Computational Strategy for Analyzing Effective Properties of Random Composites-Part III: Machine Learning.

Area of Science:

• Materials Science
• Computational Mechanics
• Machine Learning Applications

Background:

• Previous work (Parts I and II) analyzed two-dimensional dispersed random composites.
• Existing analytical methods are extended by incorporating machine learning (ML) for microstructure classification.

Purpose of the Study:

• To quantitatively classify microstructures in two-dimensional dispersed random composites using ML.
• To demonstrate the dependence of the effective tensor on geometric probabilistic distributions and computational protocols.

Main Methods:

• Decomposition of effective tensor expressions into geometrical and physical parts (structural sums and physical constants).
• Application of machine learning analysis using feature vectors derived from structural sums.
• Development of an analytical algorithm for computing effective elastic constants in two-phase composites with circular inclusions.

Main Results:

• Explicit demonstration of the effective tensor's dependency on inclusion distribution and computational realization.
• Successful classification of dispersed random composites that are indistinguishable by simple observation.
• A computationally effective strategy for classifying microstructures is developed.

Conclusions:

• The integration of ML with analytical decomposition provides a powerful tool for microstructure characterization.
• The developed strategy offers a robust method for classifying complex composite materials.
• This approach advances the understanding and analysis of elastic fibrous composites.