Research On Fault Prognostic Of Electro-hydraulic Servo Valve Based On CNN And GRU

Electro-hydraulic servo control system who has the advantages of high control precision,fast response speed and large driving power,has been widely used in aerospace.As the key component of the control system,electro-hydraulic servo valve’s degradation greatly affects the performance and safety of the control system.Thus,this thesis studies the prognostic of electro-hydraulic servo valve,including remaining useful life(RUL)prediction and degradation type prediction.This thesis established the mathematical model of electro-hydraulic servo valve,made the prognostic dataset,designed and improved the prognostic algorithms of electrohydraulic servo valve based on artificial neural network,finally verified the feasibility and superiority of the proposed neural network structure.The main work and research results of this thesis are as follows:1.Establish the mathematical model of nozzle-flapper servo valve.According to the three degradation types,make the prognostic dataset.2.The artificial neural network is introduced into the prognostic of electro-hydraulic servo valve.A new kind of neural network structure named CNN_LSTM/GRU/IGRU,is proposed,which is based on convolutional neural network(CNN),long short term memory(LSTM),gated recurrent unit(GRU),and impoved gated recurrent unit(IGRU).The new neural network structure endows the prognostic algorithm with the ability of automatic extraction,combination of features and the ability of long-short term memory.3.An improved cost function is proposed.This theis introduces cost sensitive for RUL prediction to avoid safety problems caused by RUL overestimation.The fusing of RUL prediction cost and degradation type prediction cost,realizes the reuse of some network structures.4.Based on the advantages and disadvantages of LSTM and GRU,IGUR is proposed by modifying two structures of GRU.The overall prediction performance of CNN_IGUR structure is higher than that of CNN_LSTM/GRU.Finally,use the progonostic dataset to verify the feasibility and superiority of the newly proposed neural network structure in the prognostic of servo valve.