Research On Compound Fault Diagnosis Method Of Planetary Gearbox Based On Convolutional Neural Network

The planetary gearbox is a key transmission component used by fans,helicopters,ships,and other major equipment,which used to change the speed and torque to match the power source and load.It has the advantages of small mass,small size,large reduction ratio,and high transmission efficiency.Compared with the fixed-axis gearbox,the planetary gearbox has a more complicated structure,which causes the fault signals to be easily coupled and canceled each other during the transmission process,making it diffcult to extract valuable information that can be used to judge the characteristics of the fault state from the strong interference vibration signal.In addition,gears and bearings have a high failure rate in the gearbox,if they were failed in the same time,that are very harmful to the system.However,in actual working conditions,it is often necessary to reach a certain damage level before shutting down for maintenance.During this period,multiple failures often occur.In view of the above problems of planetary gearbox fault diagnosis,this paper first uses virtual prototype technology to model and simulate the common faults of planetary gearboxes,and studies the characteristics of common faults.And based on the characteristics of planetary gearbox faults,a lightweight diagnostic model was constructed based on convolutional neural networks.Experiments verified that the model has good diagnostic performance for single faults of planetary gearboxes.Then,by improving the structure of the diagnosis model,the diagnosis performance of the model to the compound fault of the planetary gearbox is improved.Finally,through model visualization,the reason why the model has good diagnostic performance is discussed.The main work and innovation of the paper:1.First,the virtual prototype technology is used to dynamically model the common faults of the planetary gearbox,and the characteristics of the common faults of the planetary gearbox are studied;2.For the planetary gearbox fault,the traditional convolutional neural network is improved to implement a high-performance and lightweight diagnostic model,and the diagnostic model directly uses the composite fault original signal collected by the multi-source sensor as the input;3.The convolutional neural network can automatically perform feature extraction.By setting the bypass to the network,the multiplexing of the extracted fault features is realized,and the problem of low accuracy of the compound fault diagnosis of the planetary gear box is solved.The improved diagnostic model has good performace,and the accuracy of compound faults of planetary gearbox gear bearings has been increased to 99.99%;4.Visualized the network model and discussed the reasons why the diagnostic model has good diagnostic performance.