Scientists from the Tokyo Institute of Technology have undertaken a study that demonstrates how a form of machine learning which uses “transfer learning,” can assist with the discovery of materials with desired properties, even from a very small data set. Most forms of machine learning require a high number of data inputs with which to produce meaningful analyses.
The new process to discover heat-resistant polymers draws on machine learning, data science, organic synthesis and advanced measurement technologies. The task required the use of specially designed machine learning algorithm, which was programmed for computational molecular design. The algorithm is named the iQSPR.
The algorithm was used to scan polymeric properties from PoLyInfo, which is the largest database of polymers in the world. Through this process, the researchers have managed to verify the heat transfer properties of a range of computationally designed polymers. This opens up new possibilities for materials design.
The identified polymers were subjected to physical testing. These tests showed the machine learning selected materials to possess high thermal conductivity (up to 0.41 Watts per meter-Kelvin). This level standards as 80 percent higher compared with typical polyimides (the common polymers which have been mass-produced since the 1950s and which are used in a range of everyday items).
The focus on polymers that have effective heat management will assist with the development the fifth-generation (5G) mobile communication technologies.
According to lead scientist Dr. Ryo Yoshida: “Machine learning for polymer or soft material design is a challenging but promising field as these materials have properties that differ from metals and ceramics, and are not yet fully predicted by the existing theories.”
The research demonstrates how machine learning methods can out-perform conventional ways of searching for high-performance plastic materials. The research has been reported to the journal npj Computational Materials, and the study is titled “Machine-learning-assisted discovery of polymers with high thermal conductivity using a molecular design algorithm.”
