New Design Method for Fabricating Multilayer Membranes Using CO₂-Assisted Polymer Compression Process.

Area of Science:

• Materials Science
• Chemical Engineering
• Artificial Intelligence

Background:

• Developing multilayer membranes with controlled porosity is crucial for various applications.
• The CO2-assisted polymer compression (CAPC) method offers a pathway for creating such membranes.
• Optimizing the design process for CAPC-fabricated membranes requires efficient methods.

Purpose of the Study:

• To investigate the efficacy of deep learning in designing multilayer membranes using the CAPC method.
• To explore strategies for improving the accuracy of deep learning models with limited experimental data.
• To demonstrate the unique simulation capabilities of deep learning in process design.

Main Methods:

• Utilized deep learning algorithms for the design of multilayer membranes.
• Employed the CO2-assisted polymer compression (CAPC) method for membrane fabrication.
• Expanded experimental data from two-layer to three-layer compression for model training.
• Incorporated additional three-layer experimental data to enhance model accuracy.

Main Results:

• Deep learning models were trained to predict the behavior of multilayer membranes during CAPC.
• Initial training with extrapolated data showed insufficient accuracy.
• Adding specific three-layer experimental data dramatically improved model predictive accuracy.
• Demonstrated deep learning's ability to simulate process results without explicit physical models.

Conclusions:

• Deep learning is highly effective for designing multilayer membranes fabricated via the CAPC method.
• Careful curation and augmentation of training data are critical for achieving high accuracy in deep learning models for materials design.
• Deep learning offers a powerful, data-driven approach to optimize complex material fabrication processes.