航空学报 > 2025, Vol. 46 Issue (8): 231148-231148   doi: 10.7527/S1000-6893.2024.31148

固体力学与飞行器总体设计

基于AGABP神经网络的运载火箭推力偏移损失故障在线诊断

陈海鹏1,2(), 符文星3,4,5, 闫杰3,4,5   

  1. 1.西北工业大学 航天学院,西安 710072 2.中国运载火箭技术研究院,北京 100076 3.无人飞行器技术全国重点实验室,西安 710072 4.无人机技术集成攻关大平台,西安 710072
    5.西北工业大学 无人系统技术研究院,西安 710072
  • 收稿日期:2024-09-03 修回日期:2024-10-08 接受日期:2024-11-21 出版日期:2024-12-09 发布日期:2024-11-29
  • 通讯作者: 陈海鹏 E-mail:chenhp_hit@163.com
  • 基金资助:
    国家自然科学基金(52232014)

Fault diagnosis of thrust offset loss of launch vehicle based on AGABP neural network

Haipeng CHEN1,2(), Wenxing FU3,4,5, Jie YAN3,4,5   

  1. 1.School of Astronautics,Northwestern Polytechnical University,Xi’an 710072,China
    2.China Academy of Launch Vehicle Technology,Beijing 100076,China
    3.National Key Laboratory of Unmanned Aerial Vehicle Technology,Xi’an 710072,China
    4.Integrated Research and Development Platform of Unmanned Aerial Vehicle Technology,Xi’an 710072,China
    5.Unmanned System Research Institute,Northwestern Polytechnical University,Xi’an 710072,China
  • Received:2024-09-03 Revised:2024-10-08 Accepted:2024-11-21 Online:2024-12-09 Published:2024-11-29
  • Contact: Haipeng CHEN E-mail:chenhp_hit@163.com
  • Supported by:
    National Natural Science Foundation of China(52232014)

摘要:

针对运载火箭动力系统的推力偏移损失故障,提出基于自适应遗传算法反向传播(AGABP)神经网络的推力故障在线检测和诊断方法,仅依据箭载传感器测量得到的火箭运动信息,实现对推力损失故障的低延迟、高精度在线检测和诊断。首先根据我国运载火箭数据及推力故障类型进行六自由度建模,并将过载和视加速度等对故障敏感的历史状态信息作为输入进行网络训练;其次通过自适应遗传算法调整BP神经网络中初始权重,从而得到优化后的网络参数;最后对得到的运载火箭推力偏移损失故障在线诊断模型进行六自由度在线仿真验证。数值仿真结果表明,与传统BP网络相比,基于AGABP的方法收敛速度快,迭代次数少,故障定位准确率为96.51%,故障定位延迟在0.1~2 s之间,94.19%的样本预测推力下降程度与实际推力下降程度之差在20%范围内。

关键词: 运载火箭, 动力系统故障, 故障检测与诊断, 自适应遗传算法, 神经网络

Abstract:

To address the thrust deviation loss fault in the launch vehicle’s power system, an online detection and diagnosis method for thrust faults based on the Adaptive Genetic Algorithm-based Back Propagation (AGABP) neural network is proposed. To achieve low-latency, high-precision online detection and diagnosis of thrust loss faults, this method solely utilizes the rocket motion information measured by onboard sensors. Firstly, a six-degree-of-freedom (6-DOF) modeling is established based on the data and thrust fault types of a certain type of launch vehicle in China. Historical state information sensitive to faults, such as overload and apparent acceleration, was used as inputs for network training. Secondly, the initial weights in the BP neural network are adjusted through the adaptive genetic algorithm to obtain optimized network parameters. Finally, the resulting online diagnostic model for thrust deviation loss faults in launch vehicles is verified through 6-DOF online simulations. Numerical simulation results indicate that compared with the traditional BP network, the AGABP-based method exhibits faster convergence speed with fewer iteration generations. The accuracy of fault location is 96.51%, the fault location delay is between 0.1 s and 2 s, and the difference between the predicted and actual thrust reduction degree is within 20% for 94.19% of the samples.

Key words: launch vehicle, power system fault, fault detection and diagnosis, adaptive genetic algorithm, neural network

中图分类号: