Polymer Chemistry meets A.I. – Finding and Developing New Polymers with Target Properties in the 21st Century



The thermal conductivity of polymers is a key material property concerning applications such as vehicles electrification (e.g. traction motors and battery modules), communication devices as well as electronics. For instances, implementing a 5G communication standard requires antennas and associated parts being able to sink heat.

While making my research in this context, I came across a publication of Mr. Wu and his team from the Tokyo Institute of Technology in Japan. In their Nature publication entitled “Machine-learning-assisted discovery of polymers with high thermal conductivity using a molecular design algorithm”, they report on the use of big data analytics for the purpose of discovering new compounds and polymers. Their research targeted, in particular, the finding of a higher thermal conductivity Polyimide (PI). PI is often used in communication and sensor devices. For this, machine learning was applied, i.e. computers were allowed to learn from a given data set. In a first step, training of the algorithm is done in the given database. In a next step the trained application looks into a real world database containing several thousand of polymers to find PI and/or other compounds and/or combinations thereof which can fulfill the target requirements. Identification of more than thousand “virtual” polymers could be achieved by applying this methodology. In a next step, the three most promising polymers were selected out of the big pool with the underlying boundary condition of easy synthesis and processing. In the end, all suggested polymers were polyamides: a wholly aromatic polyamide (Figure 1a), an aromatic polyhydrazide (Figure 1b), and an aliphatic–aromatic polyamide (Figure 1c).

Figure 1: Resulting polymers of the molecular design study using machine learning and AI [2].

The suggested polymers were synthesized, cast into films and their thermal conductivities were tested. Commercial PI polymer such as Kapton® (PMDA/ODA*), UPILEX-S (BPDA/p-PDA*1) and UPILEX-R (BPDA/ODA*2) were similarly tested as well. The newly suggested polymers exhibited thermal conductivities up to 0.41 W/mK, 2x higher than their commercial PI counterparts whose thermal conductivities ranged from 0.19 to 0.21 W/mK. Previously, these values have only been reached by adding fillers such as boron nitrides to commercial PI’s.
 

Conclusion

Mr. Wu has shown that the use of machine learning and AI combined and big data analytics can be a very efficient and effective tool for materials design. Polymer chemists and data science work together hand in hand expanding the landscape of how to carry out research in the 21st century.

Thank you for reading!

Best regards,
Herwig Juster

If you liked this post, share and like!
Interested in my monthly blog posts – then subscribe here.
New to my Find Out About Plastics Blog – check out the start here section
Check out also my personal webpage.

Literature:

[1] https://www.scienceandtechnologyresearchnews.com/successful-application-of-machine-learning-in-the-discovery-of-new-polymers/

[2] S. Wu et al.: Machine-learning-assisted discovery of polymers with high thermal conductivity using a molecular design algorithm (https://www.titech.ac.jp/english/news/2019/044593.html)

* pyromellitic dianhydride and 4,4 –oxydianiline

*1 3, 3, 4, 4 -biphenyltetracarboxylic dianhydride and p-phenylenediamine

*2 3, 3, 4, 4 -biphenyltetracarboxylic dianhydride and 4,4 –oxydianiline