Reserach On Fault Diagnosis Method Of Planetary Gearbox Based On Deep Belief Network

Planetary gearboxes were widely used in wind turbine,marine and other large mechanical equipment machines.The harsh operation conditions caused the problem that planetary gearboxes were easily broken under alternating loads.At the same time,the number of components of the planetary gearboxs was large,the meshing between the gears was frequent,and the vibrations were coupled with each other to cause the vibration to exhibit obvious nonlinearity,so it was difficult to extract the fault characteristic information.Therefore,the planetary gearbox fault diagnosis can detect and eliminate the sudden problems in the operation process in time,and effectively prevent major accidents.It had important research significance and engineering value.Traditional fault diagnosis methods relied on signal processing and diagnostic experience to extract fault features,and then used machine learning models for diagnosis.The matching degree between the extracted features and the applied pattern recognition algorithms was difficult to evaluate.The two were not organic integration,affecting the reliability of the diagnosis.results.The Deep Belief Network organically combined fault feature learning with pattern recognition,and had been widely used in the field of fault diagnosis,and had achieved good results.However,most of the solutions were fault diagnosis under one working condition.When multiple working conditions were alternated,fault information coupling and training samples were limited,there was no good diagnostic effect.Based on this,the fault diagnosis method of planetary gearbox based on Deep Belief Network was studied.The main research work of this paper was as follows:Aiming at the problem of adaptive feature extraction and intelligent diagnosis of planetary gearboxes,a planetary gearbox fault diagnosis method was proposed based on Deep Belief Network(DBN).Firstly,in order to fully present the feature of the fault planetary gearbox signal,prevent the loss of information and facilitate feature selflearning,the spectrum of the original signal was used as the input of the deep belief network.Then,we used the multi-layer self-learning network of DBN to hierarchically express the input signal,and we transformed the low-level features layer by layer to form abstract deep feature to obtain the distributed feature expression of the original signal.Finally,softmax multi-classifier was added at the end of the feature output layer to integrate the adaptive feature extraction and fault pattern recognition to realize the intelligent diagnosis of planetary gearbox faults.The feasibility and effectiveness of the proposed method were proved by the fault diagnosis experiments of planetary gearboxes with different faults and two working conditions.Aiming at the problem of the data obtained for the planetary gearbox fault diagnosis is difficult that the training and test data are independent and the same and the training data is sufficient,resulting in poor diagnostic results,a planetary gearbox fault diagnosis method based on Deep Belief Network migration learning was proposed.Firstly,the original signal spectrum of the auxiliary marker data was used as the input of the DBN network,and the input signal is hierarchically expressed by updating the weight and offset value of the network layer by layer to obtain the distributed feature expression,and the DBN pre-model based on the auxiliary marker sample was obtained.Then used a small number of target mark samples to fine-tune the weight and offset value of the DBN premodel to realize the migration of the weight and offset value of the DBN network from the source domain to the target domain to adapt to the new target sample identification,and finally improve the target domain sample the effect of fault classification accuracy.The feasibility and effectiveness of the proposed method under the limited conditions of the target training samples were proved by the planetary gearbox fault simulation experiment.In summary,this paper proposed method based on deep belief network for the selflearning and intelligent diagnosis of planetary gearbox fault diagnosis.This method improved fault diagnosis efficiency of planetary gearboxes by using the original signal spectrum as the input of the deep belief network.Aiming at the low accuracy of planetary gearbox fault diagnosis under limited conditions,a planetary gearbox fault diagnosis method based on Deep Belief Network migration learning is proposed,which improves the planetary gearbox fault diagnosis under limited samples.Finally,the methods were validated by using the planetary gearbox fault simulation experiment data The experimental results proved the effectiveness and feasibility of the proposed method.