Automatically Predicting Material Properties with Microscopic Images: Polymer Miscibility as an Example

Area of Science:

• Materials Science
• Computer Science
• Polymer Science

Background:

• Material properties are often assessed via microscopic imaging, like scanning electron microscopy (SEM).
• Polymer miscibility is crucial but typically judged subjectively from SEM images, which is inefficient and difficult to quantify.
• Existing methods for assessing polymer miscibility are time-consuming and labor-intensive.

Purpose of the Study:

• To develop an automated, accurate, and quantitative method for polymer miscibility recognition using computer vision.
• To overcome the limitations of subjective human judgment in analyzing SEM images for material characterization.
• To establish a quantitative criterion for polymer miscibility assessment.

Main Methods:

• Utilized convolutional neural networks (CNNs) and transfer learning for image analysis.
• Developed a machine learning model for automatic recognition of polymer miscibility from SEM images.
• Implemented computer image recognition to provide quantitative judgments.

Main Results:

• Achieved up to 94% accuracy in automatic polymer miscibility recognition.
• Successfully developed a quantitative criterion for assessing polymer miscibility.
• Demonstrated the potential for accurate and quantitative material characterization.

Conclusions:

• The proposed AI-driven method offers a significant improvement over manual analysis of SEM images for polymer miscibility.
• This approach provides accurate, quantitative, and efficient characterization of polymer microstructure.
• The method is broadly applicable to microstructure and property characterization across various materials.