Analysis Of Acoustic Emission Characteristics Of Cracks And Life Prediction Of Harmonic Reducer

The performance status of harmonic reducers directly determines the stability and reliability of high precision robotic arms,which in turn affects the accuracy and quality of products.To ensure the continuous and efficient operation of the equipment,it is vital to know the health status of the harmonic reducer.Since harmonic reducers are placed inside the joints of robotic arms,they require high installation accuracy,are inconvenient to stop and disassemble periodically for inspection,and have a lag in sensing the damage status of the components.Therefore,there is a need to find an inspection method that can overcome the shortcomings of traditional methods of detecting defects with difficulty,low sensitivity,and poor integrity.Acoustic emission detection is independent of the shape,material,and location of the defect and is sensitive to minor and dynamic damage within the structure.The fracture is the most important factor affecting the life of a harmonic reducer,it is important to study the relationship between its acoustic emission characteristics and harmonic reducer damage,and to establish a life prediction model based on acoustic emission characteristic parameters for the health status assessment of harmonic reducer.First,the basic structure and transmission characteristics of the LHSG-20-50-C-I harmonic reducer were introduced,and the main failure forms of this component were summarized.For the load characteristics of the flexure wheel,an equivalent model of the flexure wheel was established to study its stress distribution,and the conclusions were verified by finite element simulation.A small angular crack was introduced at the maximum stress node,and the stress field parameters at the crack tip were analyzed to predict the crack expansion capability and trend.Based on the extended finite element method,the crack expansion behavior of the flex wheel was analyzed,and the characteristics of the flex wheel crack during its complete expansion were summarized with the corresponding experiments.Then,the acoustic emission detection theory was studied,and the selection of each part of the acoustic emission acquisition system was completed according to the application object.The acoustic emission signal acquisition experimental bench was built,and the acquisition settings were preset,and the acoustic emission sensors were arranged in multiple directions.The origin of acoustic emission events generated by harmonic reducers with different activity levels was analyzed.The time-frequency characteristics of acoustic emission were extracted,and the selected characteristics were determined by correlation analysis of acoustic emission characteristics parameters,which could be used for the establishment of machine learning models.Then,based on the accumulated acoustic emission characteristics,the crack damage was divided into three stages by combining the simulation results with the experimental results,and the evolution law of the acoustic emission characteristics parameters in each stage was studied in turn.Based on the principal component analysis and DBSCAN,the Euclidean distance of the sample points in the principal component feature space was calculated,and the samples were classified into eight categories and their acoustic emission characteristics were analyzed based on this parameter.Finally,based on the acoustic emission characteristics in the degradation process,a multiple linear regression model,a light gradient boosting machine,and an extreme gradient boosting were developed to establish a harmonic reducer life prediction model.The life prediction performance of the three regression models was evaluated using the mean absolute error and goodness of fit,and the prediction accuracy of the two regression tree models was compared and analyzed.It was found that the prediction performance of the regression tree model was significantly better than that of the linear regression model in the time series analysis;the sampling error and prediction accuracy of the two tree models were similar,but the prediction trends were different.In terms of regression effect,XGBoost has higher prediction precision than Light GBM,with MAE of 21.9737 and R2 of 0.8113,which identifies XGBoost as the integrated optimal model for regression performance.The prediction results for the destabilization and failure states of the harmonic reducer were about 1168 s and 3337 s earlier than the real values.