火箭发动机涡轮泵故障检测及预测算法研究

Research of Fault Detection and Prediction Algorithm for Rocket Engine Turbopump

涡轮泵作为液体火箭发动机的核心部件,恶劣的工作环境和极高的转速使其易发生组件断裂、烧蚀等问题。为了对液体火箭发动机的涡轮泵进行健康管理,提出针对某型液体火箭发动机涡轮泵的数据驱动故障检测、故障预测及健康状态评估方法。在某型液体火箭发动机试车数据集上,通过对涡轮泵轴、径、切向振动数据进行对应的时域、频域特征处理后,送入训练好的ResNet网络、自主设计的图像特征识别算法以及退化模式线性回归模型,分别实现了对该型液体火箭发动机涡轮泵的故障检测、预测及健康状态评估,具有较高的准确性。

As the core component of liquid rocket engine,turbopump is prone to component fracture and ablation due to its harsh working environment and extremely high rotating speed.In order to carry out health management for the turbopump of liquid rocket engine,we propose data-driven fault detection,fault prediction and health evaluation methods for a certain type of liquid rocket engine turbopump.On the test data set of a certain type of liquid rocket engine,after processing the corresponding time-domain and frequency-domain features of the turbopump shaft,radial and tangential vibration data,we feed them into the trained residual neural network,the self designed image feature recognition algorithm and the degenerate mode linear regression model to implement high accuracy fault detection,fault prediction and health evaluation of liquid rocket engine turbopump,respectively.