Fault Diagnosis Method Of Gearbox Based On Deep Belief Networks

As the key component of power and torque transmission in oilfield,once the gearbox fault occurs,it will directly affect the normal operation of the equipment.Accurate diagnosis of mechanical faults is significant to the safety,stability,continuous and efficient production of oilfields.The practice of fault diagnosis shows that the gearbox fault is not the single fault,but the compound faults.Different fault signals are coupled and interfered with each other,which brings many challenges to fault diagnosis.Therefore,it is necessary to carry out composite fault diagnosis.In this paper,fault signal denoising,fault feature extraction,and compound fault diagnosis are studied.The main contents are as follows:(1)A gearbox fault denoising method based on adaptive variational mode decomposition(VMD)and maximum correlation kurtosis deconvolution(MCKD)is proposed.The fault signal is decomposed by adaptive VMD,the decomposed mode is filtered and denoised by MCKD,and the optimal filter length is determined by an equal step search to achieve the best noise reduction effect.The method is applied to the gearbox fault noise reduction,and the impact fault feature is prominent,which proves that the method has good noise reduction effect.(2)A gearbox intelligent diagnosis model based on improved DBN is proposed.By improving the activation function in the process of fine-tuning,DBN model solves the problem of random gradient disappearance and accelerates the convergence speed.Through applying in the oilfield gearbox fault diagnosis,self-adaptive fault features extraction,modes recognition and classification are realized.The descending dimension and visualization of feature vectors are used by t-SNE,which verifies that the method has higher classification accuracy and stronger adaptability by comparative analysis.(3)A gearbox compound fault diagnosis method based on VMD-MCKD and DBN is proposed.The compound fault signal is decomposed by VMD for noise reduction.According to the different fault characteristic frequencies of gear and rolling bearing,the signal is filtered by MCKD to realize the separation of compound faults.The separated fault signals are input into the improved DBN model,which realize the composite fault diagnosis of gearbox.Different feature data sets are constructed,and the DBN model is trained respectively.The results show that the compound faults has good separation effect,and the method has its superiority and stability.