Studying Complex Evolution of Hyperelastic Materials under External Field Stimuli using Artificial Neural Networks...

Area of Science:

• Materials Science
• Computational Chemistry
• Machine Learning

Background:

• Deep learning (DL) excels at modeling complex material structure-property relationships, offering a new paradigm for material design.
• Training accurate DL models often requires large datasets, which are frequently impractical to obtain in experimental materials science.

Purpose of the Study:

• To demonstrate the necessity of low-dimensional, accurate descriptors for high-precision DL models using small experimental datasets.
• To develop a novel DL model capable of effectively utilizing limited data for material design.

Main Methods:

• Utilized a dataset of 483 porous silicone rubber observations from additive manufacturing.
• Developed a convolutional bidirectional long short-term memory (BiLSTM) model with spatiotemporal feature extraction.
• Implemented a hierarchical learning mechanism to maximize data information utilization.

Main Results:

• Established that low-dimensional descriptors are crucial for high-precision DL on small material datasets.
• The proposed convolutional BiLSTM model demonstrated effective performance with limited experimental data.
• The model's hierarchical learning reduced data requirements by leveraging data information efficiently.

Conclusions:

• The developed approach is a powerful tool for innovative material design using small experimental datasets.
• The methodology can be applied to explore material structure-property evolution under complex conditions.
• Highlights the importance of descriptor development in machine learning for materials science.