论文部分内容阅读
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。
Applying the neural network method, a real-time fault detection algorithm of liquid rocket engine is proposed. The neural network uses the non-linear identification technology to approach the working process of the engine, and outputs the identification error signal that includes the fault information of the engine. If the identification error becomes larger than a certain threshold, the detection logic to predict engine failure. The neural network is trained off-line during the engine start-up phase and off-line or online learning algorithms can be used in the steady-state process of the engine. Experimental results show that the neural network can be successfully applied to the fault detection of large-scale pump-type liquid propellant rocket engines.