基于GSP仿真和SDAE的航空发动机故障诊断

为了深入研究航空发动机故障机理,提出基于航空燃气涡轮发动机性能仿真软件(GSP)和堆栈降噪自编码器(SDAE)的航空发动机故障诊断方法。通过GSP性能仿真方法模拟发动机在不同设计参数下的部件故障,并得到对应的运行状态参数;从每种故障类型下的长时间序列的状态参数中提取出向量化的曲线特征,构成故障样本;将故障样本带入SDAE模型中进行深度特征提取,经过前向传播和反向微调得到训练好的模型用于发动机故障诊断。结果表明:GSP能够通过参数更改来模拟微弱故障下的状态参数,从而构建多故障样本集;SDAE的重构误差和反向传播误差能够快速收敛到较小值,SDAE的故障诊断正确率为99.5%;与深度信念网络(DBN)、人工神经网络(ANN)以及经典机器学习方法支持向量机(SVM)相比,SDAE的故障分类正确率分别提高了0.8%、6.9%和10.1%。

Aeroengine Fault Diagnosis Based on GSP Simulation and SDAE

In order to deeply study the aeroengine fault mechanism,an aeroengine fault diagnosis method based on Gas turbine Simulation Program(GSP)and Stacked Denoising Autoencoder(SDAE)was proposed.The GSP performance simulation method was used to simulate the failure of aeroengine components under different design parameters,and the corresponding operation state parameters were obtained.The vectorized curve features were extracted from the state parameters of the long time series under each fault type to form the fault samples.The failure samples were brought into the SDAE model for deep feature extraction.After forward propagation and reverse fine tuning,the trained model was used for aeroengine fault diagnosis.The results show that GSP can simulate the state parameters under weak fault through parameter change,so as to construct multi fault sample set.The reconstruction error and back propagation error of SDAE can quickly converge to a small value,and the fault diagnosis accuracy of SDAE is 99.5%.Compared with Deep Belief Network(DBN),Artificial Neural Network(ANN)and classical machine learning method Support Vector Machine(SVM),the fault classification accuracy of SDAE is improved by 0.8%,6.9%and 10.1%respectively.