Inverse design of aero-engine nacelles using conditional denoising diffusion probabilistic models

Nacelle designs play a crucial role in the overall efficiency of high-bypass ratio turbofan engines. This study proposes a novel inverse design framework based on conditional denoising diffusion probabilistic models (CDDPM) to generate optimal nacelle geometries from aerodynamic performance metrics. The framework comprises a CDDPM for generating high-fidelity pressure distributions and a deep learning-based mapping network for precise geometric reconstruction. A high-fidelity dataset of 9600 samples spanning freestream Mach numbers from 0.82 to 0.88 and mass flow capture ratios from 0.65 to 0.75 is constructed to train the model. Comparative evaluations demonstrate that the CDDPM-based framework significantly outperforms conditional variational autoencoders (CVAE) and conditional generative adversarial networks (CGAN) in terms of generation quality and physical consistency. Specifically, the CDDPM achieves a coefficient of determination R ² of 0.992 for drag prediction on the validation set, while the mapping network ensures high reconstruction accuracy with R ² ' 0.99. In inverse design tasks targeting 5% and 10% drag reductions at the design condition, the proposed framework generates geometries with drag predictions closely matching the target values and significantly lower variance than CGAN and CVAE, while satisfying aerodynamic constraints on the peak isentropic Mach number M_is ' 1.30 with success rates exceeding 90%. The framework is further validated at two off-design cruise conditions targeting 10% drag reduction, demonstrating its generalization capability beyond the design point.