Predicting the Properties of High-Performance Epoxy Resin by Machine Learning Using Molecular Dynamics Simulations

Area of Science:

• Materials Science
• Polymer Chemistry
• Computational Chemistry

Background:

• Epoxy resins are versatile adhesives with properties dependent on formulation.
• Traditional optimization methods are time-consuming and costly.
• Predictive modeling can accelerate the discovery of optimal epoxy formulations.

Purpose of the Study:

• To develop a predictive model for epoxy resin adhesive properties.
• To overcome limitations of traditional trial-and-error formulation methods.
• To explore the relationship between epoxy composition and performance.

Main Methods:

• Utilized molecular dynamics (MD) simulations to generate datasets.
• Employed machine learning (ML), specifically artificial neural networks (ANN), for property prediction.
• Performed linear correlation analysis to identify key compositional factors.

Main Results:

• Identified strong correlations between specific components (DICY, TGMDA, DGEBA) and properties (cohesive energy density, modulus, glass transition temperature).
• Developed an optimized ANN model with high predictive accuracy (R^2: 0.835-0.986).
• Demonstrated the model's effectiveness in predicting epoxy adhesive properties.

Conclusions:

• MD simulations and ML offer an efficient alternative to traditional methods for epoxy resin optimization.
• The predictive technique can guide the formulation of epoxy systems with desired adhesive properties.
• This approach has potential for broader application in materials science and engineering.