Deep learning–based artificial intelligence approach has provided an extremely efficient means for evaluating aerodynamic pressure over aircraft surfaces. Yet current approaches face two major challenges: (1) the capacity to represent features in geometrically complex regions is insufficient, leading to low prediction accuracy in those areas, and (2) they usually adopt a shallow, fixed fusion strategy for flight conditions, ignoring the fact that different flight conditions influence different geometric regions to varying degrees, which limits accuracy improvement under multiple flight conditions. To tackle these two challenges, we propose a multi-scale, condition-adaptive deep-learning framework. First, to cope with complex-geometry representation, we design the Laplacian Geometric Feature Extractor. It employs discrete Laplacian operators to compute additional local features, markedly enhancing the model's perception of intricate geometries. Second, to achieve deep and adaptive fusion between flight conditions and geometric shapes, we build a multi-scale conditional-fusion pipeline: the Point-level Conditional Fusion module dynamically adjusts the influence of flight conditions on every single point, enabling fine-grained interaction, while the Global Conditional Fusion module optimizes the interaction between global geometry and flight conditions at a coarse scale, allowing condition effects to be adaptively tuned for different shapes. Experiments on four pressure-prediction datasets show that, under various aircraft geometries and flight conditions, the proposed framework achieves relative errors of 6.45%, 5.49%, 7.14%, and 2.88%, outperforming existing deep-learning methods. Ablation studies and fine-tuning experiments further verify the effectiveness, generalization capability, and transferability of our approach.
Xue et al. (Thu,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: