应用动态神经网络在线辨识方法，提出了一种液体火箭发动机故障实时检测与分离基本系统。检测逻辑通过度量包含发动机故障信息的辨识残差信号实现火箭发动机故障检测，故障分离通过分析辨识误差相关函数的不同空间特征来实现。仿真研究表明动态神经网络可成功地应用于泵压式液体火箭发动机故障检测与分离。 On the basis of dynamic online neural network identification method, a basic system for real-time detection and separation of liquid rocket engine fault is proposed. The detection logic realizes the rocket engine fault detection by measuring the identification residual signal containing the engine fault information. The fault separation is realized by analyzing and identifying the different spatial characteristics of the error correlation function. Simulation results show that dynamic neural network can be successfully applied to the detection and separation of pump-type liquid propellant rocket engine.