Applied Research On Aeroengine Bearing Fault Diagnosis Based On Extreme Learning Machine

Aero-engine bearings are one of the core components of the aero-engine rotor systems.And the state changes of aero-engine bearings will directly affect the performance of aero-engine.The position of aero-engine bearings are difficult to monitor and even if the same bearings but their life is very different.Therefore,many scholars at home and abroad have studied this and carried out real-time monitoring and diagnosis of bearing faults,which not only conducive to mastering the performance of aero-engine,but also making full use of bearings to save maintenance costs.Based on bearing vibration signal and artificial intelligence algorithm to diagnose the bearing fault signal.The article will conduct in-depth research from the following aspects:1)Introduce the mechanism of bearing failure.Some bearing faults that can be used for engineering recurrence will be selected.According to the test requirements,the test equipment platform is built,and the preconditions of test signal acquisition are expounded.2)It is difficult to extract the characteristics of bearing fault vibration signals.Firstly,a nosie reduction method based on the combination of empirical mode decomposition(EMD)and fast independent component analysis(FastICA)is proposed.Secondly,the 12 time-domain characteristic parameters and 14 frequency-domain characteristic parameters are extracted from the bearing vibration signals after noise reduction.Finally,considering the coupling and high dimensionlity between time-domain and frequency-domain characteristic parameters,genetic algorithm(GA)is used for dimensionality reduction,the 8 optimal time-domain characteristic parameters and 6 optimal frequency-domain characteristic parameters are selected.And this optimal parameters are used as input sources in the subsequent diagnosis process.3)Based on optimal parameter feature data set,extreme learning machine(ELM)is used to diagnose bearing fault.In order to better realize bearing fault diagnosis,the data set is normalized to eliminate the dimension and magnitude differences between the data,and the data set is reasonably divided.Considering the in fluence factors of ELM model,the hidden layer neurons and activation functions of hidden layer of ELM model are analyzed.ELM network and some single hidden layer feedforward neural networks(SLFN)are compared and analyzed,which is used to verify the validity of ELM model in bearing fault diagnosis.4)Considering the defect caused by random generate connection weights and threshold of hidden layer of ELM model,GA is proposed to optimize the ELM model.Firstly,considering the some problems in the classical GA,an improved genetic algorithm(IGA)based on crossover and mutation operation is proposed.IGA is used to optimize ELM(IGA-ELM),and IGA-ELM is used to diagnose bearing fault.Secondly,Bloch spherical coordinate genetic algorithm(BQGA)is used to optimize ELM(BQGA-ELM).Considering the BQGA search area limitation problem,chaotic disturbance is introduced to improve BQGA(CBQGA),and CBQGA is used to optimize ELM(CBQGA-ELM).Finally,the optimization method used in this paper is compared with many other algorithms to verify the feasibility of the above method.