Research On Fault Diagnosis Method Of Planetary Gearbox Based On Improved Capsule Network

As a common component in large mechanical equipment,planetary gearbox is widely used in sea and air exploration and military fields because of its smooth transmission,compact design and large load-bearing capacity.However,because the planetary gearbox works in the environment of high vibration and high erosion for many years,which makes its internal structure highly susceptible to pitting,cracking and tooth breakage.Therefore,in order to reduce the loss of personnel and property,it is of paramount relevance to study the high-precision fault diagnosis technology of planetary gearbox.The traditional fault diagnosis method based on the characteristics of the signal itself needs to rely on a large number of prior knowledge and expert experience,which faces the problem of low diagnosis accuracy and efficiency under massive data.In response to the above problems,deep learning provides new ideas for fault diagnosis techniques.This thesis takes the planetary gearbox as the subject of study and investigates a fault diagnosis method for planetary gearboxes based on an improved capsule network.Starting from the structural composition and fault mechanism of the planetary gearbox,this thesis analyzes the time-frequency characteristics of different faults,and explores the high-precision diagnosis method of planetary gearbox under different speed and different noise using time-frequency data.The specific research content of the thesis is as follows:1.For the components of planetary gearboxes,this thesis analyzes the formation mechanism of common faults and explains the advantages and disadvantages of the traditional methods by combining the time frequency knowledge.To address the problem of over-dimensionality of time-frequency data,the linear interpolation algorithm is applied to the processing of time-frequency data on the basis of time-frequency transformation,which circumvents the manual feature clipping and scaling and realizes the fully automated process from data acquisition to processing.2.Fault diagnosis under variable working conditions is an important problem in the field of rotating machinery fault diagnosis.To address this problem,capsule network is applied to the field of variable working condition diagnosis of planetary gearbox in the thesis.The network consists of ordinary convolutional modules and capsule modules.The convolution module realizes the preliminary feature extraction of input data.The capsule module uses the feature extraction advantages of vector neurons to reduce the impact of rotating speed on fault features to a certain extent.3.In order to improve the fault diagnosis accuracy of planetary gearbox in noise environment,a convolution capsule network with improved structure is proposed by combining residual structure and attention mechanism with capsule network.The proposed network filters different features through the attention mechanism to improve the effect of network training.The robustness of the network is enhanced by the way of channel layering and asymmetric convolution with improved residual structure.Experiments show that the proposed network has excellent diagnosis effect on noisy signals.