Fault Diagnosis Of Planetary Gearbox Based On Data Drive

The planetary gearbox is a key transmission link of large mechanical equipment.Its reliability and stability have a crucial impact on the safe operation of mechanical equipment.With the increase of service time and the complicated changes of working conditions,the planetary gearbox will inevitably produce fault conditions.Therefore,it is extremely important to diagnose the failure of the planetary gearbox quickly and effectively.At present,the analysis methods and intelligent diagnostic methods of diagnostic signal are widely used in planetary gearbox fault diagnosis.However,the operating environment of heavy machinery is very terrible,and is full of various noises,which will lead to a decline in the diagnostic performance of traditional signal processing methods.Therefore,how to extract representative fault features from complex vibration signals is particularly important,which is also the research topic of this paper.This paper takes planetary gearboxes as the research object and researches its fault diagnosis.This paper proposes a parameter optimization strategy of characteristic index and two fault diagnosis methods to improve the diagnosis of signal processing methods.The research results of this article are as follows:(1)In order to determine the problem of selecting two parameters in the improved sideband energy ratio(MSER),namely the bandwidth width and the number of sidebands.This paper proposes a parameter selection method based on extreme gradient boosting(XGBoost).This method uses XGBoost to classify and diagnose MSER values calculated from different parameter combinations.Then based on the diagnosis results,to select a set of parameters with the highest diagnostic accuracy as MSER parameters.In addition,this parameter selection method leads to a diagnosis method based on traditional signal indicators,that is,by extracting fault indicators of diagnostic signals and using shallow machine learning algorithms for fault diagnosis.(2)In order to avoid the tedious manual extraction of signal failure indicators and achieve "end-to-end" training,this paper proposes a "deep autoencoder-small volume convolutional neural network structure(DAE-MCNN)".The DAE part of the network can automatically extract features from diagnostic signals without supervision,while the MCNN part can quickly complete iterative parameter updates from a small number of data samples due to its small size and few network parameters,which improves the diagnostic accuracy and avoids the occurrence of network overfitting.(3)In order to apply the model trained by vibration signals of constant speed to fluctuating speed,this chapter proposes a DAE-MCNN-based transfer learning method that combines three types of migration strategies for deep transfer learning.The feature extraction ability only extracts the fault characteristics of the signal,and ignores the signal amplitude transformation caused by the fluctuation speed.Thereby simplifying the problem.During the migration process,all training parameters of the DAE-MCNN network are frozen,and then two layers of adaptive networks are added to improve the model's feature learning ability and generalization ability.Experiments prove that the proposed strategy has better performance.In this paper,the vibration signals obtained from the planetary gearbox experimental platform are used to verify and evaluate the effectiveness of the three proposed fault methods.The experimental analysis results show that the proposed method is superior to the existing fault diagnosis methods.Finally,the problems and future work of this method are summarized.