ChemNet: A Deep Neural Network for Advanced Composites Manufacturing

Area of Science:

• Materials Science
• Chemical Engineering
• Artificial Intelligence

Background:

• Advanced manufacturing of fiber-reinforced polymer-matrix composites (FRPCs) is crucial for aerospace, marine, automotive, and energy sectors.
• Frontal polymerization (FP) offers significant time and energy savings but faces challenges in process control due to thermosetting polymer cure kinetics.
• Optimizing cure kinetics is essential for reliable and efficient FRPC fabrication using FP.

Purpose of the Study:

• To develop a deep learning model for predicting and optimizing thermosetting polymer cure kinetics parameters in FRPCs.
• To address the inverse problem of relating front characteristics to cure kinetics for desired fabrication strategies.
• To enhance the design and control of frontal polymerization processes.

Main Methods:

• Development of ChemNet, a 9-layer fully connected FeedForward deep neural network.
• Training the model on one million examples of cure kinetics data.
• Predicting activation energy and reaction enthalpy based on frontal characteristics (speed, maximum temperature).

Main Results:

• ChemNet achieved highly accurate predictions for cure kinetics parameters.
• The model demonstrated a Mean Squared Error (MSE) of 5.58 × 10^-6 on a hidden test dataset.
• A Maximum Absolute Error (MAE) of 1 × 10^-3 was recorded, indicating strong predictive performance.

Conclusions:

• ChemNet effectively solves the inverse problem for thermosetting FRPC cure kinetics.
• The deep learning approach enables precise control and optimization of frontal polymerization processes.
• This facilitates faster, more energy-efficient manufacturing of advanced composite materials.