Research On Fault Diagnosis Of Marine Gearbox Based On Deep Belief Network

With the implementation of marine economic powerful nation strategy in recent years,the shipbuilding industry has entered a new development stage.The requirement on the reliability of ships and marine equipment is becoming higher.As the main transmission in marine machinery and equipment,the health of the gearbox is closely related to the operating state of the entire mechanical equipment system.If the health of the gearbox can be evaluated in time and the type of failure is accurately identified,ship safety can be improved.To a certain extent,it can also guide post-maintenance and overhaul work.Therefore,gearbox fault diagnosis has gradually become an important research direction in ship machinery.Aiming at the difficulty in extracting fault features in the gearbox vibration signal and the feature selection in traditional diagnostic methods relying too much on subjective experience,this paper studies the intelligent fault diagnosis methods of the key components in gearbox based on the deep belief network(DBN).In the fault diagnosis methods,the gear box is taken as the research object,and the deep learning technology is used to develop the fault diagnostic models.The main research contents of this paper are as follows:(1)According to the DBN's network framework and basic principles,the training methods of DBN forward greedy learning and backward fine tuning are expounded.Simulation signal is used to explore the influence of DBN main parameters on network classification performance,and a practical DBN hyper-parameter setting strategy is proposed.(2)By analyzing the parts that are prone to failure in the gearbox,fault modes of rolling bearings and gears and their corresponding vibration signal characteristics are discussed,and the vibration model at the gear meshing point is derived.The vibration data acquisition system under the Labview platform is designed.Experiment is carried out on the mechanical failure simulation test bench,and the fault diagnosis data set under different working conditions is obtained.(3)A DBN fault diagnosis method for bearing combining with variational mode decomposition(VMD)algorithm is proposed.By integrating VMD method,energy entropy,time domain frequency domain statistical features and feature selection strategy,sensitive feature vector sets are obtained.This set of feature vectors is trained in the optimized DBN,and the diagnostic model is obtained.Compared with other diagnostic methods,the fault diagnostic model developed in this chapter has higher accuracy and better stability.(4)A fault diagnostic method for gear based on DBN and wavelet denoising is proposed.By using the strong capability of wavelet analysis method on reducing the signal noise and the DBN on reconfiguring the original input data,the wavelet denoised signal is directly input into the DBN for automatic multi-level feature extraction.The fully trained DBN model can better identify and classify gear states under different working conditions.The effectiveness and superiority of the method are verified by comparing traditional diagnostic methods.