Design of Thermosetting Polymers with High Thermal Stability and Enhanced Processability via ML-assisted Material Genome Approach

Phthalonitrile
(PN), due to its good mechanical and thermal properties,
stands out as an ideal high-performance resin. However, the conflict
between heat resistance and processing property has always hindered
the development and application of PN resins. We proposed a machine-learning-assisted
materials genome approach to enable the practical design of PNs with
excellent heat resistance and expansive processing windows. Combining
domain knowledge, we constructed three property prediction models
through machine learning. Subsequently, an extensive collection of
candidate structures was created to screen optimal structures based
on the machine learning predictions. Three PN structures were selected
for proof-of-concept experiments. The measure properties of PNs agree
well with the predictions, confirming the effectiveness of our strategy.
The optimal PN surpasses existing performance constraints and extends
processing boundaries without compromising heat resistance. In addition,
the integrated gradients method was extended to analyze chemical “genes”,
which deepens the understanding of the structure–property relationship
of the PNs, and takes a step forward in implementing interpretable
machine learning methods. The insights derived from gene analysis
can assist in the design of heat-resistant polymers with excellent
processing properties.