Application Of Image Features And Gaussian Process In Research On Fault Diagnosis Of Planetary Gearbox Gears

The planetary gearbox is small in size but has a large carrying capacity and a more stable operating advantage,so it is widely used in many fields such as aviation and new energy power.However,the planetary gearbox has been running in the harsh working environment with high strength and high load,and the internal key components are very easy to fail.If it is allowed to further develop,it will cause damage to the equipment and even cause more serious consequences.In order to accurately identify its failure mode and achieve the goal of safe and long-term operation of mechanical equipment,therefore,the use of image features and Gaussian process methods to carry out diagnostic research on planetary gearbox gear failure.This article first analyzes in detail the characteristics of the planetary gearbox structure,the characteristics of the vibration signal,the failures of key components and the causes.On this basis,the signal acquisition and research work used the model HFXZ-1 planetary gearbox fault diagnosis test platform.The planetary gearbox vibration signal acquisition experiment was completed in six working conditions.And by analyzing the characteristics of the vibration signal time and frequency spectrum,it is found that the collected vibration signal components are complex.It is concluded that it is difficult to accurately diagnose the failure mode through the traditional spectrogram.Second,it is proposed to use the time-frequency analysis method to process the signal combining the current development status of the planetary gearbox fault diagnosis research and the difficulty of the research technology.The simulation signal proves that the S-transform time-frequency analysis method can retain the signal characteristics without loss,and is not disturbed by the linear cross term to avoid the problem of image distortion.And successfully obtained the time-frequency distribution imageof the vibration signal.Third,considering that the extraction of traditional feature values will be affected by many factors,so that the problem of the nature of the fault cannot be fully and accurately reflected,a technical method in the field of fusion image processing is proposed,The local binary mode is used to extract the image texture features of the time-frequency distribution image of the planetary gearbox vibration signal to characterize its running state.Fourth,with the advantages of Gaussian process models in machine learning recognition,a Gaussian process model suitable for planetary gearboxes is constructed through specific kernel functions and feature sets and fault diagnosis is performed to obtain the final recognition result of the failure mode.Finally,a comparative experiment is performed based on the signal time domain characteristics and ELM.The experimental results prove the effectiveness of the time-frequency analysis method and image features proposed in this paper,and verify that the integration of multi-domain technologies can more accurately and comprehensively reflect the operating status information of the planetary gearbox.The number of iterations and time consumption of this method is less and the accuracy rate is91.67%,which is superior to other methods.It also further proves the superiority of the fault diagnosis method using image features and Gaussian process in identifying planetary gearbox failure modes.