Damage Detection And Performance Evaluation Of Harmonic Reducer Based On Sound Signal And Acoustic Emission

Harmonic reducer is an important part of industrial robot and other high precision equipment,and its health state directly affects the performance of the equipment.The internal component of harmonic reducer is subjected to continuous alternating load in the work,which is easy to cause the internal damage of each component.Damage detection is a necessary link before leaving the factory.It mainly judges the existence of damage and can reduce the huge economic loss caused by product recall.In the process of service,it is necessary to study the type of damage,clear the law of damage evolution,judge the performance state of harmonic reducer,so as to realize predictive maintenance.Therefore,in order to realize the whole life cycle damage detection and performance evaluation,it is necessary to find a compound detection means to ensure the continuous and efficient operation of the equipment.The sound signal has the advantages of simple acquisition and non-contact measurement,which can realize the rapid inspection of harmonic reducer.In service,AE technology is used to detect early defects,enable predictive maintenance,and assess performance degradation.Therefore,in order to meet different scenarios,this paper uses the detection means combined with acoustic emission to complete two different tasks of damage detection and performance evaluation.Firstly,the basic structure of the harmonic reducer is introduced,the interaction of each component is analyzed,the main failure modes are summarized,and two damage types of wear and crack are obtained from different failure modes.The damage types correspond to different AE sources.Based on the crack as the main AE source,the mapping relationship between AE characteristic parameters and cracks is established.It is proved theoretically that AE characteristic parameters such as ringing count and energy can reflect the crack growth rate and the degree of performance degradation.Secondly,the experimental scheme is designed to collect sound and acoustic emission data and extract features.A factory detection platform based on sound signals was built,harmonic reducer was excited by force hammer to obtain damaged sound data,and feature engineering was constructed by three signal processing methods: time domain feature extraction,frequency domain feature extraction and variational mode decomposition feature extraction.An accelerated life test platform was built to collect acoustic emission data during service,and simplified waveform parameters were used to extract features.Then,based on the collected sensor data,the existence of damage before delivery is detected by sound signal,and the types of damage in service are distinguished by acoustic emission signal.The sound signal is segmented,the sound data set of damage detection is expanded,and the damage features are extracted adaptively by the method of variational mode decomposition.More obvious difference of frequency components can be observed in the harmonic reducer with or without damage.By comparing the test set accuracy of different machine learning algorithms,different feature engineering and different number of mode decomposition,it is obtained that the accuracy of VMD-SVM algorithm is the highest when the number of mode decomposition K=5,which is 99.16 %on verification set and 99.53 % on test set.Then,based on the AE data of the whole life cycle,the AE experience graph analysis method was used to distinguish the two damage types of wear and crack in service,and the analysis results were consistent with the final experimental failure results.Finally,the acoustic emission data of the whole life cycle is used to evaluate the performance of the harmonic reducer based on the method of damage sources-outliers accumulation.According to the transformation trend of energy and average level,combined with the main damage-crack propagation law,four periods were divided into normal non-damage period,crack initiation period,crack stable propagation period and crack unstable propagation period.In order to fit the variation trend of acoustic emission parameters,an outlier accumulation algorithm based on principal component analysis was proposed to obtain the performance degradation trend of harmonic reducer.In the performance degradation curve obtained by this method,a sharp increase of anomalies can be observed,which can be used as the alarm threshold,700 h earlier than the alarm threshold of One-Class SVM anomaly detection algorithm.