Aero Engine Gas Path Fault Diagnosis And Prognosis Based On Machine Learning Methods

Aero engine fault diagnosis and prognostic technology is the key of engine prognostic and health management(PHM)system,which guarantees the flight safety and reduces maintenance costs.In this thesis,turbofan engine gas path fault diagnosis and prognosis based on extreme learning machine(ELM)and deep learning methods are mainly studied.To deal with the problem of the instability of ELM,some relevant improvements are proposed.And the main contents are as follows:A novel extreme learning model based on restricted Boltzmann ELM,constructs a feature mapping and recursively tune the weights between input neurons and hidden neurons,which leads to more accurate and stable model.The proposed methodology is evaluated on UCI benchmark datasets for classification issue,and then extended to gas path fault diagnosis for a turbofan engine.The experimental results confirm the superiority to plain ELM.This paper develops a restricted Boltzmann strategy to learn topological parameters in ELM input layer,and it is then applied to aero-engine gas path fault diagnosis.Aimed at online sequential learning problem,this thesis proposed an adaptive ensemble learning strategy for ensemble of OS-ELM algorithm(EOS-ELM)based on linear Kalman filtering algorithm.The coefficient vector tuned according to the regression accuracy of each individual at the latest several steps in the KEOS-ELM network.Furthermore,the analytical redundancy of sensors is proposed based on KEOS-ELM to improve the efficiency of drift faults diagnosis.To further improve the stability of OS-ELM,a new training approach of the OS-ELM using Kalman filter called KFOS-ELM is proposed,and state propagation is combined into extreme learning process to obtain the OS-ELM's topological parameters.Besides,an enhanced multi-sensor prognostic model based on KFOS-ELM and logistic regression model was designed for performance degradation prognostic to effectually predict the failure time of aero engine.Finally,the remaining useful life prediction of aero engine was presented based on long short-term memory(LSTM)neural network and auto-regression-intergrated-moving-average(ARIMA)model.The prediction method estabilished the health indicator model base based on historical degradation data of aero engine and evaluated the degradation state according to the sensor parameters predicted by ARIMA to calculate the remaining useful life and probability density distribution.