchers using machine learning to improve 3D printing.
As 3D printers are used for different means, the materials used for printing need to be different. For example, you wouldn’t use the same materials for printed meat, microscopes and replicated human organs. But determining the exact chemical combinations to create these materials is costly, requires specific domain expertise and extensive testing. Machine learning allows scientists to more efficiently predict the right combination of chemicals to create desired materials.
The typical process, Mihai reports, includes a chemist, with domain expertise, mixing ingredients together and testing the combination extensively to determine the right formula, but at this point the ingredients are already mixed so you are testing in a reactionary way and if you are wrong, the materials are wasted. Being wrong during experimentation processes is part of the deal, but when being wrong is costly and leads to large volumes of waste, it can make the necessary experimentation processes hard to stomach. Beyond the waste, this process presents several challenges including: 1) relying heavily on domain expertise to predict what combinations will be successful; 2) requiring time and effort for extensive experimentation; and 3) optimizing for only one property at a time instead of the best combination of the properties.
Machine learning allows scientists to be more informed before they begin mixing. The algorithm created by these scientists learns from prior experiments. The algorithm then takes in information regarding the chemical ingredients and the desired properties of the material, and the output is a recommendation of different combinations of the ingredients to optimize for the desired material.
Machine learning becomes extremely powerful when partnered with other emerging technologies like 3D printing. As more scientists experiment with machine learning in their fields, we will likely see innovations increase exponentially. This means advances in lab developed meat, international development and healthcare, among many others.