.png){kind=icon}
.png){kind=icon}
Return
Variable-Stiffness Composite Optimization Using Dynamic and Exponential Multi-Fidelity Surrogate Models
.png){kind=icon}
PRE
AI
DOI：10.1016/j.ijmecsci.2023.108547.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Variable-stiffness composite laminates with spatially varied orientation angles always require refined finite element models to accurately model the spatial variation characteristics, thus resulting in high computation costs. Further, practical restrictions in fiber steering should be imposed to generate manufacturable designs, making the design problem more challenging. To address these challenges, this paper presents a new framework assisted by multi-fidelity surrogate models for variable-stiffness composite optimization with manufacturing constraints. An initial sampling strategy is originally developed for the case of involving the fiber steering constraints, improving the accuracy of the surrogate model in the concerned space. Based on Gaussian process regressions, a new type of multi-fidelity model corrected with an exponential function is proposed by fusing many cheap low-fidelity models and a few expensive high-fidelity models. Using genetic algorithm as the optimizer, new data points are generated from the optimization process and then employed to dynamically update the constructed multi-fidelity model. The proposed optimization strategy is applied to case studies of buckling optimization for both a composite plate and a composite cylinder, demonstrating that the developed framework requires significantly less computation.
Keywords:
Variable-stiffness composite
Multi-fidelity surrogate model
Optimal design
Fiber steering constraint
Gaussian process regression
Journal
IF:
9.4
Papers:
1.0W
Citations:
4.5W