Aerodynamic Reduced Order Model For Blade To Upstream Wake Based On Neural Network

The design of modern aero-engines is increasingly developing towards the goals of high thrust-to-weight ratio,high efficiency,high reliability and low fuel consumption.In order to meet its development needs,compressors have always been developing along the direction of high pressure ratio,high speed and high efficiency,its blades has also become lighter and thinner,the axial length has been continuously reduced,and the structure has become more compact,making the interaction between the upstream and downstream inside the engine more intense.In an aero-engine,the high-speed relative rotation between the upstream and downstream blades causes the downstream blades to undergo aeroelastic vibration under periodic disturbances such as wakes and shock waves from upstream blade rows,which is one of the important factors that cause fatigue damage to the blades,and seriously threatens the safety and reliability of the engine.The high-performance numerical simulation method is the most reliable method to study the aeroelastic vibration of the blade under the internal wake excitation of the engine,but its calculation is time-consuming and expensive.With the continuous deepening of the application of neural networks in the field of aeroelasticity,this paper establishes a wake-driven blade aerodynamics reduced-order model based on neural networks,which provides a fast calculation method for studying the aeroelastic vibration of downstream blades with the effect of upstream and downstream interference.In this paper,firstly,the reduced order model of blade aerodynamic force under wake excitation is established based on BP neural network.BP neural network with input delay is used for training modeling,and the blade aerodynamic force under random wake excitation of different amplitudes is predicted,which is compared with the CFD calculation results.The results show that the reduced order model with input delay,which is trained by random wake excitation,can better predict the aerodynamic response of blades under different amplitudes of random wake excitation,but the prediction ability and generalization ability of the model are limited.Subsequently,it was proposed to establish a model of blade aerodynamic reduction under wake excitation based on NARX neural network,and to predict the blade aerodynamic response under different wake excitation,and compared with the CFD calculation results.The results show:The NARX neural network model of blade aerodynamic reduction,which is trained by periodic signal,can predict the blade aerodynamic response under periodic wake excitation rapidly and accurately,and the blade aerodynamic response under simple harmonic wake excitation can also be predicted well,but the prediction accuracy is relatively low.In this paper,it was proposed to establish a reduced order model of blade aerodynamic force under wake excitation based on RBF neural network.RBF neural network with input delay is used for training modeling,and the aerodynamic response of blade under wake excitation with different amplitude and different steady-state total pressure under period wake excitation is predicted,and compared with CFD calculation results,the analysis shows:The RBF neural network model with input delay,which is trained by periodic signal,can predict the aerodynamic response of the blade under the periodic wake excitation of different amplitudes with high accuracy,and the prediction accuracy of the aerodynamic response of the blade under the periodic wake excitation of different steady-state total inlet pressure is low with limited generalization ability.Finally,the recursive RBF neural network was used to establish the aerodynamic order reduction model of the blade under wake excitation,and the blade aerodynamic response under different wake excitation is predicted,and compared with the CFD calculation results.The RRBF neural network blade aerodynamic order reduction model obtained by periodic signal training can not only quickly and accurately predict the aerodynamic response of the blade under different wake amplitudes and different steady-state inlet pressures,but also quickly and accurately predict different aerodynamic responses.The blade aerodynamic response under the excitation of simple harmonic wake with different amplitude and total steady-state inlet pressure has strong predictive ability and generalization ability.