Research On Fault Diagnosis Technology Of Planetary Gearbox Based On Deep Belief Network

The planetary gearbox is an important part of the vehicle transmission.It can meet the requirements of different gears and different driving conditions by changing the transmission ratio and transmitting high power.Due to the complex structure of the planetary gearbox and the harsh operating conditions,the internal gears and other components often fail,so it is very necessary to conduct research on the fault diagnosis technology of the planetary gearbox,which is to ensure the safety and reliability of the operation of the vehicle transmission Sex is of great significance.Aiming at the problem of planetary gearbox fault diagnosis,this paper studies the application of a deep learning method — deep belief network(DBN)in the typical fault diagnosis of planetary gearbox.Firstly,the concept of Deep Belief Network and related theories are elaborated in detail.Secondly,build a simulation planetary gearbox test bench,determine the location of the measuring point,the experimental working conditions and the type of fault,analyze the collected vibration signals in the time domain and frequency domain,and group the experimental data in groups.Then the grouped experimental data is input into the Deep Belief Network.According to the structure of the Deep Belief Network,the influence of different model parameters on the diagnosis results is discussed.In order to verify the ability of each hidden layer feature extraction of DBN,the principal component analysis method(PCA)is introduced,through which the first three principal component components are retained after dimensionality reduction of the data,and then each hidden layer is visualized in three-dimensional space to extract Characteristics.Finally,by inputting the vibration signals of previous different experimental conditions to the back propagation network(BP),extreme learning machine(ELM),and deep belief network,the comparative analysis results verify the deep belief network in feature extraction and classification recognition The superiority also confirms the ability of the deep belief network to process experimental data of planetary gearbox failures.Through repeated experiments and verification comparisons,it is shown that the deep belief network is intelligent in fault diagnosis.This method can automatically extract fault features from the original data,and adding a classifier can achieve classification recognition,which not only has high recognition accuracy,but also greatly saves Time cost,the results show that this method has certain application prospects in the field of planetary gearbox fault diagnosis.