Design of PMMA-Cotton Composite Textile with Tunable Properties via a Physics-Aware Bidirectional Neural Network...

Area of Science:

• Materials Science
• Textile Engineering
• Computational Materials Science

Background:

• Natural fibers like cotton offer sustainable composite bases but often lack tunable mechanical properties.
• Developing methods to enhance cotton's mechanical response is crucial for advanced material applications.

Purpose of the Study:

• To create a tunable composite material from cotton textiles using Polymethyl methacrylate (PMMA) impregnation.
• To develop and validate a physics-aware artificial neural network (ANN) for predicting and controlling the composite's mechanical behavior.

Main Methods:

• Vacuum-assisted impregnation of cotton fabric with varying concentrations of acetone-borne PMMA.
• Characterization of mechanical properties (elastic modulus, tensile strength, etc.) and thermal properties (DSC, Tg).
• Development of a physics-aware bidirectional ANN incorporating material science constraints (rule-of-mixtures, porosity, thermal).

Main Results:

• Optimal PMMA loading for load transfer was identified between 0.5-1.0 wt.%.
• The ANN achieved sub-percent prediction errors for both forward and inverse modeling.
• Physics regularization significantly improved ANN accuracy, reducing errors by fivefold (forward) and one order of magnitude (inverse).

Conclusions:

• The developed PMMA impregnation process effectively tunes cotton textile properties from flexible to stiff.
• Physics-aware ANNs provide a robust framework for accurate prediction and optimization of composite material behavior.
• Integrating physical constraints into AI models is essential for reliable material science predictions.