YYYing Zhe Ye

Ying Zhe Ye

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Ying Zhe Ye has a total of 5 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    A multi-fidelity deep neural network method for predicting tiltrotor aerodynamic interaction during transition flight

    by Ying Zhe Ye, Yu Liang, Xiaowen Shan, Kefu Huang

    Tiltrotor electric vertical takeoff and landing aircraft experience strong rotor–rotor and rotor–wing aerodynamic interactions during transition flight, producing localized and operating-condition dependent wing-pressure and rotor-disk loading distributions. Unsteady Reynolds-averaged Navier-Stokes (URANS) simulations resolve these responses, but their high cost limits sampling of the operating space. Blade-element-momentum (BEM) calculations are much cheaper but exhibit systematic discrepancies in strongly three-dimensional interaction regions. This work develops a residual two-step multi-fidelity deep neural network (MF-DNN) for rapid prediction of high-dimensional aerodynamic snapshots comprising wing 𝐶𝑝 and rotor-disk 𝐹𝑧 and 𝐹𝜃 distributions.The HF and LF outputs are mapped onto common structured surface and polar disk grids. Thedatabase contains 40 URANS samples and 256 BEM samples, with eight URANS cases reservedfor independent testing. MF-DNN(16HF) uses 16 HF samples and all LF samples and is compared with single-fidelity DNNs trained using the same 16 or all 32 available HF training samples. Its LF response network learns the broad parametric trends, while the HF residual network uses the limited URANS data to correct the remaining fidelity discrepancies.Across the eight test cases, MF-DNN(16HF) gives the lowest mean relative 𝐿2 error for all ninedistribution-prediction tasks. Its mean 𝐶𝑝 error is 12.21%, compared with 18.21% and 16.56% for DNN(16HF) and DNN(32HF), respectively. Relative to DNN(32HF), it requires approximately half the training-data generation cost. The proposed framework therefore enables rapid prediction of wing pressure and forward- and rear-rotor loading distributions with a reduced HF-data requirement.