对于存在气动失谐的涵道风扇，传统基于固定节径假设的空间截断模型难以表征模态局部化响应，因而不足以准确评估其颤振稳定性。为突破这一限制，本文采用时域流固耦合方法对其颤振行为进行分析。为了降低流固耦合高昂计算成本，本文提出了一种基于GPU加速的异构并行框架。与96核CPU计算相比，该框架基于8张GPU可实现约47倍挂钟时间加速，计算费仅为CPU计算的5.2%。在此基础上，本文分别对均匀叶尖间隙和叶尖间隙失谐条件下的涵道风扇开展流固耦合颤振分析。分析表明，非均匀叶尖间隙会显著改变叶片间非定常气动耦合关系，使振动能量的周向分布更加不均匀，并诱发模态局部化响应，恶化气弹稳定性。上述结果进一步表明，全环时域流固耦合方法能够有效捕捉气动失谐条件下的颤振特征，可为失谐风扇颤振稳定性评估提供有效手段。

For ducted fans with aerodynamic mistuning, the conventional spatial truncation model based on a fixed nodal-diameter assumption cannot be used to characterize mode-localized responses, and is therefore insufficient for accurately evaluating flutter stability. To overcome this limitation, this study employs a time-domain fluid–structure coupled method to analyze the flutter behavior. To reduce the high computational cost of fluid–structure coupled analyses, a GPU-accelerated heterogeneous parallel framework is proposed. Compared with 96-core CPU computation, the proposed framework achieves a speedup factor of 47 in wall-clock time using eight GPU cards, while its monetary computing cost is only 5.2% of that of the CPU-based computation. Based on this framework, fluid–structure coupled flutter analyses are performed for a ducted fan under uniform tip clearance and tip-clearance mistuning conditions. The results show that non-uniform tip clearance significantly alters the unsteady aerodynamic coupling among blades, leading to a more non-uniform distribution of vibrating energy, inducing mode localization, and worsening the aeroelastic instability. These findings further demonstrate that full-annulus time-domain fluid–structure coupled analysis can effectively capture local aeroelastic response characteristics under rotor geometric mistuning, providing an effective approach for flutter stability assessment of mistuned fans.