Multi-fidelity neural network-based aerodynamic optimization framework for propeller design in electric aircraft

Wu, Xiaojing; Zuo, Zijun; Ma, Long; Zhang, Weiwei

doi:10.1016/j.ast.2024.108963

Cited by 5 publications

(1 citation statement)

References 38 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Nagawkar et al [18] employed a multi-fidelity deep neural network model for the aerodynamic optimization design of airfoils and wings and obtained multi-fidelity data through calculations with grids of different resolutions. Wu et al [19] describe a Multi-Fidelity Neural Network (MFNN)-based optimization framework for the optimization design of electric aircraft propellers and improve the accuracy of the propeller's aerodynamic force model through multi-fidelity data fusion. Many scholars employ different methods to construct surrogate models for obtaining various aerodynamic data.…”

Section: Introductionmentioning

confidence: 99%

High-Efficiency Data Fusion Aerodynamic Performance Modeling Method for High-Altitude Propellers

Zhang,

Jiao,

Zhang

et al. 2024

Drones

View full text Add to dashboard Cite

During the overall design phase of solar-powered unmanned aerial vehicles (UAVs), a large amount of high-fidelity (HF) propeller aerodynamic performance data is required to enhance design performance, but the acquisition cost is prohibitively expensive. To improve model accuracy and reduce modeling costs, this paper constructs a multi-fidelity aerodynamic data fusion model by associating data with different fidelity. This model utilizes a low-fidelity computational method to quickly determine the design space. The constrained Latin hypercube sampling based on the successive local enumeration (SLE-CLHS) method and the expected improvement (EI) criterion were adopted to achieve the efficient initialization and fastest convergence of the Co-Kriging surrogate model within the design space. This modeling framework was applied to acquire the aerodynamic performance of high-altitude propellers, and the model was evaluated using various performance indicators. The results demonstrate that the proposed model has excellent predictive performance. Specifically, when the surrogate model was constructed using 350 high-fidelity samples, there were improvements of 13.727%, 12.241%, and 5.484% for thrust, torque, and efficiency compared with the surrogate model constructed from low-fidelity samples.

show abstract

Section: Introductionmentioning

confidence: 99%