Machine Learning-Enabled Nanoscale Phase Prediction in Engineered Poly(Vinylidene Fluoride).

Area of Science:

• Materials Science
• Polymer Science
• Machine Learning Applications

Background:

• Engineered poly(vinylidene fluoride) (PVDF) exhibits diverse crystalline phases critical for piezo-, pyro-, ferro-, and tribo-electric devices.
• Accurate phase detection is vital for understanding structure-property relationships in PVDF materials.
• Traditional characterization methods face limitations in effectively distinguishing PVDF phases.

Purpose of the Study:

• To develop and validate a multimodal data-driven machine learning (ML) approach for distinguishing PVDF crystalline phases.
• To overcome the limitations of traditional characterization techniques in PVDF phase identification.
• To accelerate materials selection for PVDF-based devices by providing an autonomous phase distinction method.

Main Methods:

• Employed multimodal data-driven techniques combined with a machine learning (ML) model.
• Trained the ML model using a combination of empirical and theoretical data.
• Evaluated the model's performance in classifying different PVDF phases.

Main Results:

• Achieved a classification accuracy exceeding 94% for distinguishing PVDF phases.
• Demonstrated a 15% improvement in noise resilience compared to unimodal approaches.
• Showcased an 11% increase in accuracy when utilizing multimodal data.

Conclusions:

• The developed multimodal ML model offers an effective alternative for autonomous PVDF phase distinction.
• This approach significantly reduces the need for repetitive experiments, saving resources and time.
• The findings accelerate the process of materials selection for various PVDF applications.