Predicting Crystallization Tendency of Polymers Using Multifidelity Information Fusion and Machine Learning.

Area of Science:

• Polymer Science and Engineering
• Materials Informatics
• Computational Materials Science

Background:

• Polymer crystallinity significantly influences mechanical, ion transport, and gas permeability properties, impacting applications from composites to batteries.
• Experimental determination of polymer crystallinity is often time-consuming and expensive, hindering rapid material screening.
• Existing data for polymer crystallinity under consistent processing conditions are scarce, posing challenges for predictive modeling.

Purpose of the Study:

• To develop a data-driven machine learning model for rapid and cost-effective prediction of polymer crystallinity.
• To address data scarcity and variability by integrating experimental and computational (group contribution) datasets.
• To enable efficient screening of polymers for diverse applications based on their crystallization tendency.

Main Methods:

• Developed a machine learning model trained on a high-fidelity experimental dataset (107 polymers) and a low-fidelity computational dataset (429 polymers).
• Employed a multifidelity information fusion strategy to leverage diverse data sources while maintaining high prediction accuracy.
• Modeled the 'tendency to crystallize' to account for inherent process variability and predict the 'most-likely' crystallinity.

Main Results:

• Successfully created a machine learning model capable of predicting the most-likely polymer crystallinity.
• The multifidelity approach effectively combined limited experimental data with broader computational data for improved predictions.
• The model offers a significantly more cost-effective and efficient method for estimating polymer crystallization tendency compared to traditional experiments.

Conclusions:

• The developed machine learning model provides a powerful tool for accelerating polymer discovery and development.
• This data-driven approach overcomes limitations of experimental data scarcity and cost in predicting polymer properties.
• The model facilitates faster, cheaper screening of polymers for applications requiring specific crystallinity levels.