【摘要】 基于深度迁移学习优化的民机故障诊断模型有效解决变工况民机故障诊断任务，进而保障民机部件在复杂工况下的故障诊断任务需求。然而，民机的高安全性需求造成设备故障样本稀缺，进而导致变工况故障诊断任务的高置信度故障样本严重不足且影响故障模型推断。针对上述研究短板，提出了基于条件扩散的故障诊断算法(condition diffusion based fault diagnosis, CDFD)。首先，结合去噪扩散生成模型对有限的高置信度故障样本进行条件生成增广，解决因样本缺失导致的模型过拟合。同时，针对现有条件扩散技术仅关注样本的条件分布推断的短板，联合考虑故障样本的生成与决策进行耦合优化，有效保障故障样本的生成质量，进而显著提升故障诊断模型的决策泛化能力。试验环节分别在仿真与真实故障数据上进行试验验证，论证了所提出算法在解决真实民机故障任务的高效性。更多还原

【Abstract】 Fault diagnosis models optimized through deep transfer learning have proven effective in addressing civil aircraft fault diagnosis tasks under variable working conditions, ensuring component reliability in complex operational environments. However, the scarcity of high-confidence fault samples, particularly under varying conditions, due to the stringent safety requirements in civil aviation hinders the model’s inference capabilities and increases the risk of overfitting. To overcome these challenges, a condition diffusion based fault diagnosis(CDFD) algorithm was proposed. In the algorithm, a denoising diffusion model was integrated to conditionally generate high-confidence fault samples, thereby alleviating overfitting caused by sample scarcity. Unlike traditional diffusion methods that focus solely on sample distribution inference, the CDFD algorithm couples fault sample generation with decision-making optimization, ensuring the quality of generated samples and significantly enhancing the diagnostic model’s generalization. The experimental validation on both simulated and real-world fault data demonstrates the efficiency of the proposed algorithm in handling real civil aircraft fault diagnosis tasks.更多还原