【摘要】 为实现故障辨识、定位与估计等功能，同时又面向导弹舵回路诊断的实时性与准确性需求，提出一种基于长短记忆网络-自编码器(LSTM自编码器)与长短记忆网络-全卷积网络(LSTM-FCN)算法相结合的诊断框架。采用LSTM自编码器学习数据间的时序关系并建立正常飞行状态的预测模型。将预测模型输出与当前飞行状态比较，生成残差形成故障特征。基于LSTM-FCN网络建立了集故障辨识、定位与估计功能于一体的导弹舵回路诊断模型；并引入压缩激发(SE)模块以增强诊断算法对故障特征的敏感性，利用舵回路半物理仿真平台对所提算法进行了验证。试验表明：针对气动参数与质量不确定性条件的导弹舵回路，所提算法提升了故障诊断的准确性、实时性和鲁棒性。更多还原

【Abstract】 In order to realize multiple functions of fault identification, localization and estimation, while meeting the requirements for real-time diagnostic loops for missile rudder systems, a diagnostic framework based on a combination of a long-short-term memory network-autoencoder(LSTM autoencoder) and a long-short-term memory network-full convolutional network(LSTM-FCN) algorithm is proposed. The LSTM autoencoder is used to learn the temporal relationships of the data and to build a predictive model of normal flight state data. The model output is used to generate residuals to form the fault feature input. The LSTM-FCN network is used to form a diagnostic model of the missile rudder loop system that integrates fault identification, localization and estimation functions, and the sensitivity of the diagnostic framework to fault features is enhanced by the introduction of the squeeze-and-excitation(SE) block. The proposed algorithm is validated using a rudder loop semi-physical simulation platform, which shows that the proposed algorithm improves the accuracy and real-time performance of fault diagnosis, as well as the robustness under the uncertainty of aerodynamic parameters and quality.更多还原