基于飞行过程数据的航空发动机故障诊断方法研究

Aircraft Engine Fault Diagnosis Based on Flight Process Data

关键词:  航空发动机  飞行过程数据  故障诊断  GRU网络  深度神经网络

Key words:  Aircraft engine  Flight process data  Fault diagnosis  GRU network  Deep neural network

摘要:

针对航空发动机飞行过程数据，结合门控循环单元（GRU）动态网络和深度神经网络（DNN），提出了一种数据驱动的航空发动机故障诊断结构。首先，从飞行数据中抽取发动机健康数据，并通过一组GRU网络建立发动机在健康状态下的动态模型。其次，通过GRU动态模型的预测值与真实测量信号生成残差信号，残差信号作为DNN网络的输入预测发动机健康参数。最后，通过诊断决策模块实现对发动机的故障检测与识别。使用仿真生成的真实飞行工况数据集对提出的故障诊断系统进行了验证。结果表明，相比于直接使用传感器测量数据，基于GRU网络的残差结构能够大幅提升故障检测和识别性能，故障检测和识别准确率分别可达96.51%和95.06%，并且对训练数据样本数量的依赖性较小，较少的训练样本也能获得很好的预测结果。

Abstract:

Aiming at the data of the aircraft engine flight process， a data-driven aircraft engine fault diagnosis structure is proposed by combining the gated recurrent unit （GRU） dynamic network and the deep neural network （DNN）. Firstly， the engine health data was extracted from the flight data， and the dynamic model of the engine in a healthy state was established through a group of GRU networks. Secondly， the residual signal was generated by the predicted value of the GRU dynamic models and the real measurement signal， and the residual signal was used as the input of the DNN network to predict the engine health parameters. Finally， the engine fault detection and identification were realized by the diagnostic decision module. The proposed fault diagnosis system was verified by using the real flight condition data set of the engine generated by simulation. The results show that compared with the direct use of sensor measurement data， the residual structure based on GRU network can greatly improve the performance of fault detection and identification， and the fault detection accuracy and fault identification accuracy can reach 96.51% and 95.06%. The dependence on the number of training data samples is small， and good prediction results can be obtained with few training samples.