Multi-Objective Optimization of Thin-Walled Composite Axisymmetric Structures Using Neural Surrogate Models and...

Area of Science:

• Materials Science and Engineering
• Computational Mechanics

Background:

• Composite shells offer excellent strength-to-weight ratios but require precise parameter optimization.
• Stochastic optimization methods like genetic algorithms are computationally intensive.
• Surrogate models using neural networks can approximate complex functions efficiently.

Purpose of the Study:

• To investigate the use of neural network surrogate models for multi-objective optimization of composite shells.
• To enhance the accuracy and reliability of optimization processes through mode shape identification and network ensembles.
• To evaluate the computational efficiency and efficacy of the proposed methodology.

Main Methods:

• Deep neural networks were employed as surrogate models to approximate input parameter-objective function relationships.
• Mode shape identification was incorporated to improve accuracy in multi-criteria optimization.
• Network ensembles were utilized to enhance model robustness and reliability.
• Efficiency analysis compared computational costs against traditional methods like Monte Carlo simulations.

Main Results:

• The surrogate model approach, enhanced by mode shape identification and network ensembles, demonstrated high accuracy and reliability.
• The methodology proved efficient in handling complex input parameters and intricate designs.
• A favorable trade-off between computational cost and accuracy was achieved.

Conclusions:

• The integration of network ensembles as surrogate models and mode shape identification significantly enhances multi-objective optimization of composite shells.
• This efficient and accurate approach has broad implications for advanced engineering design and optimization methodologies.