BP神经网络用于火箭增压输送系统故障检测

BP neural network applied in fault detection of rocket pressurized delivery system

增压输送系统是火箭产生推力的核心系统,其工作可靠性直接关系到运载火箭安全性。以某型号运载火箭第三级作为故障检测对象,将遥测数据抽样和预处理,降低数据规模和提高故障检测效率,模拟常见故障和偶发故障扩大故障数据规模,使得故障数据和正常数据均衡,通过添加归一化时间序列区分火箭工况。经处理后的遥测数据用于训练BP神经网络,故障检测结果表明:BP神经网络可以利用火箭遥测数据对增压输送系统进行有效的故障检测,正常数据误报率和故障数据漏报率均较低,且可以满足实时故障检测需求。

The pressurized delivery system is the core system of rocket to generate thrust, and its working reliability is directly related to the safety of launch vehicle. Taking the third stage of a certain type of launch vehicle as the fault detection object, the telemetry data is sampled and preprocessed to reduce the data scale and improve the fault detection efficiency.By simulating common faults and occasional faults to expand the fault data scale, the fault data and normal data are balanced, and the rocket working conditions are distinguished by adding normalized time series. The processed telemetry data is used to train BP neural network.The fault detection results show that BP neural network can effectively detect the fault of pressurized delivery system by using therocket telemetry data, and the false alarm rate of normal data and false alarm rate of fault data are both low, and it can meet the real-time fault detection requirements.