Joint Health State Anomaly Detection Of Industrial Robot Based On Current Signal

Industrial robots have greatly improved production efficiency,but with the diversification of their functions and the complexity of their structures,it is an important task to ensure the high reliability of industrial robots.At present,there are some studies on the anomaly detection and fault diagnosis of industrial robots.Most of the studies are based on the vibration signals of industrial robots.In the process of vibration signal acquisition,there are some disadvantages such as inconvenient sensor installation and large environmental interference.The industrial robot current signal acquisition is more convenient,and the signal is not easily disturbed.Therefore,based on the current signal of industrial robot,this study develops an anomaly detection method of joint health status of industrial robot based on current signal.This paper analyzes the basic mechanism and failure mode of industrial robots,and introduces the accelerated life test platform and data acquisition process of industrial robots in detail.Firstly,the characteristics of industrial robot joint current signal are analyzed,such as periodicity,coupling and time-frequency characteristics.Aiming at the problem of signal cycle division and motion stability,the signal cycle division can be completed by autocorrelation function and peak finding function.For the problem of phase difference of signals collected at different time,the phase difference between signals can be found by cross-correlation method.Aiming at the problem of working condition division of industrial robots,the time-frequency information obtained by shorttime Fourier transform of joint current signal can be connected with the working condition information.Aiming at the coupling problem between different joints of industrial robot joint current signal,it is found that the signal length between different joints is modulated,and the decoupling analysis is difficult,and the follow-up research is no longer carried out.Based on the above analysis,the understanding of industrial robot joint current signal is strengthened.Secondly,aiming at the problem of industrial robot joint anomaly detection,an industrial robot joint anomaly detection model based on criterion fusion is established.The classical statistical model is established by the method of skewness index and control chart.The model has good interpretability,but the anomaly detection results have some misjudgment.In order to find the potential rules in the joint current signal,an anomaly detection model based on machine learning is established by time domain analysis and spectral clustering method.The anomaly detection results are more accurate,but the interpretability of the model is poor.The criterion fusion method fuses the results of the two models,so that the model has good interpretability and improves the accuracy of model detection.Finally,in the face of the unbalanced data set of industrial robot joint current signal(normal and abnormal sample ratio 10 : 1),the sliding window method is processed into a balanced data set(normal and abnormal sample ratio 1 : 1).A deep learning model for industrial robot joint anomaly detection based on 1D-CNN is established,with an average accuracy of over 99% on different industrial robot datasets..Aiming at the problem of anomaly detection of industrial robots at the group level,in the face of specific problems such as insufficient data labels and no abnormal labels for new equipment,an anomaly detection model of joint health state of industrial robots based on transfer learning is established.Among them,1D-CNN is used as the feature extractor of the model,and the maximum mean difference is used as the measurement index of the distribution difference of current signals of different equipment.It is verified in the No.2 joint data set,and the accuracy rate is more than 75 %.The joint anomaly detection of industrial robots across devices is completed.This study helps enterprise engineers to find effective equipment status information from a large number of industrial robot joint current signals,capture the early abnormalities and degradation trends of equipment,and provide a scientific basis for enterprises to formulate reasonable maintenance plans.