Multiscale Computational and Artificial Intelligence Models of Linear and Nonlinear Composites: A Review.

Area of Science:

• Materials Science and Engineering
• Computational Mechanics
• Multiscale Modeling

Background:

• Composite materials, both hard (polymers, metals, fibers) and soft (biological tissues like brain white matter), require sophisticated modeling techniques.
• Existing numerical models range from molecular dynamics to finite-element (FE) analysis, but challenges remain in computational resources, model fidelity, and repeatability.
• The need for advanced constitutive models, including viscoelastic and fractional viscoelastic types, is critical for accurately capturing material behavior.

Purpose of the Study:

• To describe state-of-the-art multiscale modeling methods for diverse composite materials.
• To outline key challenges and limitations in current multiscale modeling approaches.
• To present recent advancements and future trends in computational modeling of composites.

Main Methods:

• Review of molecular dynamics simulations, finite-element (FE) analyses, and machine learning/deep learning surrogate models.
• Summarization of constitutive material models: viscoelastic, hyperelastic, and fractional viscoelastic.
• Exploration of advanced FE techniques like meshless methods and hybrid machine learning (ML) and FE models.

Main Results:

• Identification of challenges including meshing, data variability, nonlinearity-driven uncertainty, computational constraints, and model repeatability.
• Presentation of latest advancements in FE modeling and data-driven models utilizing extensive experimental data.
• Demonstration of data-driven models developed across various length and time scales.

Conclusions:

• Recent advancements in FE and data-driven models offer a clearer outlook on futuristic trends in composite multiscale modeling.
• Data-driven models provide essential tools for real-time monitoring and lifecycle prediction of composite structures.
• The integration of advanced mathematical and numerical techniques with large experimental datasets is crucial for developing robust digital models.