基于MEF-MSCNN-LSTM模型的飞机电控舵面故障检测研究

Research on Fault Detection of Aircraft Electrically Controlled Rudder Surface Based on MEF-MSCNN-LSTM

摘要:

随着航空系统智能化水平的提升，飞行器对飞控系统的实时响应与自主诊断能力提出更高要求；电控舵面作为关键执行部件，其故障将直接影响飞行稳定与安全；为提升识别精度与响应效率，研究构建融合多尺度卷积与长短期记忆网络的故障检测模型，并引入多阶段信号降噪机制以优化特征提取；模型基于多源飞行数据训练，精准捕捉电压、电流与偏转角等关键序列特征的演化模式；实验表明，模型在完全训练后准确率达96.7%，漏报率低于1.08%，可在异常发生前2.2 s实现高置信度预警，具备出色的时效性与稳定性；同时在典型巡航工况下识别出微弱间歇性迟滞故障，分类置信度为0.974，体现了良好的早期检测能力；研究验证了该模型在复杂非线性场景中的鲁棒性与实用性，为飞控系统故障诊断由被动响应向智能预测转变提供了技术支撑。

Abstract:

With the advancement of intelligent aviation systems, modern aircraft demand higher levels of real-time responsiveness and autonomous diagnostic capability from flight control systems. As a critical actuator component, failures in the electrically controlled rudder surface can directly affect flight stability and safety. To enhance detection accuracy and response efficiency, this study proposes a fault detection model combining multi-scale convolutional neural networks (MSCNN) and long short-term memory (LSTM) networks. A multi-stage signal denoising mechanism is integrated to optimize feature extraction. The model is trained on multi-source flight data to accurately capture the temporal evolution patterns of key sequences such as voltage, current, and deflection angle. Experimental results show that the model achieves an accuracy of 96.7% and a miss rate below 1.08% after full training. It can generate high-confidence early warnings 2.2 seconds before the occurrence of abnormal events, demonstrating excellent timeliness and stability. Additionally, the model successfully identifies weak intermittent lag faults during typical cruise conditions with a classification confidence of 0.974, indicating strong early detection capability. The study confirms the robustness and practical applicability of the model under complex nonlinear scenarios and provides technical support for shifting fault diagnosis in flight control systems from passive response to intelligent prediction.