Study On Health Prediction Method Of Key Components Of Gas Turbine

As the power device of an aircraft,the health state of an aero gas turbine has a great influence on the aircraft's normal completion of the flight mission.In this paper,the key part of a military aircraft's gas turbine main pump is taken as the research object,and its health prediction method is studied,to improve the reliability of gas turbine.Aim at de-noising problem for the health monitoring parameter exhaust gas temperature margin(EGTM)of the main pump of gas turbine,In this paper,a de-noising method combining EMD with SVD is proposed.In this method,the original monitoring signal is decomposed by EMD to obtain the IMF and RES components.The trending component is extracted from the original signal by selecting the appropriate time scale,then the residual signal is denoised by SVD method and the signal is reconstructed thus achieving the signal recovery and noise removal.Research on health prediction method for main pump of gas turbines,single health prediction models of key components of gas turbine based on LSSVM,WNN and ARMA are established respectively.The results show that the three prediction models can effectively predict the health of gas turbines.At the same time,a combination forecasting method based on LSSVM-WNN is also proposed in this paper,the prediction precision of combination prediction method is higher,the LSSVM-WNN combination forecasting method combined with the advantages of LSSVM and WNN,LSSVM-WNN combined forecasting method not only has faster convergence speed,higher accuracy and advantage,and strong approximation ability and generalization ability.In order to get better results,particle swarm optimization algorithm is introduced to optimize LSSVM and ARMA to optimize the parameters,and also can improve the forecast precision.This paper also introduces the fuzzy integral algorithm to fuse the two optimized prediction models at the decision-making level.The average prediction error after the fusion reduces to 0.23% and the maximum relative error decreases to 0.34%.