Application Study Of Transfer Learning In Early Fault Diagnosis Of Aero-engine Rotor System

The aero engine is the core part of the aircraft,its normal and stable operation is directly related to the flight safety of the aircraft.Due to its complicated structure,especially the rotor system working under the conditions of high temperature,high pressure and high speed for a long time,it is subject to the coupling of multiple physical fields such as force field,flow field,temperature field,etc.The vibration problem is very complicated and it is easy to cause the rotor system to malfunction,resulting in accidents.Therefore,the study of early fault diagnosis of aero engine is of great significance to avoid flight accidents,avoid economic losses and improve the stability and safety of aero-engines.The current fault diagnosis is mainly based on mathematical models,signal processing,and intelligent diagnosis methods.However,early fault signals of aero engine are submerged in strong noise,which results in weak fault signals,difficulty in extraction and difficulty in collecting early fault samples.In this paper,a feature extraction method based on stack sparse self-coding and principal component analysis is proposed,which is combined with migration learning method to build a fault diagnosis network model based on migration learning,and then a set of fault diagnosis system is developed.The main research contents are as follows:(1)Aiming at the problems of weak early fault signals and difficult to extract fault features,the sparse self-encoding principle was studied and its limitations were improved.A feature extraction method combining stacked sparse self-encoding and principal component analysis was proposed and used by rolling bearing data of the test bench verified its feature extraction performance.The experimental results show that the feature extraction performance of this method is better and it has more advantages compared with traditional methods.(2)Aiming at the problem of lack of early failure samples,this paper studied the transfer learning theory and different transfer learning methods,combined with the subject,studied the maximum mean difference algorithm in transfer learning and applied it to the fault diagnosis network to construct fault diagnosis network model based on transfer learning.(3)Taking rolling bearing data and aero-engine vibration data as the objects,the faultdiagnosis model is established to identify the fault.The influence of different network parameters on their diagnosis results is studied separately and compared with traditional methods and different diagnostic network models.The experimental results show that the fault learning network based on migration learning constructed in this paper has more advantages and higher diagnostic accuracy than traditional methods and other models.(4)Based on the software platforms of LabVIEW and MATLAB,a fault diagnosis system integrating feature extraction and fault classification and integration was developed by mixed programming.It makes the fault diagnosis process more convenient and intuitive,easy to operate and follow-up research.The validity of the system is verified by the system running test.