Research On Oil-Vibration Online Intelligent Monitoring Method For Gear Wear Damage Based On Deep Learning

With the advancement of the intelligent upgrade of major equipment,the demand for intelligent operation and maintenance of equipment has intensified.Gear is a key component of mechanical equipment,and gear wear is the main cause of its failure.In the current monitoring methods,sensors cannot monitor the wear and damage of the gear surface.Therefore,the fault diagnosis and fault monitoring of gears still mainly rely on oil analysis.Methods and vibration monitoring means.However,using oil analysis or vibration monitoring methods alone has problems such as incomplete monitoring information,low monitoring efficiency,and insensitive fault characteristics.For this reason,this paper takes the online monitoring of gear wear damage as the research object.Based on the oil-vibration multi-source and multi-dimensional monitoring data obtained online,based on the deep learning theory,the research on the oil-vibration online intelligent monitoring method for gear wear and damage is carried out.The main research contents include:(1)In order to deeply analyze and interpret the oil and vibration signals,the common fault forms of gears,the types of gear wear and the types of abrasive particles are interpreted,and the mechanism of abnormal vibration during gear faults is introduced.The research is oil-vibration The analysis and feature extraction of monitoring data provides a theoretical basis.(2)Carry out research on the information perception and feature extraction method of wear and abrasive particles in lubricating oil.Aiming at the complex background of the spectrum images obtained by online ferrography technology and the difficulty in identifying the wear and abrasive particles,a multi-objective gear wear and abrasive particle identification method based on the yolov5 model is proposed.Six types of abrasive abrasives,including sliding wear abrasives,spherical abrasives,oxidative abrasives,and cutting abrasives,are automatically identified,with a m AP of 77.6%,which provides a way for the identification and statistics of different types of abrasives.(3)Aiming at the problem of insufficient fault information of oil single-source monitoring data,carry out research on intelligent monitoring methods based on oil-vibration multi-source heterogeneous monitoring data.Based on the oil-vibration multi-source heterogeneous monitoring data,extract the 3-dimensional oil index and 14-dimensional vibration signal characteristic index of the number of abnormal abrasive particles,the total number of abrasive particles and the concentration of abrasive particles,and construct a multi-dimensional characteristic index system based on LSTM.The neural network monitors and evaluates different stages of gear wear,and the feasibility of the method proposed in this paper is verified by the sample data of different wear stages.(4)In order to further verify the method proposed in this paper,about 150 hours of gear life cycle wear accelerated test research was carried out,and 5460 sets of oil and vibration monitoring data were obtained online.The processing results of the method proposed in this paper show that due to the oil-vibration monitoring data with complementary advantages,the monitoring accuracy of the gear wear damage operating state can reach 83.68%,which is significantly higher than the results predicted by the LSTM model on the single-source monitoring data set.This research provides a theoretical basis for gear wear damage monitoring,and is of great significance to ensure the safe,efficient,and long-life operation of gear transmission systems,and to avoid unnecessary waste of resources and economic losses caused by overmaintenance and under-repair.