Multiscale Analysis of Composite Structures with Artificial Neural Network Support for Micromodel Stress...

Area of Science:

• Computational mechanics
• Materials science
• Artificial intelligence

Background:

• Multiscale finite element analysis is crucial for composite materials.
• Homogenized properties in macroscale models yield inaccurate stress predictions.
• Accurate stress analysis in heterogeneous microstructures is computationally intensive.

Purpose of the Study:

• To decrease computation time for multiscale analysis of composite structures.
• To apply machine learning for efficient stress prediction in heterogeneous materials.
• To improve the accuracy of stress localization in composite materials.

Main Methods:

• Utilized a computational multiscale approach for analyzing composite structures.
• Employed artificial neural networks (ANNs) trained on carefully prepared data.
• Validated the methodology with a numerical example of a short glass fiber-reinforced epoxy resin.

Main Results:

• Machine learning significantly decreased computation time for multiscale analyses.
• ANNs learned the relationships between macroscale and microscale behavior.
• The efficiency of the multiscale approach was substantially increased.

Conclusions:

• Machine learning offers a powerful tool for accelerating multiscale simulations of composites.
• This approach enhances the prediction of stress distributions in complex material structures.
• The validated methodology provides a more efficient alternative to traditional time-consuming analyses.