Research On Fault Diagnosis Technology Of Turbofan Aeroengine Based On Attention Mechanism

As the power core of aviation equipment,aeroengine is the epitome of aviation technology and industrial accumulation.Due to the bad working environment of engine,the performance of engine is often degraded or even shut down,which seriously threatens its normal operation and brings great harm and economic losses.Thus the fault diagnosis of aero-engine is helpful to improve the reliability of aero-engine.However,aero-engine is a nonlinear and multivariable complex thermal system,so the fault diagnosis technology of aero-engine is not mature and still needs to be explored.Aeroengine fault diagnosis can be roughly divided into fault diagnosis based on thermodynamic model and data-driven fault diagnosis.Data-driven fault diagnosis has become a common fault diagnosis algorithm for aero-engines due to its good approximation performance and no prior information construction.However,most of the current data-driven diagnostic methods still have limitations,such as only diagnosing the stable state points and discarding the time dimension information of the data,which leads to the poor diagnostic accuracy in the field data.This paper focuses on the data driven fault diagnosis technology of aeroengine: The main work is to establish the thermodynamic model of aeroengine and obtain the simulation data set according to the characteristics of implanted faults.After verification,the error between the established thermodynamic model and the actual data is not more than 1.87%,which meets the accuracy requirement.And a neural network model which contains attention mechanism is trained on simulation data set.Compared with other neural networks,the neural network based on attention mechanism model improves the overall accuracy by 3%,which not only can recognize the dynamic characteristics,but also has stronger fault feature extraction capability.Then transfer learning algorithm is used to improve the generalization ability of the trained network model.In the absence of all fault data,the diagnostic accuracy of aero-engines with slightly degraded components decreased by only about 1%.Transfer learning makes it have better application value,which provides new ideas and attempts for fault diagnosis of aeroengine.