Study On Methodologies Of Fault Diagnosis Of Planetary Gearbox Under Varying Rotational Speed Based On Deep Convolutional Neural Networks

Planetary gearbox plays vital transmission roles in many electromechanical systems.Once fault occurs somewhere in it,it will cause huge economic loss and casualty.Thus it is more than necessary to do condition monitoring and fault diagnosis to the planet gearbox.However,it is quite difficult to extract fault features from raw vibration signal without being interfered by time-varying components in the field of intelligent fault diagnosis to planetary gearbox under varying speed conditions.To conduct fault diagnosis in the unknown speed conditions is a even more tough task.Aiming at the problems raised above,this paper preliminarily analyzed the vibration characteristics of planetary gearbox with planet gear fault,based on which a vibration model of planetary gearbox with faulty planet gear was established.Accompanying the real test planetary gearbox vibration signals,it was validated that the planetary gearbox vibration signal under varying working speed has strong non-stationarity and trait of multi-frequency modulation.Combining the traits of the vibration signals of planetary gearbox,an intelligent fault diagnosis scheme of planetary gearbox under varying working speed based on deep convolutional neural networks(DCNN)was designed.The designed scheme is able to automatically extract fault features,and able to diagnose two undistinguishable planet gear faults under known varying working speed conditions.Directing at the cross-domain diagnosis problem for intelligent diagnosis model under unknown speed conditions,this paper proposed three optimization strategy.First,by adding a small number of validation samples under unknown working speed into training process,the model can achieve cross domain diagnosis under unknown speed conditions.Secondly,considering that one may not be able to obtain vibration signals under target working speed when training the model,an optimized deep convolutional neural network architechture was designed based on drop-out strategy and residual block to enhance its generalizability.Finally,generative adversarial networks were utilized to generate signals under unknown working speeds in order to augment training data set,which further improved the fault diagnosis ability of the model.Experiments have been conducted to validate that the proposed intelligent fault diagnosis scheme based on DCNN is capable to realized fault diagnosis under varying working speed conditions for planetary gearbox,and the three optimization strategy could effectively improve the cross-domain diagnosis performance of the diagnosis model.This paper provided a valid intelligent diagnosis scheme for planetary gearbox under varying speed conditions,and also provided a new idea and solid solution to the cross-domain diagnosis problem for intelligent diagnosis model under unknown speed condtions.