Aircraft Engine Fault Prediction And Application Based On Deep Learning

As a highly integrated complex system,aircraft runs in harsh working environment for a long time and may fail at any time.Aircraft engine as the core of the aircraft,but also the power source of the aircraft,if the engine is not reliable,the consequences can not be imagined.In order to prevent the aircraft engine failure accident even plane crash,airlines need to spend huge cost for aircraft maintenance and repair,security against aircraft engine and aircraft engine maintenance budget two seemingly opposing problem,many airlines are looking for more efficient aircraft engine health management and maintenance decision-making,Therefore,it can guarantee the safety and reliability of navigation and take into account the economy of engine maintenance.However,the traditional fault prediction methods have the disadvantages of complex modeling and poor generality,so they cannot accurately predict the faults of aircraft engines.In this paper,the deep learning method will be used to mine the relationship between aircraft engine condition monitoring data and Remaining Useful Life(RUL)of aircraft engine,to predict engine faults,and to provide decision-making basis for engine operation and maintenance,so as to develop a reasonable aircraft engine health management programFor aircraft engine fault prediction is usually based on statistical or physical methods such as creating a model,through analysis of engine operation principle and failure mechanism,create a research and analysis mathematical model,create the process model is not only difficult,complex and for different types of fault or equipment,the lack of generality,unable to realize the multiple fault prediction at the same time.The data in this study comes from the turbofan engine degradation simulation data set in the NASA predictive data repository,which simulates the engine condition monitoring data under different operating conditions and different fault modes.Through the analysis of the data,it is found that the data have the characteristics of large amount of data and high data dimension.According to the feature points,the data were preprocessed,and the Principal Component Analysis(PCA)method was used to extract the feature from the condition monitoring data,so as to remove useless information and noise and improve the accuracy of prediction processing.Three commonly used classification algorithms,Support Vector Regression(SVR),Recurrent Neural Network(RNN)and Long Short-Term Memory(LSTM).Based on the current research status,principal component analysis is proposed to extract features from engine state detection data.Then,a Bi-directional Long-Short Term Memory(BLSTM)network was used to mine the relationship between condition monitoring data and engine residual service life.It provides decision basis for the formulation of engine maintenance and health management program.The conditions monitoring data preprocessing,PCA-BLSTM model building process,training process,super parameter selection and other problems are introduced in detail.The PCA-BLSTM hybrid model to predict the remaining service life of aircraft engine were compared.Support vector regression,long and short-term memory network,and two-way long and short-term memory network were built to compare with the PCA-BLSTM hybrid model,and the results show that the performance of PCA-BLSTM model is better than the previous three methods.Finally,the predictive maintenance strategy is proposed.Taking the failure of aircraft engine compressor as an example,the fault mode of the compressor is analyzed,and the fault severity is classified,and the fault causes and detection methods are analyzed.Based on the Failure Mode and Effects Analysis(FMEA)results,the PCA-BLSTM model is proposed to predict the remaining service life of aircraft engines,and the health status of aircraft engines is classified.Use FMEA method to troubleshoot the cause of the fault and analyze the influence of the fault,and develop reasonable aircraft engine health management strategy.