火箭发动机基于神经网络非线性辨识的故障检测

应用神经网络方法，提出了一种液体火箭发动机故障实时检测算法。神经网络采用非线性辨识技术贴近发动机的工作过程，并输出包合发动机故障信息的辨识误差信号。若辨识误差变大超过一定阈值，检测逻辑就预报发动机故障。在发动机启动阶段离线训练神经网络，在发动机稳态过程可以采用离线或在线学习算法。实验研究表明神经网络可以成功地应用于大型泵压式液体火箭发动机的故障检测。

ROCKET ENGINE FAULT DETECTION BASED ON NONLINEAR IDENTIFICATION OF NEURAL NETWORK

sing the method of an artificial neural network, a real-time algorithm is proposed for the fault detection of a liquid propellant rocket engine in this paper. Neural network used for fault detection can approximate the working process of rocket engine by the way of nonlinear identification technology,and it can also generate an identified residual signal that contains the engine fault message at the same time. If the identified residual signal becomes bigger than a given threshold, then there occurs an engine fault. Neural network can be trained for off-lime condition in all flight or for on-line condition in the stabilizing process of the rocket engine running. The experimental research has shown that the artificial neural network has been successful employed in the fault detection of the large turbo-pump liquid rocket engine.