Research On Fault Diagnosis Method Of Aircraft Engine Based On QAR Data

Flight safety and punctuality are important factors related to the lives and property safety of airlines and passengers,and timely diagnosis of aircraft failures is an important means to greatly improve flight safety and punctuality.In order to monitor the aircraft's health status,major airlines have long installed QAR systems,which use QAR data to indicate the health status of aircraft during flight.This project aims to diagnose 8 types of aircraft engine faults in time by mining 8 types of data related to aircraft engines,and provide a reference for airlines to reasonably arrange aircraft maintenance plans.For this,this project mainly completed the following tasks:1.Select QAR data of 8 attributes such as aircraft engine exhaust temperature(EGT),the purpose of which is to diagnose faults such as fifteen-level bleed air leakage by mining the data.Since the QAR data during the cruise phase of the aircraft is relatively stable,it can more accurately represent the aircraft system status.First,we intercepted the QAR data of the cruise stage using the relevant standards of the EHM software as our sample data through the relevant standards of the manual,and processed the sample data by k-means method for missing values and outliers.2.For the fault diagnosis model,we consider research through the theoretical framework of deep learning,and establish ConvLstm-ELM network models.A Conv Lstm-ELM fault detection model is established.This model uses sliding windows to segment the original QAR data,and reconstructs the segmented data.After that,it uses ConvLstm to extract features from the reconstructed data.In this paper,using Conv Lstm to extract the data makes up for the shortcoming that the LSTM can only extract the temporal features but not the spatial features.Finally,the extracted spatio-temporal features are classified by ELM for fault diagnosis.This paper compares the Conv Lstm-ELM fault diagnosis model with other deep learning models to measure its fault detection performance.In addition,it compares the diagnostic performance and diagnosis time of the spatio-temporal features extracted by Conv Lstm through the Softmax classifier and ELM classifier.ELM takes less time and has higher diagnostic accuracy.In this paper,the Conv Lstm-ELM model was compared with the SVM multi-class classification model and the PS-SVM model to verify the superiority of convlstm-elm in the diagnostic accuracy?3.Optimize the fault detection model from the data perspective to improve the accuracy of engine fault diagnosis.In order to solve the problem of low dimensions and weak spatiality of QAR data,a sparse autoencoder is used to train the original QAR data to obtain high-dimensional encoded data.The high-dimensional encoded data obtained from sparse autoencoder training is used as the input data of ConvLstm-ELM.The diagnostic performance of the AE-Conv Lstm-ELM model and the ConvLstm-ELM model is compared.The diagnostic performance of the fault diagnosis model with the sparse autoencoder is significantly better than that without the sparse autoencoder.