Adaptive Data-Driven Deep-Learning Surrogate Model for Frontal Polymerization in Dicyclopentadiene.

Area of Science:

• Materials Science
• Chemical Engineering
• Computational Science

Background:

• Frontal polymerization (FP) is a rapid, energy-efficient thermoset polymer curing method.
• Traditional simulation techniques like FEM are computationally intensive, hindering process optimization.
• Efficient simulation is crucial for analyzing sensitivity, uncertainty, and optimizing FP manufacturing.

Purpose of the Study:

• To develop an adaptive surrogate deep-learning model for frontal polymerization (FP) of dicyclopentadiene (DCPD).
• To achieve orders-of-magnitude speedup in predicting temperature and cure evolution compared to FEM.
• To enhance computational efficiency and accuracy in FP process modeling.

Main Methods:

• Developed an adaptive deep-learning surrogate model for FP simulations.
• Utilized automatic differentiation to calculate residual errors of governing equations.
• Employed a probability density function based on residual error for efficient training sample selection.
• Generated training data using 2D Finite Element Method (FEM) simulations.

Main Results:

• The adaptive surrogate model predicts temperature and cure evolution significantly faster than FEM.
• The adaptive sampling strategy proved more efficient and accurate than random sampling.
• The model achieved orders-of-magnitude speedup in predictive capabilities.
• Key FP characteristics like front speed, shape, and temperature were rapidly extracted.

Conclusions:

• The adaptive deep-learning model offers a computationally efficient and accurate approach for FP simulation.
• This method accelerates the analysis and optimization of thermoset polymer manufacturing.
• The surrogate model enables rapid extraction of critical process parameters from predicted fields.