Single-Criterion Optimisation with Consideration of Uncertainties of the Composite Multi-Layer Slabs

Area of Science:

• Computational Mechanics
• Materials Science
• Optimization Techniques

Background:

• Optimizing the fundamental frequency of composite slabs is crucial for structural integrity.
• Traditional Finite Element Method (FEM) based optimization is computationally expensive.
• Incorporating uncertainties due to manufacturing tolerances is essential for robust designs.

Purpose of the Study:

• To develop a novel, efficient computational framework for optimizing the fundamental frequency of multi-layered composite slabs.
• To integrate uncertainty quantification into the optimization process for enhanced robustness.
• To significantly reduce the computational cost associated with traditional optimization workflows.

Main Methods:

• Utilized Finite Element Method (FEM) for data generation.
• Employed Deep Neural Network (DNN) surrogate modeling to replace expensive modal analyses.
• Implemented deterministic optimization using Genetic Algorithm (GA) and uncertainty-aware quantile-based optimization.

Main Results:

• The DNN-GA framework achieved near-identical optimal non-dimensional parameter values (Ω) compared to literature values (Bayesian Optimization), with <0.5% discrepancy.
• Uncertainty-aware optimization, by maximizing the 5% quantile of Ω, yielded robust solutions with minimal performance loss.
• The surrogate model significantly reduced optimization evaluation costs for various boundary conditions.

Conclusions:

• The proposed DNN-GA framework offers a fast and accurate method for optimizing composite laminate fundamental frequencies.
• The approach provides both deterministic and robust design recommendations.
• This computational framework achieves significant computational cost savings compared to traditional FEM-based optimization.