Deep Learning-based Baseline Model And Trend Prediction Of Aeroengine Gas Path Parameter

Aeroengine is one of the most core components of the aircraft,and its health will directly affect the safety and reliability of the aircraft.In order to ensure the operational safety and the economic benefits of maintenance of the aeroengine,baseline model and trend prediction of gas path parameters of aeroengines can provide technical support for Airlines.Due to the complexity of the aeroengine,the precise physics-based model is difficult to construct,the intelligent learning model based on data-driven becomes an effective method for baseline model and trend prediction of gas path parameters.The complex operating environment and the process of performance degradation of the aeroengine will bring huge challenge to baseline model and trend prediction of gas path parameters,whose results will affect the judgment of performance status and the precise of maintenance decisions.Therefore,study on the approach of the baseline model and trend prediction of gas path parameters is significant.Based on the monitoring data of the aeroengine,the gas path parameters can be mined and the modeling methods of baseline and trend prediction of gas path parameters will be proposed,then the purpose of overrun monitoring and trend prediction can be realized.The main contents of this paper includes the following three aspects:Firstly,due to the influence of coarse error and noise in the raw data collected by sensors,the method of preprocessing based on monitoring data of the aeroengine is studied.The detection and processing of coarse error based on Laida criterion is proposed.The monitoring data should be corrected after the coarse error is removed.Then,the stability points of each parameter in each cruise phase are extracted to represent the performance status of aeroengines based on the method of sliding window;Pearson correlation is used to extract the critical factors related to the gas path parameters,which will be used to the input vector of the baseline model;Min-Max is employed to achieve the normalization of the input vector in order to eliminate the influence of the dimension between the parameters.Secondly,the gas path parameters of the aeroengine are affected by many QAR parameters,traditional and single intelligent learning model is difficult to extract the nonlinear features and achieve good predi ctive results.Therefore,a hybrid model based on Deep Belief Network(DBN)and Support Vector Regression(S VR)is proposed to establish the baseline model of gas path parameters of the aeroengine.DBN is employed to mine the inherent nonlinear features and invariant structures of the Quick Access Recorder(QAR)data associated with gas path parameters.Then,the nonlinear features extracted by layer-wise pre-training based DBN are input to SVR to predict the values of gas path parameters.The experimental results show that the hybrid model can not only extract the data features of the input vector,but also improve the prediction accuracy and establish a high-precision baseline model.Finally,the accuracy of the traditional prediction methods of time series is insufficient,and it is difficult to realize the problem of parameters prediction based on multivariate inputs.In the paper,the Long-Short Term Memory(LSTM)network is used to predict the future trend of gas path parameters of the aeroengine.The history sequence of gas path parameters and working condition parameters are used as the input vector of the model to predict the future trend of gas path parameters.By considering the working condition parameters,the accuracy of the prediction model can be improved.The research in this thesis can enrich the system of the aeroengine EHM,and it can improve the operational level of the aircraft.In addition,the methods in this thesis has practical significance for airlines to improve the management level of the aeroengine.